A systematic review and thematic analysis of gender bias in AI

A systematic review and thematic analysis of gender bias in AI | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Systematic Review A systematic review and thematic analysis of gender bias in AI Jingyue Liu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9240627/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 14 You are reading this latest preprint version Abstract Artificial intelligence (AI) systems have been applied across various fields and the problems of AI’s gender bias become widely recognized, attracting researchers’ attention. However, there is a lack of review articles that comprehensively consider publication, research design and research focus aspects related to gender bias in AI. Therefore, this study conducted a systematic review of 29 articles based on PRISMA to explore them. The findings revealed an overall increasing research trend of AI’s gender bias and identified 10 domains within AI that exhibited gender bias. Besides, it was found that the causes of AI’s gender bias could be attributed to data bias, human bias, algorithmic bias and social bias. Solutions were correspondingly proposed from data level, ethical and social level, algorithmic and model level, and assessment and diagnosis. The study also identified that amplification of existing inequalities, gender stereotypes and unequal treatment were three main consequences and the challenges of mitigating gender bias were generated from the performance of models, culture and value, and ethical concerns. Further study could proceed to study users’ perception of AI’s gender bias to provide a comfortable experience environment and promote a healthy development of AI. AI Gender bias Gender stereotype Algorithmic bias Human bias Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 1 Introduction Considered a powerful technology, artificial intelligence (AI) has fundamentally changed technical environments (Dwivedi et al., 2023 ) and is regarded as the center of the Fourth Industrial Revolution (Khan & Ewuoso, 2024 ). With its widespread and increasing application in hiring (Gagandeep et al., 2024), medicine (Buslón et al., 2023 ) and interviews (Suen & Hung, 2023 ), etc., problems of bias in AI become visible, mainly including racial bias, gender bias, religion bias and cultural bias (Tubadji et al., 2021 ). Among them, gender bias garners large attention for its profound impact on individuals and society (Baker & Hawn, 2021 ; Galos & Coppock, 2023 ). Gender bias in AI does not only impact individuals’ rights and opportunities but also can be amplified by the AI system itself. Thus, addressing gender bias should be solved as the priority and seen as the task in the research and policy agenda (Hall & Ellis, 2023 ), which could promote gender equality, social justice and healthy development of AI technology. Gender bias is the tendency to prefer one gender over another, deriving from individuals’ unconsciousness and automatic emotions with skipping rational thinking (Coiro & Pollak, 2019 ). Gender bias in AI refers to the unfair or discriminatory treatment towards individuals based on gender, often deriving from imbalanced data, algorithmic prejudices and human biases embedded in AI systems. This bias not only undermines the fairness and trustworthiness of AI but also impedes its widespread adoption. The lack of clear regulations and the opaque nature of AI systems exacerbate these issues (Noble, 2018 ). Historically, the issue of gender bias in AI is often overlooked, with early research focusing primarily on technical performance rather than ethical implications (Alami et al., 2020 ). Recent studies have highlighted the multifaceted harms of gender bias in AI. For instance, Ellis and Hall ( 2023 ) noted that AI systems often amplify existing societal biases, particularly in image search results where engines like Bing and Google display imbalanced gender representations (Górska & Jemielniak, 2023 ). In the hiring process, gender bias is particularly obvious (Chang, 2023 ). Gagandeep et al. (2024) revealed that Amazon’s AI-based resume filtering system unintentionally favored male candidates, thereby perpetuating gender prejudice. The emergence of generative AI models, such as ChatGPT and LLaMA, has introduced new approaches but also challenges in addressing gender bias. These models while transformative, have been shown to perpetuate and even amplify existing biases. For instance, some studies reveal that ChatGPT often generates text that reinforces traditional gender roles, such as associating independent and adventurous traits with men and inclusive and emotional roles with women (Gross, 2023 ). Similarly, LLaMA’s open-source nature has raised concerns about its possible abuse in applications that may deepen gender inequalities, such as biased content generation in hiring tools or educational materials. Fang et al.’s ( 2024 ) study claimed that the representative large language models (LLMs), like ChatGPT and LLaMA, demonstrated significant bias against women and Black people. To address these challenges, researchers proposed various mitigation techniques. One promising technique is prompt engineering, where carefully designed inputs are used to guide generative models toward more neutral and unbiased outputs (Dwivedi et al., 2023 ). For example, a prompt like “Describe a nurse without referencing gender” can help reduce stereotypical associations. Besides, the development of more diverse and representative datasets has been emphasized as a critical step in tackling biases. In addition to technological efforts, the existing research also emphasizes individuals’ greater awareness of gender bias and the necessity of legal regulations (Bernagozzi et al., 2021 ; Buslón et al., 2023 ). Like in the healthcare field, the American Medical Association passes a policy, calling for the development of thoughtfully designed, high-quality and clinically proven AI technologies (Rigby, 2019 ). Despite these efforts, there is a scarcity of review articles analyzing gender bias in AI technology. 7 review articles about bias in AI were identified, including systematic review (N = 3), scoping review (N = 3) and narrative review (N = 1). Among these, only two reviews specifically focus on gender bias. 3 results just discuss the bias generally (Chen et al., 2024 ; Delgado et al., 2022 ; Paul et al., 2022 ), and others explore bias related to ethics (Daneshjou, 2021), race, or age (Chu et al., 2023 ). The reviews of gender bias are all performed through the systematic review, which utilizes a more strict methodology to thoroughly examine the literature in a clearly defined manner (Gregory & Denniss, 2018 ). Furthermore, although a few papers touch on the causes of bias (Daneshjou, 2021; Delgado et al., 2022 ; Hall & Ellis, 2023 ), or propose mitigation methods (Chen et al., 2024 ; Hall & Ellis, 2023 ; Shrestha & Das, 2022 ), none of them comprehensively address the causes, mitigation methods, consequences and challenges of mitigating gender bias simultaneously (Table 1 ). Table 1 The detailed information of review articles of bias in AI Author Year Title Type Focus Daneshjou 2021 Lack of Transparency and Potential Bias in Artificial Intelligence Data Sets and Algorithms: A Scoping Review Scoping review Data bias in Clinical AI algorithms; sources of data bias; ethnic or race bias to patients through the recognition of skin color Paul et al. 2022 Bias Investigation in Artificial Intelligence Systems for Early Detection of Parkinson’s Disease: A Narrative Review Narrative review Bias in the health-related AI model; computing the risk of bias Delgado et al. 2022 Bias in algorithms of AI systems developed for COVID−19: A scoping review Scoping review Causes and consequences of bias from the ethical perspective Shrestha & Das 2022 Exploring gender biases in ML and AI academic research through systematic literature review Systematic review Gender bias in the machine learning and AI assistant automated systems; mitigation and detection methods Hall & Ellis 2023 A systematic review of socio-technical gender bias in AI algorithms Systematic review Solutions, causes and consequences of gender bias in AI algorithms from the socio-technical framework Chu et al. 2023 Age-related bias and artificial intelligence: A scoping review Scoping review Identifying how AI systems encode, produce, or reinforce age-related bias; main domains that present the age bias of AI (age recognition and facial recognition systems) Chen et al. 2024 Unmasking bias in artificial intelligence: A systematic review of bias detection and mitigation strategies in electronic health record-based models Systematic review Bias detection and mitigation strategies in the health-related model; six bias types (algorithmic, confounding, implicit, measurement, selection and temporal) This study 2024 Does gender bias in AI exist: A systematic review and thematic analysis Systematic review Gender bias in AI; research trend; causes, solutions, consequences and challenges of AI’s gender bias To bridge these gaps, the present study adopts a systematic review approach, providing a comprehensive analysis of gender bias in AI systems. Unlike the above reviews that center on algorithmic or general bias, this study takes a broader perspective, examining gender bias from the macroscopic areas that manifest it to the specific application systems of AI, then to the model algorithms behind them. By reviewing the underlying causes, consequences, challenges and mitigation strategies of gender bias, this study aims to offer a more complete understanding of the issue. Findings in this review also attempt to provide insightful views into the root causes of gender bias, enabling the design of fairer and more inclusive AI systems for AI developers. Besides, the review summarized the societal consequences of gender bias in AI, offering policymakers a foundation for making targeted policies to mitigate these effects. To achieve these goals, the following research questions were proposed from the publication aspect, research design aspect and research focus aspect. RQ1: What is the current trend of AI’s gender bias, including year publication and journal of publication? RQ2: What is the distribution of research methodologies in the study of AI’s gender bias? RQ3: What are the domains that present gender bias in AI RQ4: What causes of AI’s gender bias are proposed? RQ5: What mitigation approaches for AI’s gender bias are proposed? RQ6: What are the challenges of mitigating AI’s gender bias? RQ7: What are the consequences of AI’s gender bias? 2 Methods 2.1 Data collection On 3 May, 2024, we selected five databases to retrieve articles, including the Web of Science (WOS), Elsevier, Taylor & Francis, Wiley and Sage. To ascertain the search terms, we first determined the basic keywords AI and gender bias to gain the relevant research articles or reviews (e.g. Delgado et al., 2022 ; Shrestha & Das, 2022 ), and then summarized the other expressions having the same meaning. Finally, we formed the two search strings in Table 2 . The logic connection among the terms in the same category was OR and the AND was used to link these two categories. Table 2 Keyword search scheme Technology Gender “AI” AND “Gender bias” “Artificial intelligence” “Sexism” “AI agent” “Feminism” “Artificial intelligence agent” “Gender inequality” “Chatbot” “Gender prejudice*” “Chatterbot” “Gender stereotype*” In the WOS, the Core Collection was chosen, including six indexes: Science Citation Index Expanded (SCI-EXPANDED, 2013 to present), Social Sciences Citation Index (SSCI, 2008 to present), Arts & Humanities Citation Index (A&HCI, 2008 to present), Emerging Sources Citation Index (ESCI, 2018 to present), Current Chemical Reactions (CCR-EXPANDED, 1985 to present) and Index Chemicus (IC, 1993 to present). 164 results were gained for “AI” OR “Artificial intelligence” OR “AI agent” OR “Artificial intelligence agent” OR “Chatbot” OR “Chatterbot” AND “Gender bias” OR “Sexism” OR “Feminism” OR “Gender inequality” OR “Gender prejudice*” OR “Gender stereotype*”. We selected the document type of articles (N = 148). In Wiley and Taylor & Francis, through setting these keywords in the abstract, 3 and 11 articles remained respectively. In Elsevier Science Direct, we set title, abstract, keywords: “AI” OR “Artificial intelligence” OR “AI agent” OR “Artificial intelligence agent” OR “Chatbot” OR “Chatterbot” AND Title: “Gender bias” OR “Sexism” OR “Feminism” OR “Gender inequality” OR “Gender prejudice” OR “Gender stereotype”, there were 6 research articles. The symbol “*” was not allowed in Elsevier Science Direct different from the other four databases. In Sage, by keying these keywords in the abstract, 2 articles remained. In total, 170 articles were imported into EndNote X9 for further screening (Fig. 1 ). 2.2 Inclusion and exclusion criteria The selection of papers was rigorously based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Page et al., 2021 ). We first skimmed through titles and abstracts to filter the articles that 1) were not written in English and 2) were duplicates. Some irrelevant articles could not be excluded by the retrieval terms because of the different combinations and logic of these keywords. Thus, we proceeded to a more detailed screening, where we excluded articles that 1) focused on the influence of gendered AI (e.g. Brown, 2022 ); 2) centered on the influence of users’ gender on the perception of AI; or 3) emphasized the identification and avoidance of gender bias by AI (e.g. Guevara-Gómez et al., 2021 ; Koo, 2022 ; Pisanelli, 2022 ). In this stage, we gained 36 articles. Then, we carefully looked through these 36 full papers to ensure they met our inclusion criteria, which required studies to 1) employ clear and rigorous methodologies to investigate gender bias in AI; 2) explicitly identify gender bias in AI systems; 3) discuss the societal, ethical, or technical impacts of gender bias; 4) explore the underlying causes or mechanisms of gender bias or; 5) propose or evaluate strategies to mitigate gender bias. During this stage, 7 results were removed because of inadequate discussion on gender bias or lack of clear methodology. Ultimately, 29 articles were used for data analysis (Fig. 1 ). 2.3 Quality assessment These 29 results were evaluated under the guidelines established by the American Educational Research Association (AERA) to guarantee the quality (Durán et al., 2006 ). The guideline has eight concrete standards, including problem formulation, design and logic, sources of evidence, measurement and classification, analysis and interpretation, generalization, ethics in reporting and title, abstract, and headings. Each of them was scored from 1 point to 5 points by its validity. Thus, the total scores varied from 8 to 40 points. Two authors marked them independently and only included the articles that scored at least 20 points referring to the scoring system in Khosravi et al.’s ( 2024 ) systematic review article. Two researchers evaluated the articles with high inter-rater reliability (k = 0.8) and all results were eligible. 2.4 Data extraction and thematic analysis The data extraction was conducted through a rigorous application of thematic analysis, following the guidelines proposed by Braun and Clarke ( 2006 ). Thematic analysis is a qualitative method of recognizing and forming themes within data and we positioned our thematic analysis as primarily descriptive with interpretive elements, concentrating on summarizing and categorizing patterns related to gender bias in AI and also exploring the underlying meanings of these patterns (Braun & Clarke, 2022 ). This study employed both inductive and deductive approaches to guarantee a comprehensive understanding of the data. The data extraction process began with a thorough review of all selected articles. Each article was read several times to identify relevant text segments related to gender bias in AI. These segments were extracted based on their connection to research questions.The inductive approach suggests that the theme or category in research is totally data-driven (Wæraas, 2022 ), while the deductive approach is based on the preexisting theoretical frameworks. (Fife & Gossner, 2024 ). During the inductive phase, the author conducted open coding, where text segments were labeled with codes. These codes were then grouped into preliminary categories based on their similarities. The deductive process began with five categories derived from the literature (e.g. Hall & Ellis, 2023 ; Jeon et al., 2023 ), including basic information (e.g. year, publication journal), methodology (e.g. research design), domains of gender bias of AI (e.g. decision-making system, AI-generated contents), causes of gender bias of AI (e.g. biased data), mitigation approaches of gender bias (e.g. debiasing techniques) and consequences of gender bias of AI (e.g. social inequality). However, the five categories could not cover the finalized codes, so the categories of challenges of mitigating gender bias were introduced, forming the final code book. This iterative process of refining the coding framework is consistent with the principles of the thematic analysis, which emphasize flexibility and responsiveness to the data (Braun & Clarke, 2006 ). To ensure the reliability of the coding process, the second author independently coded using the same code book. If two authors coded differently, they would discuss, finally forming three themes, including publication aspect, research design and research focus aspect (Table 3 ). The inter-coder reliability was assessed by comparing the consistency of codes assigned to the same study by two or more coders (Kurasaki, 2000 ). A high percentage agreement of 98% was observed, indicating strong consistency in the coding process (Belur et al., 2021 ). The results of the coding schema were shown in detail in Table 3 . Table 3 Coding schema Theme Category Definition Sub-category Included studies Publication aspect Basic information Information about the publication Year of publication All 29 results Journal of publication Research design aspect Methodology Method of systematically designing a study to ensure reliable results Quantitative study Cecere et al., 2024 ; Chung et al., 2021 ; Górska & Jemielniak, 2023 ; Gupta et al., 2022 ; Hort et al., 2023 ; Kaplan et al., 2024 ; Kelley et al., 2022 ; Larrazabal et al., 2020 ; Noguero et al., 2023 ; Park et al., 2022 ; Rizhinashvili et al., 2022 ; Yoder-Himes et al., 2022 Qualitative study Adams & Loideain, 2019 ; Buslón et al., 2023 ; Gross, 2023 ; Howard & Borenstein, 2017 ; Manasi et al., 2022 ; Newstead et al., 2023 ; O’Connor & Liu, 2023 ; Wang, 2020 Mixed study Bernagozzi et al., 2021 ; Dwivedi et al., 2023 ; Fang et al., 2024 ; Gagandeep et al., 2024; Nah et al., 2024 ; Oca et al., 2023 ; Prates et al., 2020 ; Sun et al., 2023 ; Wang et al., 2023 Research focus aspect Domains of gender bias of AI The specific situations or general AI technologies reflecting AI’s gender bias Health and medicine Buslón et al., 2023 ; Chung et al., 2021 ; Larrazabal et al., 2020 Translation Bernagozzi et al., 2021 ; O’Connor & Liu, 2023 ; Prates et al., 2020 AI assistant Adams & Loideain, 2019 ; Gross, 2023 ; Manasi et al., 2022 ; Wang, 2020 AI-generated content Fang et al., 2024 ; Górska & Jemielniak, 2023 ; Kaplan et al., 2024 ; Nah et al., 2024 ; Newstead et al., 2023 ; Sun et al., 2023 Recommender system Cecere et al., 2024 ; Gupta et al., 2022 ; Oca et al., 2023 ; Wang et al., 2023 Decision-making system Gagandeep et al., 2024; Hort et al., 2023 ; Kelley et al., 2022 ; Assessment system Noguero et al., 2023 ; Park et al., 2022 Recognition system Howard & Borenstein, 2017 ; Rizhinashvili et al., 2022 ; Yoder-Himes et al., 2022 Large Language Model Dwivedi et al., 2023 NLP O’Connor & Liu, 2023 Causes of gender bias of AI Reasons behind AI’s gender bias Data bias Buslón et al., 2023 ; Chung et al., 2021 ; Dwivedi et al., 2023 ; Larrazabal et al., 2020 ; O’Connor & Liu, 2023 ; Prates et al., 2020 ; Sun et al., 2023 ; Wang et al., 2023 ; Yoder-Himes et al., 2022 Human bias Fang et al., 2024 ; Górska & Jemielniak, 2023 ; Manasi et al., 2022 ; Nah et al., 2024 ; Newstead et al., 2023 ; Wang et al., 2023 Algorithmic bias Cecere et al., 2024 ; Górska & Jemielniak, 2023 ; Kelley et al., 2022 ; Manasi et al., 2022 ; O’Connor & Liu, 2023 Social bias Dwivedi et al., 2023 ; Oca et al., 2023 ; O’Connor & Liu, 2023 ; Wang, 2020 ; Wang et al., 2023 Mitigation approaches of AI’s gender bias Strategies for reducing gender bias Data level Buslón et al., 2023 ; Cecere et al., 2024 ; Chung et al., 2021 ; Gross, 2023 ; Gagandeep et al., 2024; Kelley et al., 2022 ; Manasi et al., 2022 ; Park et al., 2022 ; Rizhinashvili et al., 2022 Algorithmic and model level Gross, 2023 ; Dwivedi et al., 2023 ; Fang et al., 2024 ; Kaplan et al., 2024 ; Manasi et al., 2022 ; Noguero et al., 2023 ; O’Connor & Liu, 2023 ; Park et al., 2022 ; Sun et al., 2023 Ethical and social level Adams & Loideain, 2019 ; Buslón et al., 2023 ; Gross, 2023 ; Górska & Jemielniak, 2023 ; Manasi et al., 2022 ; O’Connor & Liu, 2023 ; Prates et al., 2020 ; Sun et al., 2023 ; Wang, 2020 Assessment and diagnosis Adams & Loideain, 2019 ; Bernagozzi et al., 2021 ; Fang et al., 2024 ; Howard & Borenstein, 2017 ; Kaplan et al., 2024 ; Larrazabal et al., 2020 ; Manasi et al., 2022 ; Newstead et al., 2023 ; O’Connor & Liu, 2023 ; Sun et al., 2023 ; Wang et al., 2023 Challenges of mitigating AI’s gender bias Obstacles in reducing AI’s gender bias Performance of models Bernagozzi et al., 2021 ; Hort et al., 2023 ; Sun et al., 2023 Culture and value Adams & Loideain, 2019 ; Buslón et al., 2023 ; Howard & Borenstein, 2017 ; Kaplan et al., 2024 Ethical concerns Dwivedi et al., 2023 ; O’Connor & Liu, 2023 ; Sun et al., 2023 ; Wang et al., 2023 Consequences of gender bias of AI Negative effects of AI’s gender bias Inequalities amplification Buslón et al., 2023 ; Gross, 2023 ; Górska & Jemielniak, 2023 ; Oca et al., 2023 Gender stereotypes Adams & Loideain, 2019 ; Dwivedi et al., 2023 ; Górska & Jemielniak, 2023 Unequal treatment Gross, 2023 ; Górska & Jemielniak, 2023 ; Yoder-Himes et al., 2022 Table 4 Publication quantity and topic by journal Group Journal N Topic 1 ACM Transactions on Interactive Intelligent Systems; Applied Soft Computing; Information Systems Frontiers; IEEE Internet Computing; Neural Computing and Applications; Australasian Journal of Information Systems; Journal of Computer-Mediated Communication; Electronics 8 Computer science and information technology 2 AI&Society; Gender, Technology and Development; Rupkatha Journal on Interdisciplinary Studies in Humanities; Feminist Media Studies; Organizational Dynamics 5 Social sciences and humanities 3 Social Sciences; Scientific Reports; Science And Engineering Ethics; Proceedings of the National Academy of Sciences of the United States of America; Technological Forecasting & Social Change 5 Multidisciplinary sciences 4 Cureus; Frontiers in Global Women’s Health; Frontiers in Physiology 3 Biomedicine and health 5 Journal of Medical Internet Research; Manufacturing & Service Operations Management 3 Engineering and management 6 Cambridge International Law Journal 1 Law 7 International Journal of Communication; JMIR Formative Research; Multimedia Tools and Applications 3 Media and Communication 8 Frontiers in Education 1 Education Total 29 8 3 Results After analyzing 29 articles, the section reported the results from the publication aspect, research design and research focus aspect in the field of AI’ s gender bias. 3.1 Publication aspect 3.1.1 Publication years Figure 2 illustrated the annual publication distribution, revealing a progressive increase in scholarly attention to gender bias in AI systems. After the first article was published in 2017, no article was documented in 2018. The number of articles remained relatively low from 2019 to 2021 but appeared to have nearly doubled growth in 2022 and 2023. The latest publications counted on May 3, 2024 were 4, which surpassed the total annual publications before 2022, indicating a sustained growth trend. 3.1.2 Journal of publication The 29 results were mostly from different sources and only two of them were in the same journal called the Journal of Medical Internet Research. According to the main aims of these journals, the publications on gender bias of AI were categorized into 8 general topics, including computer science and information technology, social sciences and humanities, multidisciplinary sciences, biomedicine and health, engineering and management, law, media and communication and education. 3.2 Research design aspect Regarding research methodology (Fig. 3 ), 8 studies were qualitative and the majority of them utilized case studies to describe and explain how AI contributes to gender prejudice in particular fields (e.g. O’Connor & Liu, 2023 ; Wang, 2020 ). 12 results adopted a quantitative research design, most of which were about testing the proficiency and accuracy of models through concrete statistical parameters to mitigate AI’s gender bias (e.g. Chung et al., 2021 ; Kelley et al., 2022 ). The mixed method was used in 9 articles that quantitatively analyzed the output of AI-generated content to show how gender bias was manifested and qualitatively interpreted the results (e.g. Fang et al., 2024 ; Prates, 2020; Sun et al., 2023 ). 3.3 Research focus aspect 3.3.1 Domains of gender bias in AI The 29 results involved 10 domains that were classified in Fig. 4 . AI is a technology that simulates human intelligence and Natural Language Processing (NLP) is a core subfield of AI that studies how to enable computers to understand and process human language. LLM, on the other hand, is a crucial technology in NLP and the classification of the application is the the actual and specific application direction of NLP/LLM. For example, if an article is about the technical principles of NLP, it will be classified as NLP, and if it is a study of the application of NLP in medical diagnosis, it will be classified as an application layer. In terms of these 10 branches, it could be concluded that the discussion of gender bias of AI in the selected articles was conducted from the perspectives of both macroscopic technologies and microscopic applications. The most common research was on the gender bias of AI-generated text/image (N = 6), followed by the recommender system (N = 4), AI assistant (N = 4), health and medicine (N = 3), decision-making system (N = 3), recognition system (N = 3), assessment system (N = 2), translation (N = 3), LLM (N = 1) and NLP (N = 1). AI-generated text/image was discussed in 6 papers. Fang et al. ( 2024 ) and Nah et al. ( 2024 ) analyzed the gender bias in AI-generated news by providing broad topics or existing titles in previous news. The former results showed a higher gender bias in AI-generated news, but the latter found that the bias in AI-generated news was not less than in human-made ones. For AI-generated images, there was an imbalance between AI-generated males and females in some professions, with males being disproportionately high (doctor, lawyer, engineer, scientist), indicating a bias toward females (e.g. Górska & Jemielniak, 2023 ; Sun et al., 2023 ). For different application systems, 3 studies focused on recommender systems that addressed specific areas of recommendation, including healthcare, career guidance, and advertising (Cecere et al., 2024 ; Oca et al., 2023 ; Wang et al., 2023 ). The decision-making system and assessment system were primarily applied in the hiring and health fields respectively (e.g. Gagandeep et al., 2024; Park et al., 2022 ; Wang, 2020 ). In terms of the recognition system, Howard and Borenstein ( 2017 ) verified the low accuracy of the voice recognition system for female voices. The same year saw the identification of gender bias within facial recognition systems (Yoder-Himes et al., 2022 ). Regarding other more concrete domains, voice-based and text-based assistants were explored, such as Apple’s Siri, Google, ChatGPT and Microsoft’s Cortana (Adams & Loideáin, 2019; Manasi et al., 2022 ; Wang, 2020 ). The default use of female voices in most voice assistants reinforced gender stereotypes, positioning women as submissive or service-oriented roles. Notably, Google was the only major platform that avoided this gendered default (Adams & Loideain, 2019 ). ChatGPT, mainly based on texts, attributed different personalities to different genders and did not provide any traits for gender-diverse people (Gross, 2023 ). In health and medicine, studies have paid attention to the negative effect of AI on the gender ratio of employees in this occupation and declared the relatively low performance of models when trained with data from female patients (Buslón et al., 2023 ; Chung et al., 2021 ; Larrazabal et al., 2020 ). In terms of translation, most studies centered on the Google translator (e.g. O’Connor & Liu, 2020; Prates et al., 2020 ) and tested its gender bias usually through the sentence form: He/She is + occupation. The results suggested that the translator did not present the real distribution of gender in some jobs, especially in STEM. 3.3.2 Causes of gender bias of AI A total of 25 articles were analyzed to identify the primary factors contributing to gender bias in AI. The findings were categorized into four distinct causes: data bias (N = 9), human bias (N = 6), algorithmic bias (N = 5) and social bias (N = 5). Data bias was the most influential factor, which suggested that gender bias would occur when the model was trained without gender-balanced data (Chung et al., 2021 ). Dwivedi et al. ( 2023 ) claimed that gender bias was not generated from air but from data. For instance, men’s faces were recognized more easily than women’s faces possibly because more male images were collected in the training set (Yoder-Himes et al., 2022 ). Additionally, data leakage, a phenomenon where information from the test set inadvertently influences the training process, can introduce gender bias. O’Connor and Liu ( 2023 ) gave examples of presenting a dataset featuring an equal representation of women and men engaged in cooking activities. While this balance did not inherently introduce bias, the presence of a child in the image may lead to bias. Given that children were frequently depicted alongside women rather than men in various images, the model might link “children” with “cooking.” Consequently, this could result in women being disproportionately categorized as “cooking” compared to men. The essence of data bias was human bias (Wang et al., 2023 ), which was in line with Nah et al.’s ( 2024 ), Manasi et al.’s ( 2022 ) and Newstead et al.’s ( 2023 ) studies, indicating that designers and creators could choose preferred trained data due to gender stereotypes or commercial interests, along with their own biases. In the present research, human bias and social bias were similar but coded differently based on their distinct origins and manifestations. Human bias mainly referred to the prejudices and stereotypes held by individuals in the design and development of AI systems, such as developers, data scientists and engineers. Nah et al.’s ( 2024 ) study claimed that developers may introduce bias by prioritizing certain keywords or perspectives that met their own beliefs, which resulted in AI systems producing news that associated abortion with women-related keywords. Such bias operated at the individual level and was directly tied to the actions and decisions of AI creators. While social bias mainly involved the broader, historically remaining prejudices that existed within society. These biases were rooted in cultural norms, institutional practices, and social structures, which shaped the collective behaviors and attitudes of individuals. Wang et al.’s ( 2023 ) research explored AI systems designed to recommend careers to students, revealing such systems often recommended computer science or criminal justice careers to males and nursing or English to females. This result reflected deeply ingrained societal stereotypes that connected men to technical fields and women to caregiving or language-related domains. As for algorithmic bias, it was highlighted in Manasi et al.’s ( 2022 ) study. O’Connor and Liu’s ( 2023 ) experiments identified the bias in word embedding, which was a fundamental component in the NLP system and trained on a large number of text data. Word2Vec is a widely used word embedding model and learns semantic representations of words by analyzing their co-occurrence patterns in large text corpora (Mehmood et al., 2020). However, these data might contain gender bias inherently. O’Connor and Liu (2022) claimed that even gender-neutral words can become semantically correlated with a particular gender due to societal stereotypes encoded in the training data. For example, words like “nurse” and “homemaker,” though grammatically gender-neutral, are often associated with female pronouns in word embedding. If the descriptions of a particular gender were more positive or negative in the data, these biases would also be encoded into the embedding. 3.3.3 Mitigation approaches of gender bias Gender bias in AI could be mitigated from 4 perspectives, including data level (N = 9), algorithmic and model level (N = 9), ethical and social level (N = 9) and assessment and diagnosis (N = 11). These approaches were not designed to directly correspond to the four causes of bias. The last approach, an indirect way of reducing gender bias, was to identify and evaluate bias, which could be applicable across all types of bias reviewed and provide a comprehensive framework for addressing AI’s gender bias. Regarding data level, two critical processes were identified, specifically during the phases of data collection and data processing. For data collection, Buslón et al. ( 2023 ) recommended collecting diverse data to reflect adequate demographic features in the medicine and health field. Furthermore, recent studies have explored advanced data preprocessing techniques, such as gender masking, which involved removing gender-related information from textual data while maintaining semantic integrity. This technique resembled the approach outlined in Rizhinashvili et al.’s ( 2022 ) research, where gender parameters were eliminated to transform text into word embedding without any gender-related semantic information and this technique has been shown to reduce gender bias in the resume filtering system (Gagandeep et al., 2024). Additionally, most studies kept an eye on the data preprocessing before data training. The latest study verified the effect of preprocessing advertisement text in reducing gender bias in recommendation systems, like designing suitable text length (Cecere et al., 2024 ). The primary objective was to prevent the model from capturing and amplifying gender stereotypes or prejudicial information embedded in the text through careful adjustment of text length. Effective data preprocessing techniques included downsampling and upsampling, etc. (Kelley et al., 2022 ), whose roles were to balance gender data to create fair models (Park et al., 2022 ). One study explored the effect of applying upsampling and downsampling in the Neutral Machine Translation system, presenting that these two ways could diminish gender bias but lead to a significant decline in overall system performance (Tomalin et al., 2021 ). At the algorithmic and model level, it was tightly connected with the mitigation method from the data level. Because proper collection and processing of the data benefited the training of unbiased models. Various studies have underscored the necessity of modifying prompts in interactions with AI to offer more neutral and higher-quality outputs (Dwivedi et al., 2023 ; Fang et al., 2024 ; Kaplan et al., 2024 ). The prompt limit could also be interpreted as alleviating bias at the data level. From a purely model-training aspect, Noguero et al. ( 2023 ) were devoted to training fair and calibrated classifiers and the results proved the isotonic calibrator accurate. This post-processing technique adjusted the output probabilities of classifiers to align more closely with the actual probability distribution. Additionally, an innovative approach was introduced through a debiasing algorithm (Sun et al., 2023 ), which aimed to eliminate gender pair associations for gender-neutral terms while maintaining the utility of word embeddings in representing meaningful relationships and associations among words (O’Connor & Liu, 2023 ). For the ethical and social level, public awareness, regulations and policies of AI, and management of the AI team were three vital points based on the selected papers. Buslón et al. ( 2023 ) suggested that public awareness of AI’s gender bias should be enhanced and improved by education, which echoed Wang’s ( 2020 ) study, directly pinpointing the importance for girls to have an anti-bias education. Furthermore, some regulations and policies of AI were established in practice to guarantee human rights. The European Commission provided a framework for AI to define the responsibilities of its users and developers (DiNoia et al., 2022 ), and UNESCO ( 2020 ) held the view that gender equality should be incorporated into AI technology. For the management team, diversity was a priority. The tech company should enhance their gender equality and inclusion to let them participate in the “inner work” (Gross, 2023 ). Adams and Loideain’s ( 2019 ) study pointed to the urgency of reforming within the industry, revealing the significance of redistributing power in AI design (Górska & Jemielniak, 2023 ). Assessment and diagnosis related to identifying or auditing gender bias in the algorithm. Howard and Borenstein ( 2017 ) claimed it could be achieved by robot logic and this proposal was extended to design a computer-assisted diagnosis system in Larrazabal et al.’s ( 2020 ) study. In 2023, the idea of using AI to detect AI’s gender bias became popular (Newstead et al., 2023 ; Wang et al., 2023 ), suggesting that researchers expected to envisage a win-win situation of reducing the gender bias of AI and users as their interaction. 3.3.4 Challenges of mitigating gender bias Challenges of mitigating gender bias derived from the performance of models (N = 3), culture and value (N = 4) and ethical concerns (N = 4). Bernagozzi et al.’s ( 2021 ) study revealed the competition between achieving accuracy and mitigating gender bias of translators. Similarly, it was found that the single-objective search technique could largely reduce the gender bias of models but with sacrificing semantic correctness (Hort et al., 2023 ), signaling that overemphasizing gender neutrality may compromise model performance and yield impractical outcomes (Dwivedi et al., 2023 ). The cultural and value-related challenge implied that individuals’ stubborn attitudes or stereotypes could impact the decision-making systems of AI unconsciously, which were tough to remove (Howard & Borenstein, 2017 ). Ethical concerns pertained to the accountability and transparency of data (O’Connor & Liu, 2023 ) as well as the design process. According to Sun et al. ( 2023 ), enhancing the transparency of AI products through the involvement of researchers from diverse fields and the general public in collaborative decision-making could mitigate the risk of reinforcing existing gender biases and reduce the dominance of powerful stakeholders in AI technology. 3.3.5 Consequences of gender bias of AI Consequences of gender bias ranged from amplification of existing inequalities (N = 4) and gender stereotypes (N = 3) to unequal treatment (N = 3), with these adverse effects predominantly impacting females rather than males. In the realm of health and medical research, scholars have noted that gender bias in AI could lead to the perpetuation of inequality and discrimination within society, particularly affecting women and other marginalized groups, thereby significantly influencing both healthcare systems and individual lives (Buslón et al., 2023 ). The underrepresentation of women in specific roles depicted in AI-generated images has intensified this inequality (Fang et al., 2024 ), a trend also observed in AI-generated texts (Gross, 2023 ). These AI-based applications would perpetuate gender stereotypes to users and normalize them (Dwivedi et al., 2023 ; Górska & Jemielniak, 2023 ). As a result, women faced unequal treatment and were disadvantaged within society (Gross, 2023 ), a phenomenon explored in Yoder-Himes et al.’s ( 2022 ) study, which revealed that female students scored lower on exams due to biases in facial recognition technology. 4 Discussion 4.1 Publication and research design aspect (RQ1 and RQ2) RQ1 revealed the overall increasing research trend of AI’s gender bias and its cross-disciplinary characteristics. Based on data collected up to May 2024, the number of studies surged from 2022 to 2023 and reached its highest observed count in 2023. This surge may be attributed to the release of ChatGPT at the end of 2022, a groundbreaking language model from OpenAI that gained widespread adoption in 2023, reportedly amassing over 180 million users (Ghassemi et al., 2023 ). The relevant studies were largely conducted, including the technological advancement and model optimization of ChatGPT (Aljebreen et al., 2023 ) and ethical, privacy and risk considerations of ChatGPT (Ali & Djalilian, 2023 ), etc. RQ2 elucidated the research methodology, an aspect that had not been addressed in earlier reviews. Our findings indicated that the majority of studies employed quantitative methods to statistically assess the performance and accuracy of models utilizing gender-imbalanced datasets. In the field of AI, gender bias is often represented by large-scale datasets and algorithmic models. Therefore, to accurately assess and quantify the degree of gender bias, researchers usually rely on quantitative studies to analyze these data by counting and analyzing the gender distribution and tendency in data sets (Yoder-Himes et al., 2022 ; Noguero et al., 2023 ). 4.2 Research focus aspect 4.2.1 RQ3: What are the domains that present gender bias in AI? The domains that presented gender bias in AI were detected. Previous research (Shrestha & Das, 2022 ) examined areas such as NLP, recommender systems, decision-making systems, AI assistants, and medicine. Our study expanded this investigation to include facial recognition, translation, assessment systems, and AI-generated text/image, which have recently garnered increased scholarly attention, especially after the launch of ChatGPT, sparking a global interest in more diverse applications of AI technologies. Thus, the results of RQ3 became broader with the updated research time. Given the significance of facial recognition as a vital sector of AI, the issues of gender bias within this area have been prioritized for further exploration, especially in light of ChatGPT’s release. Besides, the impressive capabilities of ChatGPT in NLP have spotlighted gender bias in translation and encouraged researchers to consider reducing gender bias in translation systems to enhance their accuracy and fairness (Gross, 2023 ). Furthermore, ChatGPT’s proficiency in communication and its immense potential of AI in generation also stirred reflection on gender bias in the assessment system and AI-generated text/image domain (Nah et al., 2024 ; Newstead et al., 2023 ). 4.2.2 RQ4: What causes of AI’s gender bias are proposed? RQ4 explored the causes of AI’s gender bias, with human bias and data bias identified as the primary sources. These findings were consistent with Nadeem et al.’s ( 2022 ) study, which explored AI’s gender bias in decision-making systems. Algorithmic bias and social bias were not highlighted specifically in many studies, which could be related to the tight and complex connection of different causes of gender bias. Algorithmic bias is often viewed as a consequence of human bias and data bias (Selbst et al., 2019 ). In other words, algorithmic bias manifests as a result of human bias and data bias being embedded in the algorithm’s design and training. The relationship of social bias and human bias is complex but interconnected. Social bias can be seen as a foundational factor that contributes to the formation of human bias. Besides, social bias is not viewed as a direct cause of bias and usually serves as a background factor that influences both human behavior and data collection practices. For instance, societal gender stereotypes may shape the way data is labeled or the way algorithms are designed (Vallor, 2024 ). Therefore, human bias and data bias are the primary sources of gender bias in AI and are explored most, accounting for 60% of the selected papers. 4.2.3 RQ5: What mitigation approaches for AI’s gender bias are proposed? For the mitigation method, the study reviewed four kinds of methods. The solution of assessment was discussed frequently recently (Kaplan et al., 2024 ; Newstead et al., 2023 ). Regular assessment and diagnosis could systematically uncover gender bias that may be deeply embedded in AI systems and not readily apparent. The detection tool was identified as AI itself most (Newstead et al., 2023 ; Wang et al., 2023 ). Bellamy et al.’s ( 2018 ) study introduced the AI Fairness 360 (AIF360) package, which mainly aimed to enhance the understanding of fairness metrics and mitigation strategies, providing a shared platform for fairness researchers and industry professionals to exchange and evaluate their algorithms. Later, Mosteiro et al. ( 2022 ) applied the AIF 360 package to mitigate gender biases in clinical psychiatry. Furthermore, the emerging AI technologies have further expanded the scope of traditional mitigation methods mentioned above, offering innovative solutions to address gender bias. For instance, federated learning enhanced data-level mitigation by enabling the development of more diverse and representative datasets through decentralized training, thereby reducing the risk of bias (Gupta et al., 2024 ; Han et al., 2024 ; Kim et al., 2024 ). Compared with the traditional methods of mitigating bias at the data level, such as data clarity and data balance, federated learning was an extended method, more focused on distributed data training. Besides, at the algorithmic level, adversarial debiasing methods actively counteracted biases by training models to minimize discriminatory outcomes, complementing traditional algorithmic fairness techniques (Correa et al., 2023 ). DSouza and French ( 2024 ) found that adversarial debiasing had the capacity to mitigate the negative effects of unfavorable attributes within a dataset, while also showing considerable promise in decreasing the production of fake news that arose from the intrinsic biases presented in the data. These new technologies not only complemented traditional mitigation methods but also opened new avenues for addressing gender bias in AI systems. 4.2.4 RQ6: What are the challenges of mitigating AI’s gender bias? RQ6 showed the challenges of mitigating gender bias in AI, ranging from the performance of models, culture and value to ethical concerns. Certain algorithmic approaches to achieving unbiased outputs may conflict with high performance, particularly when relying on extensive training iterations or complex model architectures. For instance, some studies suggested that excessive data usage in pursuit of unbiased models could lead to performance degradation, a phenomenon referred to as “Model Autophagy Disorder” (MAD) (Alemohammad et al., 2023 ). The negative influence from the perspective of culture and value corroborated the idea of Tadimalla and Maher ( 2024 ), who claimed that the direction of AI development was determined by its creators, including individuals and organizations (Vallor, 2024 ). The results of ethical concerns were mainly due to the nature of the “black box” in AI systems for technological and ethical limits, indicating its inscrutable and complex features (Asatiani et al., 2020 ; Marcus & Teuwen, 2024 ). Karasinski and Candiotto ( 2024 ) addressed that the black box prevented the indexation of the deviation because the systems used to process information were not controlled by creators and were not understood, resulting in output with obvious bias, which will adversely affect the inclusion drive. For ethical limits, commercial and privacy considerations may contribute to AI becoming a black box, indicating that varied stakeholders had different needs for the transparency and accountability of AI systems (Yang & Zhou, 2023 ). 4.2.5 RQ7: What are the consequences of AI’s gender bias? Although AI is a powerful and useful technology, there are certain hazards for its gender bias (Newstead et al., 2023 ). The harms of amplifying inequalities and unequal treatment kept with the previous Donnelly and Stapleton’s ( 2021 ) research, pointing out that digital data systems facilitated social violence against marginalized communities and perpetuated gender bias. The consequences of gender stereotypes aligned with Fabris et al.’s ( 2020 ) study and Horvát and Bailón’s ( 2024 ) reports. The opacity of AI systems often hindered a direct understanding of the underlying logic behind their decisions. Consequently, when an AI system made a decision based on gender bias, it may be mistakenly perceived as reasonable or justified, thereby further reinforcing gender stereotypes (Fabris et al., 2020 ). If the AI model used historical biased data without correction, it may continue or even exacerbate historically gender inequality and the adverse consequences of gender bias were not just for the present, they may affect the process of gender equality in society in the long term (Franklin et al., 2024 ; Vallor, 2024 ). It is crucial to consider the mitigation methods and their challenges together, contributing to the ethical and responsible development of AI. 5 Conclusion 5.1 Major findings The AI system has become an integral part of our daily lives and gender bias in AI has become an apparent concern. The present study reviewed 29 papers published on gender bias in AI systems, which were coded within 7 categories through thematic analysis both inductively and deductively. An overall increasing research trend and the distribution of research methodology were identified. We also detailed the domains that reflect AI’s gender bias. Finally, we presented the causes, consequences, mitigation methods and challenges of reducing bias from multiple aspects. 5.2 Limitations This study acknowledges certain limitations. First, different keywords result in a varied pool of research papers. Though we established two groups of search strings referring to previous reviews, it was not enough, especially the keywords about AI. AI is a broad and general term, and retrieving related research comprehensively seems demanding, potentially resulting in the omission of some articles. Second, the classification of categories in bias causes, mitigation methods or consequences of AI’s gender bias is somewhat overlapping and connected, meaning that it is hard to clearly and wholly separate them and use them to label selected articles. 5.3 Further research directions Though gender bias can not be eliminated but mitigated (Solans et al., 2023 ), the process of mitigating bias is heavily obstructed by challenges, resulting in a dilemma that can not be easily solved. Further studies could concentrate on these challenges specifically. To balance model performance and biased output, it is essential to reconsider the acceptable levels of gender bias and establish the appropriate thresholds for model accuracy (Kiyasseh et al., 2023 ). For the contradiction between data transparency and privacy protection, it is important to explore the development of new machine learning algorithms that can effectively train AI models while safeguarding privacy and minimizing the risk of gender bias. Thus, further investigation into the efficacy of emerging techniques, such as differential privacy and federated learning, which allow for model training across diverse data sources while ensuring individual privacy, is warranted (Wang et al., 2023 ; Yang et al., 2023 ). Furthermore, there is growing attention to AI’s gender bias, but research regarding how individual factors shape perceptions and attitudes towards this bias remains limited. Gupta et al. ( 2022 ) verified that people with a culture of collectivism and uncertainty avoidance are more inclined to question AI’s gender bias. A recent study examined the influence of the social identity of individuals on treating biased AI (Ratajczak & Cockerill, 2023 ), but it focused on racial bias. Thus, further studies could proceed to explore the individual factors affecting the perception of gender bias. It is conducive to the development of policies and norms and system optimization, indicating that by understanding different personal factors, the designers can know how to improve, thus reducing bias and improving user satisfaction. Declarations Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Conflict of interest/Competing interests: The author declares there are no competing interests. Consent to Publish declaration: Not applicable. The manuscript does not contain any individual person’s data in any form. Consent to Participate declaration: Not applicable. The study did not involve human participants. Ethics approval statement: Not applicable. Data Availability statement No. Author Contribution Jingyue Liu: Conceptualization, Methodology, Investigation, Editing, and Writing-Original Draft. References Adams, R., & Loideain, N. N. (2019). Addressing indirect discrimination and gender stereotypes in AI virtual personal assistants: The role of international human rights law. Cambridge International Law Journal , 8 (2), 241–257. https://doi.org/10.4337/cilj.2019.02.04 Alami, H., Lehoux, P., Auclair, Y., de Guise, M., Gagnon, M., Shaw, J., Roy, D., Fleet, R., Ag Ahmed, M.A., & Fortin, J. (2020). Artificial Intelligence and Health Technology Assessment: Anticipating a New Level of Complexity. Journal of Medical Internet Research , 22 . https://doi.org/10.2196/preprints.17707 Alemohammad, S., Casco-Rodriguez, J., Luzi, L., Humayun, A., Babaei, H. R., LeJeune, D., Siahkoohi, A., & Baraniuk, R. (2023). Self-Consuming Generative Models Go MAD. ArXiv , abs/2307.01850 . https://doi.org/10.48550/arXiv.2307.01850 Ali, M. J., & Djalilian, A. (2023). Readership Awareness Series – Paper 4: Chatbots and ChatGPT-Ethical Considerations in Scientific Publications. Seminars in Ophthalmology , 38 (5), 403–404. https://doi.org/10.1080/08820538.2023.2193444 Aljebreen, M., Alabduallah, B., Asiri, M. M., Salama, A. S., Assiri, M., & Ibrahim, S. S. (2023). Moth Flame Optimization With Hybrid Deep Learning Based Sentiment Classification Toward ChatGPT on Twitter. IEEE Access , 11 , 104984–104991. https://doi.org/10.1109/ACCESS.2023.3315609 Asatiani, A., Malo, P., Nagbøl, P. R., Penttinen, E., Rinta-Kahila, T., & Salovaara, A. (2020). Challenges of Explaining the Behavior of Black-Box AI Systems. MIS Quarterly Executive , 19 (7), 259–278. https://doi.org/10.17705/2MSQE.00037 Baker, R. S., & Hawn, A. (2021). Algorithmic bias in education. Int. J. Artif. Intell. Educ. 32 , 1052–1092. https://doi.org/10.1007/s40593-021-00285-9 Bellamy, R. K., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., Lohia, P. K., Martino, J., Mehta, S., Mojsilovic, A., Nagar, S., Ramamurthy, K. N., Richards, J. T., Saha, D., Sattigeri, P., Singh, M., Varshney, K. R., & Zhang, Y. (2018). AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias. ArXiv , abs/1810.01943 . https://doi.org/10.48550/arXiv.1810.01943 Belur, J., Tompson, L., Thornton, A., & Simon, M. (2021). Interrater reliability in systematic review methodology: Exploring variation in coder decision-making. Sociological Methods & Research, 50(2), 837–865. https://doi.org/10.1177/0049124118799372 Bernagozzi, M., Srivastava, B., Rossi, F., & Usmani, S. (2021). Gender Bias in Online Language Translators: Visualization, Human Perception, and Bias/Accuracy Tradeoffs. IEEE Internet Computing , 25 , 53–63. https://doi.org/10.1109/mic.2021.3097604 Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology , 3 (2), 77–101. https://doi.org/10.1191/1478088706qp063oa Braun, V., & Clarke, V. (2022). Toward good practice in thematic analysis: Avoiding common problems and be(com)ing a knowing researcher. International Journal of Transgender Health , 24 (1), 1–6. https://doi.org/10.1080/26895269.2022.2129597 Brown, L. M. (2022). Gendered Artificial Intelligence in Libraries: Opportunities to Deconstruct Sexism and Gender Binarism. Journal of Library Administration , 62 (1), 19–30. https://doi.org/10.1080/01930826.2021.2006979 Buslón, N., Cortés, A., Catuara-Solarz, S., Cirillo, D., & Rementeria, M. J. (2023). Raising awareness of sex and gender bias in artificial intelligence and health. Frontiers in global women’s health , 4 , Article 970312. https://doi.org/10.3389/fgwh.2023.970312 Cecere, G., Jean, C.†., Le Guel, F., & Manant, M. (2024). Artificial intelligence and algorithmic bias? Field tests on social network with teens. Technological Forecasting and Social Change , 201 , Article 123204. https://doi.org/10.1016/j.techfore.2023.123204 Chang, X. (2023). Gender Bias in Hiring: An Analysis of the Impact of Amazon’s Recruiting Algorithm. Advances in Economics. Management and Political Sciences , 23 (1), 134–140. https://doi.org/10.54254/2754-1169/23/20230367 Chen, F., Wang, L., Hong, J., Jiang, J., & Zhou, L. (2024). Unmasking bias in artificial intelligence: A systematic review of bias detection and mitigation strategies in electronic health record-based models. Journal of the American Medical Informatics Association , 31 (5), 1172–1183. https://doi.org/10.1093/jamia/ocae060 Chen, Y. B., Zhai, Y. J., & Sun, S. J. (2024). The gendered lens of AI: Examining news imagery across digital spaces. Journal of Computer-Mediated Communication , 29 (1), zmad047. https://doi.org/10.1093/jcmc/zmad047 Chu, C. H., Donato-Woodger, S., Khan, S. S., Nyrup, R.,·Leslie, K., Lyn, A., Shi, T. Y., Bianchi, A., Rahimi, S. A., & Grenier, A. (2023). Age-related bias and artificial intelligence: A scoping review. Humanities and Social Sciences Communications , 10 , Article 510. https://doi.org/10.1057/s41599-023-01999-y Chung, H., Park, C., Kang, W. S., & Lee, J. (2021). Gender Bias in Artificial Intelligence: Severity Prediction at an Early Stage of COVID-19. Frontiers in physiology , 12 , Article 778720. https://doi.org/10.3389/fphys.2021.778720 Coiro, P., & Pollak, D. D. (2019). Sex and gender bias in the experimental neurosciences: The case of the maternal immune activation model. Trans itional Psychiatry , 9 , Article 90. https://doi.org/10.1038/s41398-019-0423-8 Correa, R., Pahwa, K., Patel, B., Vachon, C. M. (2023). Efficient Adversarial debiasing with Concept Activation Vector—Medical image case-studies. Journal of Biomedical Informatics , 149 (11), Article 104548. https://doi.org/10.1016/j.jbi.2023.104548 Daneshjou, R., Smith, M. P., Sun, M. D., Rotemberg, V., & Zou, J. (2021). Lack of Transparency and Potential Bias in Artificial Intelligence Data Sets and Algorithms: A Scoping Review. JAMA dermatology , 157 (11), 1362–1369. https://doi.org/10.1001/jamadermatol.2021.3129 Delgado, J., de Manuel, A., Parra, I., Moyano, C., Rueda, J., Guersenzvaig, A., Ausin, T., Cruz, M., Casacuberta, D., & Puyol, A. (2022). Bias in algorithms of AI systems developed for COVID-19: A scoping review. Journal of Bioethical Inquiry , 19 (3), 407–419. https://doi.org/10.1007/s11673-022-10200-z DiNoia, T., Tintarev, N., Fatourou, P., & Schedl, M. (2022). Recommender systems under European AI regulations. Communications of the ACM , 65 (4), 69–73. https://doi.org/10.1145/3512728 Donnelly, N., & Stapleton, L. (2021). Digital Enterprise Technologies: Do Enterprise Control and Automation Technologies Reinforce Gender Biases and Marginalization? IFAC-PapersOnLine. 54 (13), 551–556. https://doi.org/10.1016/j.ifacol.2021.10.507 DSouza, K. M., & French, A. M. (2024). Fake news detection using machine learning: An adversarial collaboration approach. Internet Research , 34 (5), 1664–1678. https://doi.org/10.1108/INTR-03-2022-0176 Durán, R. P., Eisenhart, M., Erickson, F. D., Grant, C. A., Green, J. L., Hedges, L. V., Levine, F. J., Moss, P. A., Pellegrino, J. W., & Schneider, B. L. (2006). Standards for Reporting on Empirical Social Science Research in AERA Publications. Educational Researcher , 35 , 33–40. https://doi.org/10.3102/0013189X035006033 Dwivedi, S., Ghosh, S., & Dwivedi, S. (2023). Breaking the Bias: Gender Fairness in LLMs Using Prompt Engineering and In-Context Learning. Rupkatha Journal on Interdisciplinary Studies in Humanities , 15 , Article 4. https://doi.org/10.21659/rupkatha.v15n4.10 Dwivedi, Y. K., Sharma, A., Rana, N. P., Giannakis, M., Goel, P., & Dutot, V.(2023). Evolution of artificial intelligence research in technological forecasting and social change: Research topics, trends, and future directions. Technological Forecasting and Social Change , 192 , Article 122579. https://doi.org/10.1016/j.techfore.2023.122579 Fabris, A., Purpura, A., Silvello, G., & Susto, G. (2020). Gender Stereotype Reinforcement: Measuring the Gender Bias Conveyed by Ranking Algorithms. Inf. Process. Manag. , 57 , Article 102377. https://doi.org/10.1016/j.ipm.2020.102377 Fang, X., Che, S. K., Mao, M. J., Zhang, H. Z., Zhao. M., & Zhao, X. F. (2024). Bias of AI-generated content: An examination of news produced by large language models. Sci. Rep ., 14 , Article 5224. https://doi.org/10.1038/s41598-024-55686-2 Fife, S. T., & Gossner, J. D. (2024). Deductive Qualitative Analysis: Evaluating, Expanding, and Refining Theory. International Journal of Qualitative Methods , 23 (6), 1–12. https://doi.org/10.1177/16094069241244856 Franklin, G., Stephens, R., Piracha, M., Tiosano, S., LeHouillier, F., Koppel, R., & Elkin, P. L.(2024). The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective. Life-basel , 14 (6), https://doi.org/10.3390/life14060652 Gagandeep, Kaur, J., Mathur, S., Kaur, S., Nayyar, A., Singh, S. P., & Mathur, S. (2024). Evaluating and mitigating gender bias in machine learning based resume filtering. Multimed Tools Appl ., 83 , 26599–26619. https://doi.org/10.1007/s11042-023-16552-x Galos, D. R., & Coppock, A. (2023). Gender composition predicts gender bias:A meta-reanalysis of hiring discrimination audit experiments. Sci. Adv., 9 (18), esde7979. https://doi.org/10.1126/sciadv.ade7979 Ghassemi, M., Birhane, A., Bilal, M., Kankaria, S., Malone, C., Mollick, E., & Tustumi, F. (2023). ChatGPT one year on: Who is using it, how and why? Nature , 624 , 39–41. https://doi.org/10.1038/d41586-023-03798-6 Górska, A. M., & Jemielniak, D. (2023). The invisible women: Uncovering gender bias in AI-generated images of professionals. Feminist Media Studies , 23 , 4370–4375. https://doi.org/10.1080/14680777.2023.2263659 Gregory, A., & Denniss, A. R. (2018). An Introduction to Writing Narrative and Systematic Reviews-Tasks, Tips and Traps for Aspiring Authors. Heart, lung & circulation , 27 (7), 893–898. https://doi.org/10.1016/j.hlc.2018.03.027 Gross, N. (2023). What ChatGPT Tells Us about Gender: A Cautionary Tale about Performativity and Gender Biases in AI. Social Sciences , 12 (8), Article 435. https://doi.org/10.3390/socsci12080435 Guevara-Gómez, A., Zárate-Alcarazo, L. O., & Criado, J. I. (2021). Feminist perspectives to artificial intelligence: Comparing the policy frames of the European Union and Spain. Inf. Polity , 26 (2), 173–192. https://doi.org/10.3233/IP-200299 Gupta, M., Parra, C. M., & Dennehy, D. (2022). Questioning Racial and Gender Bias in AI-based Recommendations: Do Espoused National Cultural Values Matter? Information systems frontiers: a journal of research and innovation , 24 (5), 1465–1481. https://doi.org/10.1007/s10796-021-10156-2 Gupta, A., Maurya, M. K., Dhere, K., & Chaurasiya, V. K. (2024). Privacy-Preserving Hybrid Federated Learning Framework for Mental Healthcare Applications: Clustered and Quantum Approaches. IEEE Access , 12 , 145054–145068. https://doi.org/10.1109/ACCESS.2024.3464240 Hall, P., & Ellis, D. (2023). A systematic review of socio-technical gender bias in AI algorithms. Online Inf. Rev., 47 , 1264–1279. https://doi.org/10.1108/oir-08-2021-045 Han, M., Zhu, T., & Zhou, W. (2024). Fair Federated Learning with Opposite GAN. Knowl. Based Syst. , 287 , Article 111420. https://doi.org/10.1016/j.knosys.2024.111420 Hort, M., Moussa, R., & Sarro, F. (2023). Multi-objective search for gender-fair and semantically correct word embeddings. Appl. Soft Comput., 133 , Article 109916. https://doi.org/10.1016/j.asoc.2022.109916 Horvát, E. A., & Bailón, S. G. (2024). Quantifying gender disparities and bias online: editors’ introduction to “Gender Gaps in Digital Spaces” special issue. Journal of Computer-Mediated Communication , 29 (1), zmad054. https://doi.org/10.1093/jcmc/zmad054 Howard, A. M., & Borenstein, J. (2017). The Ugly Truth About Ourselves and Our Robot Creations: The Problem of Bias and Social Inequity. Science and Engineering Ethics , 24 , 1521–1536. https://doi.org/10.1007/s11948-017-9975-2 Jeon, J., Lee, S., & Choi, S. (2023). A systematic review of research on speech-recognition chatbots for language learning: Implications for future directions in the era of large language models. Interactive Learning Environments , 1–19. https://doi.org/10.1080/10494820.2023.2204343 Jin, E., & Eastin, M. (2023). Gender Bias in Virtual Doctor Interactions: Gender Matching Effects of Chatbots and Users on Communication Satisfactions and Future Intentions to Use the Chatbot. International Journal of Human – Computer Interaction . https://doi.org/10.1080/10447318.2023.2279402 Kaplan, D. M., Palitsky, R., Arconada Alvarez, S. J., Pozzo, N. S., Greenleaf, M. N., Atkinson, C. A., & Lam, W. A. (2024). What’s in a Name? Experimental Evidence of Gender Bias in Recommendation Letters Generated by ChatGPT. Journal of medical Internet research , 26 , e51837. https://doi.org/10.2196/51837 Karasinski, M., & Candiotto, K. B. B. (2024). AI’s black box and the supremacy of standards. Filosofia Unisinos , 25 (1), 1–13. https://doi.org/10.4013/fsu.2024.251.13 Kelley, S., Ovchinnikov, A., Hardoon, D. R., & Heinrich, A. (2022). Antidiscrimination Laws, Artificial Intelligence, and Gender Bias: A Case Study in Nonmortgage Fintech Lending. Manufacturing and Service Operations Management , 24 (6), 2797–3306. https://doi.org/10.1287/msom.2022.1108 Khan, M., & Ewuoso, C. (2024). Epistemic (in)justice, social identity and the Black Box problem in patient care. Med Health Care and Philos., 27 , 227–240. https://doi.org/10.1007/s11019-024-10194-y Khosravi, M., Zare, Z., Mojtabaeian, S. M., & Izadi, R. (2024). Artificial Intelligence and Decision-Making in Healthcare: A Thematic Analysis of a Systematic Review of Reviews. Health Services Research and Managerial Epidemiology , 11, 1–15. https://doi.org/10.1177/23333928241234863 Kim, D., Woo, H., & Lee, Y. (2024). Addressing Bias and Fairness Using Fair Federated Learning: A Synthetic Review. Electronics , 13 (23), Article 4664. https://doi.org/10.3390/electronics13234664 Kiyasseh, D., Laca, J., Haque, T. F., Otiato, M., Miles, B. J., Wagner, C., Donoho, D. A., Trinh, Q., Anandkumar, A., & Hung, A. J. (2023). Human visual explanations mitigate bias in AI-based assessment of surgeon skills. npj Digit. Med., 6 , Article 54. https://doi.org/10.1038/s41746-023-00766-2 Koo, S. (2022). Leading or Cheer-Leading? The Gender Gap in Political Smiles. Politics & Gender , 18 (1), 183–211. https://doi.org/10.1017/S1743923X20000379 Kurasaki, K. S. (2000). Intercoder Reliability for Validating Conclusions Drawn from Open-Ended Interview Data. Field Methods , 12 (3). https://doi.org/10.1177/1525822X0001200301 Larrazabal, A. J., Nieto, N., Peterson, V., Milone, D. H., & Ferrante, E. (2020).Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. Proceedings of the National Academy of Sciences of the United States of America , 117 (23), 12592–12594. https://doi.org/10.1073/pnas.1919012117 Manasi, A., Panchanadeswaran, S., Sours, E., & Lee, S. J. (2022). Mirroring the bias: Gender and artificial intelligence. Gender, Technology and Development , 26 , 295–305. https://doi.org/10.1080/09718524.2022.2128254 Marcus, E., & Teuwen, J. (2024). Artificial intelligence and explanation: How, why, and when to explain black boxes. European Journal of Radiology , 173 , Article 111393. https://doi.org/10.1016/j.ejrad.2024.111393 Mehmood, F., Ghani, M. U., Ibrahim, M. A., Shahzadi, R., Mahmood, W., & and Asim, M. N.(2020). A Precisely Xtreme-Multi Channel Hybrid Approach for Roman Urdu Sentiment Analysis. IEEE Access , 8 , 192740–192759, https://doi.org/10.1109/ACCESS.2020.3030885. Mosteiro, P., Kuiper, J., Masthoff, J., Scheepers, F., & Spruit, M. (2022). Bias Discovery in Machine Learning Models for Mental Health. Information , 13 (5), Article 237. https://doi.org/10.3390/info13050237 Nadeem, A., Marjanovic, O., & Abedin, B. (2022). Gender bias in AI-based decision-making systems: A systematic literature review. Australas. J. Inf. Syst., 26 . https://doi.org/10.3127/ajis.v26i0.3835 Nah, S., Luo, J., Kim, S., Chen, M., Mitson, R., & Joo, J. (2024). Algorithmic Bias or Algorithmic Reconstruction? A Comparative Analysis Between AI News and Human News. International Journal of Communication , 18 , 700–729. Newstead, T. P., Eager, B., & Wilson, S. (2023). How AI can perpetuate – Or help mitigate –Gender bias in leadership. Organizational Dynamics , 52 (4), Article 100998. https://doi.org/10.1016/j.orgdyn.2023.100998 Noble, S. U. (2018). Algorithms of Oppression : How Search Engines Reinforce Racism . New York: NYU Press. https://doi.org/10.2307/j.ctt1pwt9w5. Noguero, D. S., Ramírez-Cifuentes, D., Ríssola, E. A., & Freire, A. (2023). Gender Bias When Using Artificial Intelligence to Assess Anorexia Nervosa on Social Media: Data-Driven Study. Journal of medical Internet Research , 25 , e45184. https://doi.org/10.2196/45184 O’Connor, S., & Liu, H. (2023). Gender bias perpetuation and mitigation in AI technologies: Challenges and opportunities. AI & Soc iety , 1–13. https://doi.org/10.1007/s00146-023-01675-4 Oca, M. C., Meller, L., Wilson, K., Parikh, A. O., McCoy, A., Chang, J., Sudharshan, R., Gupta, S., & Zhang-Nunes, S. (2023). Bias and Inaccuracy in AI Chatbot Ophthalmologist Recommendations. Cureus , 15 . https://doi.org/10.7759/cureus.45911 Page, M. J. et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. Systematic Reviews , 2021 (10), 89–100. https://doi.org/10.1186/s13643-021-01626-4 Park, J., Arunachalam, R., Silenzio, V., & Singh, V. K. (2022). Fairness in Mobile Phone-Based Mental Health Assessment Algorithms: Exploratory Study. JMIR formative research , 6 (6), e34366. https://doi.org/10.2196/34366 Paul, S., Maindarkar, M. A., Saxena, S., Saba, L., Turk, M., Kalra, M. S., Krishnan, P. R., & Suri, J. S. (2022). Bias Investigation in Artificial Intelligence Systems for Early Detection of Parkinson’s Disease: A Narrative Review. Diagnostics , 12 (1), Article 166. https://doi.org/10.3390/diagnostics12010166 Pisanelli, E. (2022). Your resume is your gatekeeper: Automated resume screening as a strategy to reduce gender gaps in hiring. Economics Letters , 221 , Article 110892. https://doi.org/10.1016/j.econlet.2022.110892 Prates, M. O. R., Avelar, P. H., & Lamb, L. C. (2020). Assessing gender bias in machine translation: A case study with Google Translate. Neural Comput & Applic ., 32 , 6363–6381. https://doi.org/10.1007/s00521-019-04144-6 Ratajczak, R., & Cockerill, R. G. (2023). Artificial Intelligence in Violence Risk Assessment: Addressing Racial Bias and Inequity. Journal of Psychiatric Practice , 29 (3), 239–245. https://doi.org/10.1097/PRA.0000000000000713 Rigby, M. J. (2019). Ethical Dimensions of Using Artificial Intelligence in Health Care. AMA Journal of Ethics , 21 (2), 121–124. https://doi.org/10.1001/AMAJETHICS.2019.121 Rizhinashvili, D., Sham, A. H., & Anbarjafari, G. (2022). Gender Neutralisation for Unbiased Speech Synthesising. Electronics , 11 (10), Article1594. https://doi.org/10.3390/electronics11101594 Selbst, A. D., Boyd, D., Friedler, S. A., Venkatasubramanian, S., & Vertesi, J. (2019). Fairness and Abstraction in Sociotechnical Systems. Proceedings of the Conference on Fairness, Accountability, and Transparency, 59–68. https://doi.org/10.1145/3287560.3287598. Shrestha, S., & Das, S. (2022). Exploring gender biases in ML and AI academic research through systematic literature review. Frontiers in artificial intelligence , 5 , Article 976838. https://doi.org/10.3389/frai.2022.976838 Solans N, D., Ramírez-Cifuentes, D., Ríssola, E. A., & Freire, A. (2023). Gender Bias When Using Artificial Intelligence to Assess Anorexia Nervosa on Social Media: Data-Driven Study. Journal of Medical Internet Research , 25 , e45184. https://doi.org/10.2196/45184 Suen, H., & Hung, K. (2023). Building trust in automatic video interviews using various AI interfaces: Tangibility, immediacy, and transparency. Comput. Hum. Behav., 143 , Article 107713. https://doi.org/10.1016/j.chb.2023.107713 Sun, L. H., Wei, M., Sun, Y. B., Suh, Y. J., Shen, L. W., & Yang. S. J. (2023). Smiling Women Pitching Down: Auditing Representational and Presentational Gender Biases in Image Generative AI. Journal of Computer-Mediated Communication , 29 (1), zmad045. https://doi.org/10.1093/jcmc/zmad045 Tadimalla, S. Y., & Maher, M. L. (2024). Implications of Identity in AI: Creators, Creations, and Consequences. Proceedings of the AAAI Symposium Series , 3 (1), 528–535. https://doi.org/10.1609/aaaiss.v3i1.31268 Tomalin, M., Byrne, B., Concannon, S., Saunders, D., & Ullmann, S. (2021). The practical ethics of bias reduction in machine translation: Why domain adaptation is better than data debiasing. Ethics Inf. Technol. , 23 , 419–433. https://doi.org/10.1007/s10676-021-09583-1 Tubadji, A., Huang, H., & Webber, D. J. (2021). Cultural proximity bias in AI acceptability: The importance of being human. Technol. Forecast. Soc. Change , 173 , Article 121100. https://doi.org/10.1016/j.techfore.2021.121100 UNESCO. (2020). Artificial intelligence and gender equality: Key findings of UNESCO’s Global Dialogue.`https://unesdoc.unesco.org/ark:/48223/pf0000374174. Accessed 20 May 2024 Vallor, S. (2024). The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking . New York: Oxford University Press. https://doi.org/10.1093/oso/9780197759066.001.0001 Wæraas, A. (2022). Thematic Analysis: Making Values Emerge from Texts. In G. Espedal, B. Jelstad Løvaas, S. Sirris, A. Wæraas (Eds.), Researching Values (pp. 153–170). Palgrave Macmillan, Cham. https://doi.org/10.1007/978-3-030-90769-3_9 Wang, C., Wang, K., Bian, A., Islam, R., Keya, K., Foulds, J. R., & Pan, S. (2023). When Biased Humans Meet Debiased AI: A Case Study in College Major Recommendation. ACM Transactions on Interactive Intelligent Systems , 13 , 1–28. https://doi.org/10.1145/3611313 Wang, L. (2020). The Three Harms of Gendered Technology. Australasian Journal of Information Systems , 24 . https://doi.org/10.3127/ajis.v24i0.2799 Wang, Y. L., Wang, Q., Zhao, L. C., & Wang, C. (2023) Differential privacy in deep learning: Privacy and beyond. Future Generation Computer Systems , 148 , 408–424. https://doi.org/10.1016/j.future.2023.06.010. Yang, J. J., & Zhou, C. (2023). Algorithmic black-box problem in medical artificial intelligence: Ethical challenges and solution approach. Chinese Science Bulletin , 68 (13), 1604–1610. https://doi.org/10.1360/TB-2022-1320 Yang, Q., Huang, A. B., Fan, L. X., Chan, C. S., Lim, J. H., Ng, K. W., Ong, D. S., & Li, B. (2023). Federated Learning with Privacy-preserving and Model IP-right-protection. Mach. Intell. Res., 20 , 19–37. https://doi.org/10.1007/s11633-022-1343-2 Yoder-Himes, D. R., Asif, A., Kinney, K., Brandt, T. J., Cecil, R., Himes, P. R., Cashon, C. H., Hopp, R. M., & Ross, E. (2022). Racial, skin tone, and sex disparities in automated proctoring software. Frontiers in Education , 7 , 1–16. https://doi.org/10.3389/feduc.2022.881449 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 06 May, 2026 Reviews received at journal 20 Apr, 2026 Reviewers agreed at journal 20 Apr, 2026 Reviews received at journal 18 Apr, 2026 Reviews received at journal 15 Apr, 2026 Reviewers agreed at journal 15 Apr, 2026 Reviews received at journal 15 Apr, 2026 Reviewers agreed at journal 15 Apr, 2026 Reviewers agreed at journal 15 Apr, 2026 Reviewers invited by journal 14 Apr, 2026 Editor assigned by journal 06 Apr, 2026 Editor invited by journal 02 Apr, 2026 Submission checks completed at journal 02 Apr, 2026 First submitted to journal 02 Apr, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9240627","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Systematic Review","associatedPublications":[],"authors":[{"id":626617469,"identity":"9bb18d35-ff63-482e-a6af-c1014434114b","order_by":0,"name":"Jingyue Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAz0lEQVRIiWNgGAWjYBACNv7+Dwc+VNjYyTMzHyBOC5/EAcOHM86kJRu2tyUQp0WOIcHYmLPtMGPDmTMGRDqM4UCaNAMbMzPjjJyPN94w2MnpNhDSwtxwTLqAh42PXSJ3s+UchmRjswMEbTnYJj1DggdoS+42aR6GA4nbCGtJZpPmMZBgbLiR84xYLWnMxjwJBiDvsxGpReIM48MZBxJAgWxsOceACL/I9/cwHPj47z8oKh/eeFNhJ0dQCwqQ4CEyapC1kKpjFIyCUTAKRgQAAH8mQCzYDvKDAAAAAElFTkSuQmCC","orcid":"","institution":"Beijing Language and Culture University","correspondingAuthor":true,"prefix":"","firstName":"Jingyue","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2026-03-27 06:08:32","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9240627/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9240627/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":107617364,"identity":"efa0a1a8-46f9-4b3f-8eef-ec5b70169859","added_by":"auto","created_at":"2026-04-23 09:19:39","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":521884,"visible":true,"origin":"","legend":"\u003cp\u003eThe screening process based on PRISMA\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/c44c2ec46493d40713e7174c.jpeg"},{"id":107707374,"identity":"55db740b-2622-48e1-91cf-adc94e03828a","added_by":"auto","created_at":"2026-04-24 09:20:11","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":150870,"visible":true,"origin":"","legend":"\u003cp\u003eNumber of publications by year\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/7c9718365ae6434781e457f8.jpeg"},{"id":107617369,"identity":"af535a11-ae57-424c-a358-907f35663e5a","added_by":"auto","created_at":"2026-04-23 09:19:39","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":25104,"visible":true,"origin":"","legend":"\u003cp\u003eThe distribution of research methodology\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/09b1a7f40536a2203b4daab4.jpg"},{"id":107617366,"identity":"ff5eda1a-bc4e-4495-bc7e-cb771d6422f9","added_by":"auto","created_at":"2026-04-23 09:19:39","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":265532,"visible":true,"origin":"","legend":"\u003cp\u003eDomains that reflect AI’s gender bias\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/b4489d937c11c8c13c9cc4d5.jpeg"},{"id":107707341,"identity":"959e99d3-0fb5-4b63-a517-b122ccee21ac","added_by":"auto","created_at":"2026-04-24 09:20:06","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":297537,"visible":true,"origin":"","legend":"\u003cp\u003eThe relationship of causes of gender bias in AI\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/77c883d1b18e44aa9aa53d9a.jpeg"},{"id":107617368,"identity":"0b1d0f8d-bcff-43f3-a8b8-23e7178ddb29","added_by":"auto","created_at":"2026-04-23 09:19:39","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":151407,"visible":true,"origin":"","legend":"\u003cp\u003eMitigation approaches of AI’s gender bias\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/19f6948f636117c96afc0e6d.jpeg"},{"id":107709199,"identity":"e42de4af-bc93-476e-8d9b-f12db280ca62","added_by":"auto","created_at":"2026-04-24 09:34:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2044934,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9240627/v1/3836115b-0cd0-44c1-b6e6-776c5c35677c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A systematic review and thematic analysis of gender bias in AI","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eConsidered a powerful technology, artificial intelligence (AI) has fundamentally changed technical environments (Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and is regarded as the center of the Fourth Industrial Revolution (Khan \u0026amp; Ewuoso, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). With its widespread and increasing application in hiring (Gagandeep et al., 2024), medicine (Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and interviews (Suen \u0026amp; Hung, \u003cspan citationid=\"CR83\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), etc., problems of bias in AI become visible, mainly including racial bias, gender bias, religion bias and cultural bias (Tubadji et al., \u003cspan citationid=\"CR87\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Among them, gender bias garners large attention for its profound impact on individuals and society (Baker \u0026amp; Hawn, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Galos \u0026amp; Coppock, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Gender bias in AI does not only impact individuals\u0026rsquo; rights and opportunities but also can be amplified by the AI system itself. Thus, addressing gender bias should be solved as the priority and seen as the task in the research and policy agenda (Hall \u0026amp; Ellis, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), which could promote gender equality, social justice and healthy development of AI technology.\u003c/p\u003e \u003cp\u003eGender bias is the tendency to prefer one gender over another, deriving from individuals\u0026rsquo; unconsciousness and automatic emotions with skipping rational thinking (Coiro \u0026amp; Pollak, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Gender bias in AI refers to the unfair or discriminatory treatment towards individuals based on gender, often deriving from imbalanced data, algorithmic prejudices and human biases embedded in AI systems. This bias not only undermines the fairness and trustworthiness of AI but also impedes its widespread adoption. The lack of clear regulations and the opaque nature of AI systems exacerbate these issues (Noble, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2018\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHistorically, the issue of gender bias in AI is often overlooked, with early research focusing primarily on technical performance rather than ethical implications (Alami et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Recent studies have highlighted the multifaceted harms of gender bias in AI. For instance, Ellis and Hall (\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) noted that AI systems often amplify existing societal biases, particularly in image search results where engines like Bing and Google display imbalanced gender representations (G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the hiring process, gender bias is particularly obvious (Chang, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Gagandeep et al. (2024) revealed that Amazon\u0026rsquo;s AI-based resume filtering system unintentionally favored male candidates, thereby perpetuating gender prejudice.\u003c/p\u003e \u003cp\u003eThe emergence of generative AI models, such as ChatGPT and LLaMA, has introduced new approaches but also challenges in addressing gender bias. These models while transformative, have been shown to perpetuate and even amplify existing biases. For instance, some studies reveal that ChatGPT often generates text that reinforces traditional gender roles, such as associating independent and adventurous traits with men and inclusive and emotional roles with women (Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Similarly, LLaMA\u0026rsquo;s open-source nature has raised concerns about its possible abuse in applications that may deepen gender inequalities, such as biased content generation in hiring tools or educational materials. Fang et al.\u0026rsquo;s (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) study claimed that the representative large language models (LLMs), like ChatGPT and LLaMA, demonstrated significant bias against women and Black people.\u003c/p\u003e \u003cp\u003eTo address these challenges, researchers proposed various mitigation techniques. One promising technique is prompt engineering, where carefully designed inputs are used to guide generative models toward more neutral and unbiased outputs (Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For example, a prompt like \u0026ldquo;Describe a nurse without referencing gender\u0026rdquo; can help reduce stereotypical associations. Besides, the development of more diverse and representative datasets has been emphasized as a critical step in tackling biases. In addition to technological efforts, the existing research also emphasizes individuals\u0026rsquo; greater awareness of gender bias and the necessity of legal regulations (Bernagozzi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Like in the healthcare field, the American Medical Association passes a policy, calling for the development of thoughtfully designed, high-quality and clinically proven AI technologies (Rigby, \u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite these efforts, there is a scarcity of review articles analyzing gender bias in AI technology. 7 review articles about bias in AI were identified, including systematic review (N\u0026thinsp;=\u0026thinsp;3), scoping review (N\u0026thinsp;=\u0026thinsp;3) and narrative review (N\u0026thinsp;=\u0026thinsp;1). Among these, only two reviews specifically focus on gender bias. 3 results just discuss the bias generally (Chen et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Delgado et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Paul et al., \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and others explore bias related to ethics (Daneshjou, 2021), race, or age (Chu et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The reviews of gender bias are all performed through the systematic review, which utilizes a more strict methodology to thoroughly examine the literature in a clearly defined manner (Gregory \u0026amp; Denniss, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Furthermore, although a few papers touch on the causes of bias (Daneshjou, 2021; Delgado et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Hall \u0026amp; Ellis, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), or propose mitigation methods (Chen et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Hall \u0026amp; Ellis, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Shrestha \u0026amp; Das, \u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), none of them comprehensively address the causes, mitigation methods, consequences and challenges of mitigating gender bias simultaneously (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe detailed information of review articles of bias in AI\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eYear\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTitle\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eType\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFocus\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDaneshjou\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2021\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLack of Transparency and Potential Bias in Artificial Intelligence Data Sets and Algorithms: A Scoping Review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eScoping review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eData bias in Clinical AI algorithms; sources of data bias; ethnic or race bias to patients through the recognition of skin color\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePaul et al.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBias Investigation in Artificial Intelligence Systems for Early Detection of Parkinson\u0026rsquo;s Disease: A Narrative Review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNarrative review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBias in the health-related AI model; computing the risk of bias\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDelgado et al.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBias in algorithms of AI systems developed for COVID\u0026minus;19: A scoping review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eScoping review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCauses and consequences of bias from the ethical perspective\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eShrestha \u0026amp; Das\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2022\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eExploring gender biases in ML and AI academic research through systematic literature review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGender bias in the machine learning and AI assistant automated systems; mitigation and detection methods\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHall \u0026amp; Ellis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eA systematic review of socio-technical gender bias in AI algorithms\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSolutions, causes and consequences of gender bias in AI algorithms from the socio-technical framework\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChu et al.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2023\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAge-related bias and artificial intelligence: A scoping review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eScoping review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIdentifying how AI systems encode, produce, or reinforce age-related bias; main domains that present the age bias of AI (age recognition and facial recognition systems)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen et al.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUnmasking bias in artificial intelligence: A systematic review of bias detection and mitigation strategies in electronic health record-based models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBias detection and mitigation strategies in the health-related model; six bias types (algorithmic, confounding, implicit, measurement, selection and temporal)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThis study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2024\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDoes gender bias in AI exist: A systematic review and thematic analysis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSystematic review\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGender bias in AI; research trend; causes, solutions, consequences and challenges of AI\u0026rsquo;s gender bias\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo bridge these gaps, the present study adopts a systematic review approach, providing a comprehensive analysis of gender bias in AI systems. Unlike the above reviews that center on algorithmic or general bias, this study takes a broader perspective, examining gender bias from the macroscopic areas that manifest it to the specific application systems of AI, then to the model algorithms behind them. By reviewing the underlying causes, consequences, challenges and mitigation strategies of gender bias, this study aims to offer a more complete understanding of the issue. Findings in this review also attempt to provide insightful views into the root causes of gender bias, enabling the design of fairer and more inclusive AI systems for AI developers. Besides, the review summarized the societal consequences of gender bias in AI, offering policymakers a foundation for making targeted policies to mitigate these effects. To achieve these goals, the following research questions were proposed from the publication aspect, research design aspect and research focus aspect.\u003c/p\u003e \u003cp\u003eRQ1: What is the current trend of AI\u0026rsquo;s gender bias, including year publication and journal of publication?\u003c/p\u003e \u003cp\u003eRQ2: What is the distribution of research methodologies in the study of AI\u0026rsquo;s gender bias?\u003c/p\u003e \u003cp\u003eRQ3: What are the domains that present gender bias in AI\u003c/p\u003e \u003cp\u003eRQ4: What causes of AI\u0026rsquo;s gender bias are proposed?\u003c/p\u003e \u003cp\u003eRQ5: What mitigation approaches for AI\u0026rsquo;s gender bias are proposed?\u003c/p\u003e \u003cp\u003eRQ6: What are the challenges of mitigating AI\u0026rsquo;s gender bias?\u003c/p\u003e \u003cp\u003eRQ7: What are the consequences of AI\u0026rsquo;s gender bias?\u003c/p\u003e"},{"header":"2 Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Data collection\u003c/h2\u003e \u003cp\u003eOn 3 May, 2024, we selected five databases to retrieve articles, including the Web of Science (WOS), Elsevier, Taylor \u0026amp; Francis, Wiley and Sage. To ascertain the search terms, we first determined the basic keywords AI and gender bias to gain the relevant research articles or reviews (e.g. Delgado et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Shrestha \u0026amp; Das, \u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and then summarized the other expressions having the same meaning. Finally, we formed the two search strings in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The logic connection among the terms in the same category was OR and the AND was used to link these two categories.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eKeyword search scheme\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTechnology\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;AI\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"5\" rowspan=\"6\"\u003e \u003cp\u003eAND\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Gender bias\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;Artificial intelligence\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Sexism\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;AI agent\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Feminism\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;Artificial intelligence agent\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Gender inequality\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;Chatbot\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Gender prejudice*\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ldquo;Chatterbot\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u0026ldquo;Gender stereotype*\u0026rdquo;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the WOS, the Core Collection was chosen, including six indexes: Science Citation Index Expanded (SCI-EXPANDED, 2013 to present), Social Sciences Citation Index (SSCI, 2008 to present), Arts \u0026amp; Humanities Citation Index (A\u0026amp;HCI, 2008 to present), Emerging Sources Citation Index (ESCI, 2018 to present), Current Chemical Reactions (CCR-EXPANDED, 1985 to present) and Index Chemicus (IC, 1993 to present). 164 results were gained for \u0026ldquo;AI\u0026rdquo; OR \u0026ldquo;Artificial intelligence\u0026rdquo; OR \u0026ldquo;AI agent\u0026rdquo; OR \u0026ldquo;Artificial intelligence agent\u0026rdquo; OR \u0026ldquo;Chatbot\u0026rdquo; OR \u0026ldquo;Chatterbot\u0026rdquo; AND \u0026ldquo;Gender bias\u0026rdquo; OR \u0026ldquo;Sexism\u0026rdquo; OR \u0026ldquo;Feminism\u0026rdquo; OR \u0026ldquo;Gender inequality\u0026rdquo; OR \u0026ldquo;Gender prejudice*\u0026rdquo; OR \u0026ldquo;Gender stereotype*\u0026rdquo;. We selected the document type of articles (N\u0026thinsp;=\u0026thinsp;148). In Wiley and Taylor \u0026amp; Francis, through setting these keywords in the abstract, 3 and 11 articles remained respectively. In Elsevier Science Direct, we set title, abstract, keywords: \u0026ldquo;AI\u0026rdquo; OR \u0026ldquo;Artificial intelligence\u0026rdquo; OR \u0026ldquo;AI agent\u0026rdquo; OR \u0026ldquo;Artificial intelligence agent\u0026rdquo; OR \u0026ldquo;Chatbot\u0026rdquo; OR \u0026ldquo;Chatterbot\u0026rdquo; AND Title: \u0026ldquo;Gender bias\u0026rdquo; OR \u0026ldquo;Sexism\u0026rdquo; OR \u0026ldquo;Feminism\u0026rdquo; OR \u0026ldquo;Gender inequality\u0026rdquo; OR \u0026ldquo;Gender prejudice\u0026rdquo; OR \u0026ldquo;Gender stereotype\u0026rdquo;, there were 6 research articles. The symbol \u0026ldquo;*\u0026rdquo; was not allowed in Elsevier Science Direct different from the other four databases. In Sage, by keying these keywords in the abstract, 2 articles remained. In total, 170 articles were imported into EndNote X9 for further screening (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Inclusion and exclusion criteria\u003c/h2\u003e \u003cp\u003eThe selection of papers was rigorously based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Page et al., \u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). We first skimmed through titles and abstracts to filter the articles that 1) were not written in English and 2) were duplicates. Some irrelevant articles could not be excluded by the retrieval terms because of the different combinations and logic of these keywords. Thus, we proceeded to a more detailed screening, where we excluded articles that 1) focused on the influence of gendered AI (e.g. Brown, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e); 2) centered on the influence of users\u0026rsquo; gender on the perception of AI; or 3) emphasized the identification and avoidance of gender bias by AI (e.g. Guevara-G\u0026oacute;mez et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Koo, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Pisanelli, \u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In this stage, we gained 36 articles. Then, we carefully looked through these 36 full papers to ensure they met our inclusion criteria, which required studies to 1) employ clear and rigorous methodologies to investigate gender bias in AI; 2) explicitly identify gender bias in AI systems; 3) discuss the societal, ethical, or technical impacts of gender bias; 4) explore the underlying causes or mechanisms of gender bias or; 5) propose or evaluate strategies to mitigate gender bias. During this stage, 7 results were removed because of inadequate discussion on gender bias or lack of clear methodology. Ultimately, 29 articles were used for data analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Quality assessment\u003c/h2\u003e \u003cp\u003eThese 29 results were evaluated under the guidelines established by the American Educational Research Association (AERA) to guarantee the quality (Dur\u0026aacute;n et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). The guideline has eight concrete standards, including problem formulation, design and logic, sources of evidence, measurement and classification, analysis and interpretation, generalization, ethics in reporting and title, abstract, and headings. Each of them was scored from 1 point to 5 points by its validity. Thus, the total scores varied from 8 to 40 points. Two authors marked them independently and only included the articles that scored at least 20 points referring to the scoring system in Khosravi et al.\u0026rsquo;s (\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) systematic review article. Two researchers evaluated the articles with high inter-rater reliability (k\u0026thinsp;=\u0026thinsp;0.8) and all results were eligible.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4 Data extraction and thematic analysis\u003c/h2\u003e \u003cp\u003eThe data extraction was conducted through a rigorous application of thematic analysis, following the guidelines proposed by Braun and Clarke (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). Thematic analysis is a qualitative method of recognizing and forming themes within data and we positioned our thematic analysis as primarily descriptive with interpretive elements, concentrating on summarizing and categorizing patterns related to gender bias in AI and also exploring the underlying meanings of these patterns (Braun \u0026amp; Clarke, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This study employed both inductive and deductive approaches to guarantee a comprehensive understanding of the data. The data extraction process began with a thorough review of all selected articles. Each article was read several times to identify relevant text segments related to gender bias in AI. These segments were extracted based on their connection to research questions.The inductive approach suggests that the theme or category in research is totally data-driven (W\u0026aelig;raas, \u003cspan citationid=\"CR90\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), while the deductive approach is based on the preexisting theoretical frameworks. (Fife \u0026amp; Gossner, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). During the inductive phase, the author conducted open coding, where text segments were labeled with codes. These codes were then grouped into preliminary categories based on their similarities. The deductive process began with five categories derived from the literature (e.g. Hall \u0026amp; Ellis, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Jeon et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), including basic information (e.g. year, publication journal), methodology (e.g. research design), domains of gender bias of AI (e.g. decision-making system, AI-generated contents), causes of gender bias of AI (e.g. biased data), mitigation approaches of gender bias (e.g. debiasing techniques) and consequences of gender bias of AI (e.g. social inequality). However, the five categories could not cover the finalized codes, so the categories of challenges of mitigating gender bias were introduced, forming the final code book. This iterative process of refining the coding framework is consistent with the principles of the thematic analysis, which emphasize flexibility and responsiveness to the data (Braun \u0026amp; Clarke, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2006\u003c/span\u003e). To ensure the reliability of the coding process, the second author independently coded using the same code book. If two authors coded differently, they would discuss, finally forming three themes, including publication aspect, research design and research focus aspect (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The inter-coder reliability was assessed by comparing the consistency of codes assigned to the same study by two or more coders (Kurasaki, \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2000\u003c/span\u003e). A high percentage agreement of 98% was observed, indicating strong consistency in the coding process (Belur et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The results of the coding schema were shown in detail in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eCoding schema\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTheme\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCategory\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDefinition\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSub-category\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eIncluded studies\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cb\u003ePublication aspect\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cem\u003eBasic information\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eInformation about the publication\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eYear of publication\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAll 29 results\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eJournal of publication\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cb\u003eResearch design aspect\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cem\u003eMethodology\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eMethod of systematically designing a study\u0026nbsp;to ensure reliable results\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQuantitative study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gupta et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Hort et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Larrazabal et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Noguero et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Rizhinashvili et al., \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eQualitative study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Howard \u0026amp; Borenstein, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Newstead et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMixed study\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBernagozzi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Gagandeep et al., 2024; Nah et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Oca et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Prates et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"23\" rowspan=\"24\"\u003e \u003cp\u003e\u003cb\u003eResearch focus aspect\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"9\" rowspan=\"10\"\u003e \u003cp\u003e\u003cem\u003eDomains of gender bias of AI\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"9\" rowspan=\"10\"\u003e \u003cp\u003eThe specific situations or general AI technologies reflecting AI\u0026rsquo;s gender bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHealth and medicine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBusl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Larrazabal et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTranslation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBernagozzi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Prates et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAI assistant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAI-generated content\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Nah et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Newstead et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecommender system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Gupta et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Oca et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDecision-making system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGagandeep et al., 2024; Hort et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAssessment system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNoguero et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecognition system\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHoward \u0026amp; Borenstein, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Rizhinashvili et al., \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLarge Language Model\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNLP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eO\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003e\u003cem\u003eCauses of gender bias of AI\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eReasons behind AI\u0026rsquo;s gender bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBusl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Larrazabal et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Prates et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHuman bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Nah et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Newstead et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlgorithmic bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSocial bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Oca et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003e\u003cem\u003eMitigation approaches of AI\u0026rsquo;s gender bias\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003eStrategies for reducing gender bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eData level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBusl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Cecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gagandeep et al., 2024; Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Rizhinashvili et al., \u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlgorithmic and model level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Noguero et al., \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEthical and social level\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Prates et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAssessment and diagnosis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Bernagozzi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Howard \u0026amp; Borenstein, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Larrazabal et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Newstead et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cem\u003eChallenges of mitigating AI\u0026rsquo;s gender bias\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eObstacles in reducing AI\u0026rsquo;s gender bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePerformance of models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBernagozzi et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Hort et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCulture and value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Howard \u0026amp; Borenstein, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEthical concerns\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u003cem\u003eConsequences of gender bias of AI\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eNegative effects of AI\u0026rsquo;s gender bias\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eInequalities amplification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eBusl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Oca et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGender stereotypes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAdams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eUnequal treatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePublication quantity and topic by journal\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJournal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTopic\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e1\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eACM Transactions on Interactive Intelligent Systems; Applied Soft Computing; Information Systems Frontiers; IEEE Internet Computing; Neural Computing and Applications; Australasian Journal of Information Systems; Journal of Computer-Mediated Communication; Electronics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eComputer science and information technology\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e2\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAI\u0026amp;Society; Gender, Technology and Development; Rupkatha Journal on Interdisciplinary Studies in Humanities; Feminist Media Studies; Organizational Dynamics\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSocial sciences and humanities\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e3\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSocial Sciences; Scientific Reports; Science And Engineering Ethics; Proceedings of the National Academy of Sciences of the United States of America; Technological Forecasting \u0026amp; Social Change\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMultidisciplinary sciences\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCureus; Frontiers in Global Women\u0026rsquo;s Health; Frontiers in Physiology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBiomedicine and health\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eJournal of Medical Internet Research; Manufacturing \u0026amp; Service Operations Management\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEngineering and management\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e6\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCambridge International Law Journal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLaw\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e7\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInternational Journal of Communication; JMIR Formative Research; Multimedia Tools and Applications\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMedia and Communication\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e8\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFrontiers in Education\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEducation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"3 Results","content":" \u003cp\u003eAfter analyzing 29 articles, the section reported the results from the publication aspect, research design and research focus aspect in the field of AI\u0026rsquo; s gender bias.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Publication aspect\u003c/h2\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003e3.1.1 Publication years\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e illustrated the annual publication distribution, revealing a progressive increase in scholarly attention to gender bias in AI systems. After the first article was published in 2017, no article was documented in 2018. The number of articles remained relatively low from 2019 to 2021 but appeared to have nearly doubled growth in 2022 and 2023. The latest publications counted on May 3, 2024 were 4, which surpassed the total annual publications before 2022, indicating a sustained growth trend.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003e3.1.2 Journal of publication\u003c/h2\u003e \u003cp\u003eThe 29 results were mostly from different sources and only two of them were in the same journal called the Journal of Medical Internet Research. According to the main aims of these journals, the publications on gender bias of AI were categorized into 8 general topics, including computer science and information technology, social sciences and humanities, multidisciplinary sciences, biomedicine and health, engineering and management, law, media and communication and education.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Research design aspect\u003c/h2\u003e \u003cp\u003eRegarding research methodology (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), 8 studies were qualitative and the majority of them utilized case studies to describe and explain how AI contributes to gender prejudice in particular fields (e.g. O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). 12 results adopted a quantitative research design, most of which were about testing the proficiency and accuracy of models through concrete statistical parameters to mitigate AI\u0026rsquo;s gender bias (e.g. Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The mixed method was used in 9 articles that quantitatively analyzed the output of AI-generated content to show how gender bias was manifested and qualitatively interpreted the results (e.g. Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Prates, 2020; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Research focus aspect\u003c/h2\u003e \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e \u003ch2\u003e3.3.1 Domains of gender bias in AI\u003c/h2\u003e \u003cp\u003eThe 29 results involved 10 domains that were classified in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. AI is a technology that simulates human intelligence and Natural Language Processing (NLP) is a core subfield of AI that studies how to enable computers to understand and process human language. LLM, on the other hand, is a crucial technology in NLP and the classification of the application is the the actual and specific application direction of NLP/LLM. For example, if an article is about the technical principles of NLP, it will be classified as NLP, and if it is a study of the application of NLP in medical diagnosis, it will be classified as an application layer. In terms of these 10 branches, it could be concluded that the discussion of gender bias of AI in the selected articles was conducted from the perspectives of both macroscopic technologies and microscopic applications.\u003c/p\u003e \u003cp\u003eThe most common research was on the gender bias of AI-generated text/image (N\u0026thinsp;=\u0026thinsp;6), followed by the recommender system (N\u0026thinsp;=\u0026thinsp;4), AI assistant (N\u0026thinsp;=\u0026thinsp;4), health and medicine (N\u0026thinsp;=\u0026thinsp;3), decision-making system (N\u0026thinsp;=\u0026thinsp;3), recognition system (N\u0026thinsp;=\u0026thinsp;3), assessment system (N\u0026thinsp;=\u0026thinsp;2), translation (N\u0026thinsp;=\u0026thinsp;3), LLM (N\u0026thinsp;=\u0026thinsp;1) and NLP (N\u0026thinsp;=\u0026thinsp;1). AI-generated text/image was discussed in 6 papers. Fang et al. (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) and Nah et al. (\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) analyzed the gender bias in AI-generated news by providing broad topics or existing titles in previous news. The former results showed a higher gender bias in AI-generated news, but the latter found that the bias in AI-generated news was not less than in human-made ones. For AI-generated images, there was an imbalance between AI-generated males and females in some professions, with males being disproportionately high (doctor, lawyer, engineer, scientist), indicating a bias toward females (e.g. G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFor different application systems, 3 studies focused on recommender systems that addressed specific areas of recommendation, including healthcare, career guidance, and advertising (Cecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Oca et al., \u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The decision-making system and assessment system were primarily applied in the hiring and health fields respectively (e.g. Gagandeep et al., 2024; Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In terms of the recognition system, Howard and Borenstein (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) verified the low accuracy of the voice recognition system for female voices. The same year saw the identification of gender bias within facial recognition systems (Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRegarding other more concrete domains, voice-based and text-based assistants were explored, such as Apple\u0026rsquo;s Siri, Google, ChatGPT and Microsoft\u0026rsquo;s Cortana (Adams \u0026amp; Loide\u0026aacute;in, 2019; Manasi et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang, \u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). The default use of female voices in most voice assistants reinforced gender stereotypes, positioning women as submissive or service-oriented roles. Notably, Google was the only major platform that avoided this gendered default (Adams \u0026amp; Loideain, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). ChatGPT, mainly based on texts, attributed different personalities to different genders and did not provide any traits for gender-diverse people (Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In health and medicine, studies have paid attention to the negative effect of AI on the gender ratio of employees in this occupation and declared the relatively low performance of models when trained with data from female patients (Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Larrazabal et al., \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). In terms of translation, most studies centered on the Google translator (e.g. O\u0026rsquo;Connor \u0026amp; Liu, 2020; Prates et al., \u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and tested its gender bias usually through the sentence form: He/She is +\u0026thinsp;occupation. The results suggested that the translator did not present the real distribution of gender in some jobs, especially in STEM.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003ch2\u003e3.3.2 Causes of gender bias of AI\u003c/h2\u003e \u003cp\u003eA total of 25 articles were analyzed to identify the primary factors contributing to gender bias in AI. The findings were categorized into four distinct causes: data bias (N\u0026thinsp;=\u0026thinsp;9), human bias (N\u0026thinsp;=\u0026thinsp;6), algorithmic bias (N\u0026thinsp;=\u0026thinsp;5) and social bias (N\u0026thinsp;=\u0026thinsp;5).\u003c/p\u003e \u003cp\u003eData bias was the most influential factor, which suggested that gender bias would occur when the model was trained without gender-balanced data (Chung et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Dwivedi et al. (\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) claimed that gender bias was not generated from air but from data. For instance, men\u0026rsquo;s faces were recognized more easily than women\u0026rsquo;s faces possibly because more male images were collected in the training set (Yoder-Himes et al., \u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Additionally, data leakage, a phenomenon where information from the test set inadvertently influences the training process, can introduce gender bias. O\u0026rsquo;Connor and Liu (\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) gave examples of presenting a dataset featuring an equal representation of women and men engaged in cooking activities. While this balance did not inherently introduce bias, the presence of a child in the image may lead to bias. Given that children were frequently depicted alongside women rather than men in various images, the model might link \u0026ldquo;children\u0026rdquo; with \u0026ldquo;cooking.\u0026rdquo; Consequently, this could result in women being disproportionately categorized as \u0026ldquo;cooking\u0026rdquo; compared to men.\u003c/p\u003e \u003cp\u003eThe essence of data bias was human bias (Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), which was in line with Nah et al.\u0026rsquo;s (\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), Manasi et al.\u0026rsquo;s (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) and Newstead et al.\u0026rsquo;s (\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) studies, indicating that designers and creators could choose preferred trained data due to gender stereotypes or commercial interests, along with their own biases. In the present research, human bias and social bias were similar but coded differently based on their distinct origins and manifestations. Human bias mainly referred to the prejudices and stereotypes held by individuals in the design and development of AI systems, such as developers, data scientists and engineers. Nah et al.\u0026rsquo;s (\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) study claimed that developers may introduce bias by prioritizing certain keywords or perspectives that met their own beliefs, which resulted in AI systems producing news that associated abortion with women-related keywords. Such bias operated at the individual level and was directly tied to the actions and decisions of AI creators. While social bias mainly involved the broader, historically remaining prejudices that existed within society. These biases were rooted in cultural norms, institutional practices, and social structures, which shaped the collective behaviors and attitudes of individuals. Wang et al.\u0026rsquo;s (\u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) research explored AI systems designed to recommend careers to students, revealing such systems often recommended computer science or criminal justice careers to males and nursing or English to females. This result reflected deeply ingrained societal stereotypes that connected men to technical fields and women to caregiving or language-related domains.\u003c/p\u003e \u003cp\u003eAs for algorithmic bias, it was highlighted in Manasi et al.\u0026rsquo;s (\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) study. O\u0026rsquo;Connor and Liu\u0026rsquo;s (\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) experiments identified the bias in word embedding, which was a fundamental component in the NLP system and trained on a large number of text data. Word2Vec is a widely used word embedding model and learns semantic representations of words by analyzing their co-occurrence patterns in large text corpora (Mehmood et al., 2020). However, these data might contain gender bias inherently. O\u0026rsquo;Connor and Liu (2022) claimed that even gender-neutral words can become semantically correlated with a particular gender due to societal stereotypes encoded in the training data. For example, words like \u0026ldquo;nurse\u0026rdquo; and \u0026ldquo;homemaker,\u0026rdquo; though grammatically gender-neutral, are often associated with female pronouns in word embedding. If the descriptions of a particular gender were more positive or negative in the data, these biases would also be encoded into the embedding.\u003c/p\u003e \u003cp\u003e \u003cb\u003e3.3.3 Mitigation approaches of gender bias\u003c/b\u003e \u003c/p\u003e \u003cp\u003eGender bias in AI could be mitigated from 4 perspectives, including data level (N\u0026thinsp;=\u0026thinsp;9), algorithmic and model level (N\u0026thinsp;=\u0026thinsp;9), ethical and social level (N\u0026thinsp;=\u0026thinsp;9) and assessment and diagnosis (N\u0026thinsp;=\u0026thinsp;11). These approaches were not designed to directly correspond to the four causes of bias. The last approach, an indirect way of reducing gender bias, was to identify and evaluate bias, which could be applicable across all types of bias reviewed and provide a comprehensive framework for addressing AI\u0026rsquo;s gender bias.\u003c/p\u003e \u003cp\u003eRegarding data level, two critical processes were identified, specifically during the phases of data collection and data processing. For data collection, Busl\u0026oacute;n et al. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) recommended collecting diverse data to reflect adequate demographic features in the medicine and health field. Furthermore, recent studies have explored advanced data preprocessing techniques, such as gender masking, which involved removing gender-related information from textual data while maintaining semantic integrity. This technique resembled the approach outlined in Rizhinashvili et al.\u0026rsquo;s (\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) research, where gender parameters were eliminated to transform text into word embedding without any gender-related semantic information and this technique has been shown to reduce gender bias in the resume filtering system (Gagandeep et al., 2024). Additionally, most studies kept an eye on the data preprocessing before data training. The latest study verified the effect of preprocessing advertisement text in reducing gender bias in recommendation systems, like designing suitable text length (Cecere et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The primary objective was to prevent the model from capturing and amplifying gender stereotypes or prejudicial information embedded in the text through careful adjustment of text length. Effective data preprocessing techniques included downsampling and upsampling, etc. (Kelley et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), whose roles were to balance gender data to create fair models (Park et al., \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). One study explored the effect of applying upsampling and downsampling in the Neutral Machine Translation system, presenting that these two ways could diminish gender bias but lead to a significant decline in overall system performance (Tomalin et al., \u003cspan citationid=\"CR86\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAt the algorithmic and model level, it was tightly connected with the mitigation method from the data level. Because proper collection and processing of the data benefited the training of unbiased models. Various studies have underscored the necessity of modifying prompts in interactions with AI to offer more neutral and higher-quality outputs (Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kaplan et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). The prompt limit could also be interpreted as alleviating bias at the data level. From a purely model-training aspect, Noguero et al. (\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) were devoted to training fair and calibrated classifiers and the results proved the isotonic calibrator accurate. This post-processing technique adjusted the output probabilities of classifiers to align more closely with the actual probability distribution. Additionally, an innovative approach was introduced through a debiasing algorithm (Sun et al., \u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), which aimed to eliminate gender pair associations for gender-neutral terms while maintaining the utility of word embeddings in representing meaningful relationships and associations among words (O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFor the ethical and social level, public awareness, regulations and policies of AI, and management of the AI team were three vital points based on the selected papers. Busl\u0026oacute;n et al. (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) suggested that public awareness of AI\u0026rsquo;s gender bias should be enhanced and improved by education, which echoed Wang\u0026rsquo;s (\u003cspan citationid=\"CR92\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) study, directly pinpointing the importance for girls to have an anti-bias education. Furthermore, some regulations and policies of AI were established in practice to guarantee human rights. The European Commission provided a framework for AI to define the responsibilities of its users and developers (DiNoia et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and UNESCO (\u003cspan citationid=\"CR88\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) held the view that gender equality should be incorporated into AI technology. For the management team, diversity was a priority. The tech company should enhance their gender equality and inclusion to let them participate in the \u0026ldquo;inner work\u0026rdquo; (Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Adams and Loideain\u0026rsquo;s (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) study pointed to the urgency of reforming within the industry, revealing the significance of redistributing power in AI design (G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAssessment and diagnosis related to identifying or auditing gender bias in the algorithm. Howard and Borenstein (\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e) claimed it could be achieved by robot logic and this proposal was extended to design a computer-assisted diagnosis system in Larrazabal et al.\u0026rsquo;s (\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) study. In 2023, the idea of using AI to detect AI\u0026rsquo;s gender bias became popular (Newstead et al., \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR91\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), suggesting that researchers expected to envisage a win-win situation of reducing the gender bias of AI and users as their interaction.\u003c/p\u003e \u003cp\u003e \u003cb\u003e3.3.4 Challenges of mitigating gender bias\u003c/b\u003e \u003c/p\u003e \u003cp\u003eChallenges of mitigating gender bias derived from the performance of models (N\u0026thinsp;=\u0026thinsp;3), culture and value (N\u0026thinsp;=\u0026thinsp;4) and ethical concerns (N\u0026thinsp;=\u0026thinsp;4). Bernagozzi et al.\u0026rsquo;s (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) study revealed the competition between achieving accuracy and mitigating gender bias of translators. Similarly, it was found that the single-objective search technique could largely reduce the gender bias of models but with sacrificing semantic correctness (Hort et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), signaling that overemphasizing gender neutrality may compromise model performance and yield impractical outcomes (Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The cultural and value-related challenge implied that individuals\u0026rsquo; stubborn attitudes or stereotypes could impact the decision-making systems of AI unconsciously, which were tough to remove (Howard \u0026amp; Borenstein, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Ethical concerns pertained to the accountability and transparency of data (O\u0026rsquo;Connor \u0026amp; Liu, \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) as well as the design process. According to Sun et al. (\u003cspan citationid=\"CR84\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), enhancing the transparency of AI products through the involvement of researchers from diverse fields and the general public in collaborative decision-making could mitigate the risk of reinforcing existing gender biases and reduce the dominance of powerful stakeholders in AI technology.\u003c/p\u003e \u003cp\u003e \u003cb\u003e3.3.5 Consequences of gender bias of AI\u003c/b\u003e \u003c/p\u003e \u003cp\u003eConsequences of gender bias ranged from amplification of existing inequalities (N\u0026thinsp;=\u0026thinsp;4) and gender stereotypes (N\u0026thinsp;=\u0026thinsp;3) to unequal treatment (N\u0026thinsp;=\u0026thinsp;3), with these adverse effects predominantly impacting females rather than males. In the realm of health and medical research, scholars have noted that gender bias in AI could lead to the perpetuation of inequality and discrimination within society, particularly affecting women and other marginalized groups, thereby significantly influencing both healthcare systems and individual lives (Busl\u0026oacute;n et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). The underrepresentation of women in specific roles depicted in AI-generated images has intensified this inequality (Fang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), a trend also observed in AI-generated texts (Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These AI-based applications would perpetuate gender stereotypes to users and normalize them (Dwivedi et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; G\u0026oacute;rska \u0026amp; Jemielniak, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). As a result, women faced unequal treatment and were disadvantaged within society (Gross, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), a phenomenon explored in Yoder-Himes et al.\u0026rsquo;s (\u003cspan citationid=\"CR96\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) study, which revealed that female students scored lower on exams due to biases in facial recognition technology.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Publication and research design aspect (RQ1 and RQ2)\u003c/h2\u003e \u003cp\u003eRQ1 revealed the overall increasing research trend of AI’s gender bias and its cross-disciplinary characteristics. Based on data collected up to May 2024, the number of studies surged from 2022 to 2023 and reached its highest observed count in 2023. This surge may be attributed to the release of ChatGPT at the end of 2022, a groundbreaking language model from OpenAI that gained widespread adoption in 2023, reportedly amassing over 180\u0026nbsp;million users (Ghassemi et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The relevant studies were largely conducted, including the technological advancement and model optimization of ChatGPT (Aljebreen et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e) and ethical, privacy and risk considerations of ChatGPT (Ali \u0026amp; Djalilian, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), etc.\u003c/p\u003e \u003cp\u003eRQ2 elucidated the research methodology, an aspect that had not been addressed in earlier reviews. Our findings indicated that the majority of studies employed quantitative methods to statistically assess the performance and accuracy of models utilizing gender-imbalanced datasets. In the field of AI, gender bias is often represented by large-scale datasets and algorithmic models. Therefore, to accurately assess and quantify the degree of gender bias, researchers usually rely on quantitative studies to analyze these data by counting and analyzing the gender distribution and tendency in data sets (Yoder-Himes et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Noguero et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Research focus aspect\u003c/h2\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003e4.2.1 RQ3: What are the domains that present gender bias in AI?\u003c/h2\u003e \u003cp\u003eThe domains that presented gender bias in AI were detected. Previous research (Shrestha \u0026amp; Das, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) examined areas such as NLP, recommender systems, decision-making systems, AI assistants, and medicine. Our study expanded this investigation to include facial recognition, translation, assessment systems, and AI-generated text/image, which have recently garnered increased scholarly attention, especially after the launch of ChatGPT, sparking a global interest in more diverse applications of AI technologies. Thus, the results of RQ3 became broader with the updated research time. Given the significance of facial recognition as a vital sector of AI, the issues of gender bias within this area have been prioritized for further exploration, especially in light of ChatGPT’s release. Besides, the impressive capabilities of ChatGPT in NLP have spotlighted gender bias in translation and encouraged researchers to consider reducing gender bias in translation systems to enhance their accuracy and fairness (Gross, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). Furthermore, ChatGPT’s proficiency in communication and its immense potential of AI in generation also stirred reflection on gender bias in the assessment system and AI-generated text/image domain (Nah et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Newstead et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section3\"\u003e \u003ch2\u003e4.2.2 RQ4: What causes of AI’s gender bias are proposed?\u003c/h2\u003e \u003cp\u003eRQ4 explored the causes of AI’s gender bias, with human bias and data bias identified as the primary sources. These findings were consistent with Nadeem et al.’s (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) study, which explored AI’s gender bias in decision-making systems. Algorithmic bias and social bias were not highlighted specifically in many studies, which could be related to the tight and complex connection of different causes of gender bias. Algorithmic bias is often viewed as a consequence of human bias and data bias (Selbst et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e). In other words, algorithmic bias manifests as a result of human bias and data bias being embedded in the algorithm’s design and training. The relationship of social bias and human bias is complex but interconnected. Social bias can be seen as a foundational factor that contributes to the formation of human bias. Besides, social bias is not viewed as a direct cause of bias and usually serves as a background factor that influences both human behavior and data collection practices. For instance, societal gender stereotypes may shape the way data is labeled or the way algorithms are designed (Vallor, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Therefore, human bias and data bias are the primary sources of gender bias in AI and are explored most, accounting for 60% of the selected papers.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section3\"\u003e \u003ch2\u003e4.2.3 RQ5: What mitigation approaches for AI’s gender bias are proposed?\u003c/h2\u003e \u003cp\u003eFor the mitigation method, the study reviewed four kinds of methods. The solution of assessment was discussed frequently recently (Kaplan et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Newstead et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). Regular assessment and diagnosis could systematically uncover gender bias that may be deeply embedded in AI systems and not readily apparent. The detection tool was identified as AI itself most (Newstead et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). Bellamy et al.’s (\u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e) study introduced the AI Fairness 360 (AIF360) package, which mainly aimed to enhance the understanding of fairness metrics and mitigation strategies, providing a shared platform for fairness researchers and industry professionals to exchange and evaluate their algorithms. Later, Mosteiro et al. (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) applied the AIF 360 package to mitigate gender biases in clinical psychiatry. Furthermore, the emerging AI technologies have further expanded the scope of traditional mitigation methods mentioned above, offering innovative solutions to address gender bias. For instance, federated learning enhanced data-level mitigation by enabling the development of more diverse and representative datasets through decentralized training, thereby reducing the risk of bias (Gupta et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Han et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Kim et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Compared with the traditional methods of mitigating bias at the data level, such as data clarity and data balance, federated learning was an extended method, more focused on distributed data training. Besides, at the algorithmic level, adversarial debiasing methods actively counteracted biases by training models to minimize discriminatory outcomes, complementing traditional algorithmic fairness techniques (Correa et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). DSouza and French (\u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e) found that adversarial debiasing had the capacity to mitigate the negative effects of unfavorable attributes within a dataset, while also showing considerable promise in decreasing the production of fake news that arose from the intrinsic biases presented in the data. These new technologies not only complemented traditional mitigation methods but also opened new avenues for addressing gender bias in AI systems.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003e4.2.4 RQ6: What are the challenges of mitigating AI’s gender bias?\u003c/h2\u003e \u003cp\u003eRQ6 showed the challenges of mitigating gender bias in AI, ranging from the performance of models, culture and value to ethical concerns. Certain algorithmic approaches to achieving unbiased outputs may conflict with high performance, particularly when relying on extensive training iterations or complex model architectures. For instance, some studies suggested that excessive data usage in pursuit of unbiased models could lead to performance degradation, a phenomenon referred to as “Model Autophagy Disorder” (MAD) (Alemohammad et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The negative influence from the perspective of culture and value corroborated the idea of Tadimalla and Maher (\u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e), who claimed that the direction of AI development was determined by its creators, including individuals and organizations (Vallor, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). The results of ethical concerns were mainly due to the nature of the “black box” in AI systems for technological and ethical limits, indicating its inscrutable and complex features (Asatiani et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e; Marcus \u0026amp; Teuwen, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). Karasinski and Candiotto (\u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e) addressed that the black box prevented the indexation of the deviation because the systems used to process information were not controlled by creators and were not understood, resulting in output with obvious bias, which will adversely affect the inclusion drive. For ethical limits, commercial and privacy considerations may contribute to AI becoming a black box, indicating that varied stakeholders had different needs for the transparency and accountability of AI systems (Yang \u0026amp; Zhou, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section3\"\u003e \u003ch2\u003e4.2.5 RQ7: What are the consequences of AI’s gender bias?\u003c/h2\u003e \u003cp\u003eAlthough AI is a powerful and useful technology, there are certain hazards for its gender bias (Newstead et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The harms of amplifying inequalities and unequal treatment kept with the previous Donnelly and Stapleton’s (\u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e) research, pointing out that digital data systems facilitated social violence against marginalized communities and perpetuated gender bias. The consequences of gender stereotypes aligned with Fabris et al.’s (\u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e) study and Horvát and Bailón’s (\u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e) reports. The opacity of AI systems often hindered a direct understanding of the underlying logic behind their decisions. Consequently, when an AI system made a decision based on gender bias, it may be mistakenly perceived as reasonable or justified, thereby further reinforcing gender stereotypes (Fabris et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). If the AI model used historical biased data without correction, it may continue or even exacerbate historically gender inequality and the adverse consequences of gender bias were not just for the present, they may affect the process of gender equality in society in the long term (Franklin et al., \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Vallor, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). It is crucial to consider the mitigation methods and their challenges together, contributing to the ethical and responsible development of AI.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003e \u003cb\u003e5.1 Major findings\u003c/b\u003e \u003c/p\u003e\u003cp\u003eThe AI system has become an integral part of our daily lives and gender bias in AI has become an apparent concern. The present study reviewed 29 papers published on gender bias in AI systems, which were coded within 7 categories through thematic analysis both inductively and deductively. An overall increasing research trend and the distribution of research methodology were identified. We also detailed the domains that reflect AI’s gender bias. Finally, we presented the causes, consequences, mitigation methods and challenges of reducing bias from multiple aspects.\u003c/p\u003e\u003cp\u003e \u003cb\u003e5.2 Limitations\u003c/b\u003e \u003c/p\u003e\u003cp\u003eThis study acknowledges certain limitations. First, different keywords result in a varied pool of research papers. Though we established two groups of search strings referring to previous reviews, it was not enough, especially the keywords about AI. AI is a broad and general term, and retrieving related research comprehensively seems demanding, potentially resulting in the omission of some articles. Second, the classification of categories in bias causes, mitigation methods or consequences of AI’s gender bias is somewhat overlapping and connected, meaning that it is hard to clearly and wholly separate them and use them to label selected articles.\u003c/p\u003e\u003cp\u003e \u003cb\u003e5.3 Further research directions\u003c/b\u003e \u003c/p\u003e\u003cp\u003eThough gender bias can not be eliminated but mitigated (Solans et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), the process of mitigating bias is heavily obstructed by challenges, resulting in a dilemma that can not be easily solved. Further studies could concentrate on these challenges specifically. To balance model performance and biased output, it is essential to reconsider the acceptable levels of gender bias and establish the appropriate thresholds for model accuracy (Kiyasseh et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). For the contradiction between data transparency and privacy protection, it is important to explore the development of new machine learning algorithms that can effectively train AI models while safeguarding privacy and minimizing the risk of gender bias. Thus, further investigation into the efficacy of emerging techniques, such as differential privacy and federated learning, which allow for model training across diverse data sources while ensuring individual privacy, is warranted (Wang et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e; Yang et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eFurthermore, there is growing attention to AI’s gender bias, but research regarding how individual factors shape perceptions and attitudes towards this bias remains limited. Gupta et al. (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) verified that people with a culture of collectivism and uncertainty avoidance are more inclined to question AI’s gender bias. A recent study examined the influence of the social identity of individuals on treating biased AI (Ratajczak \u0026amp; Cockerill, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), but it focused on racial bias. Thus, further studies could proceed to explore the individual factors affecting the perception of gender bias. It is conducive to the development of policies and norms and system optimization, indicating that by understanding different personal factors, the designers can know how to improve, thus reducing bias and improving user satisfaction.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest/Competing interests:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author declares there are no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Publish declaration:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. The manuscript does not contain any individual person\u0026rsquo;s data in any form.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate declaration:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable. The study did not involve human participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval statement:\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNo.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eJingyue Liu: Conceptualization, Methodology, Investigation, Editing, and Writing-Original Draft.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAdams, R., \u0026amp; Loideain, N. N. (2019). Addressing indirect discrimination and gender stereotypes in AI virtual personal assistants: The role of international human rights law.\u003cem\u003e Cambridge International Law Journal\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(2), 241\u0026ndash;257. https://doi.org/10.4337/cilj.2019.02.04\u003c/li\u003e\n\u003cli\u003eAlami, H., Lehoux, P., Auclair, Y., de Guise, M., Gagnon, M., Shaw, J., Roy, D., Fleet, R., Ag Ahmed, M.A., \u0026amp; Fortin, J. (2020). Artificial Intelligence and Health Technology Assessment: Anticipating a New Level of Complexity. \u003cem\u003eJournal of Medical Internet Research\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e. https://doi.org/10.2196/preprints.17707\u003c/li\u003e\n\u003cli\u003eAlemohammad, S., Casco-Rodriguez, J., Luzi, L., Humayun, A., Babaei, H. R., LeJeune, D., Siahkoohi, A., \u0026amp; Baraniuk, R. (2023). Self-Consuming Generative Models Go MAD. \u003cem\u003eArXiv\u003c/em\u003e, \u003cem\u003eabs/2307.01850\u003c/em\u003e. https://doi.org/10.48550/arXiv.2307.01850\u003c/li\u003e\n\u003cli\u003eAli, M. J., \u0026amp; Djalilian, A. (2023). Readership Awareness Series \u0026ndash; Paper 4: Chatbots and ChatGPT-Ethical Considerations in Scientific Publications. \u003cem\u003eSeminars in Ophthalmology\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(5), 403\u0026ndash;404. https://doi.org/10.1080/08820538.2023.2193444\u003c/li\u003e\n\u003cli\u003eAljebreen, M., Alabduallah, B., Asiri, M. M., Salama, A. S., Assiri, M., \u0026amp; Ibrahim, S. S. (2023). Moth Flame Optimization With Hybrid Deep Learning Based Sentiment Classification Toward ChatGPT on Twitter. \u003cem\u003eIEEE Access\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 104984\u0026ndash;104991. https://doi.org/10.1109/ACCESS.2023.3315609\u003c/li\u003e\n\u003cli\u003eAsatiani, A., Malo, P., Nagb\u0026oslash;l, P. R., Penttinen, E., Rinta-Kahila, T., \u0026amp; Salovaara, A. (2020). Challenges of Explaining the Behavior of Black-Box AI Systems. \u003cem\u003eMIS Quarterly Executive\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(7), 259\u0026ndash;278. https://doi.org/10.17705/2MSQE.00037\u003c/li\u003e\n\u003cli\u003eBaker, R. S., \u0026amp; Hawn, A. (2021). Algorithmic bias in education. \u003cem\u003eInt. J. Artif. Intell. Educ.\u003c/em\u003e \u003cem\u003e32\u003c/em\u003e, 1052\u0026ndash;1092. https://doi.org/10.1007/s40593-021-00285-9\u003c/li\u003e\n\u003cli\u003eBellamy, R. K., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., Lohia, P. K., Martino, J., Mehta, S., Mojsilovic, A., Nagar, S., Ramamurthy, K. N., Richards, J. T., Saha, D., Sattigeri, P., Singh, M., Varshney, K. R., \u0026amp; Zhang, Y. (2018). AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias. \u003cem\u003eArXiv\u003c/em\u003e, \u003cem\u003eabs/1810.01943\u003c/em\u003e. https://doi.org/10.48550/arXiv.1810.01943\u003c/li\u003e\n\u003cli\u003eBelur, J., Tompson, L., Thornton, A., \u0026amp; Simon, M. (2021). Interrater reliability in systematic review methodology: Exploring variation in coder decision-making. Sociological Methods \u0026amp; Research, 50(2), 837\u0026ndash;865. https://doi.org/10.1177/0049124118799372\u003c/li\u003e\n\u003cli\u003eBernagozzi, M., Srivastava, B., Rossi, F., \u0026amp; Usmani, S. (2021). Gender Bias in Online Language Translators: Visualization, Human Perception, and Bias/Accuracy Tradeoffs. \u003cem\u003eIEEE Internet Computing\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e, 53\u0026ndash;63. https://doi.org/10.1109/mic.2021.3097604\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2006). Using thematic analysis in psychology. \u003cem\u003eQualitative Research in Psychology\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(2), 77\u0026ndash;101. https://doi.org/10.1191/1478088706qp063oa\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2022). Toward good practice in thematic analysis: Avoiding common problems and be(com)ing a knowing researcher. \u003cem\u003eInternational Journal of Transgender Health\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(1), 1\u0026ndash;6. https://doi.org/10.1080/26895269.2022.2129597\u003c/li\u003e\n\u003cli\u003eBrown, L. M. (2022). Gendered Artificial Intelligence in Libraries: Opportunities to Deconstruct Sexism and Gender Binarism. \u003cem\u003eJournal of Library Administration\u003c/em\u003e,\u003cem\u003e 62\u003c/em\u003e(1), 19\u0026ndash;30. https://doi.org/10.1080/01930826.2021.2006979\u003c/li\u003e\n\u003cli\u003eBusl\u0026oacute;n, N., Cort\u0026eacute;s, A., Catuara-Solarz, S., Cirillo, D., \u0026amp; Rementeria, M. J. (2023). Raising awareness of sex and gender bias in artificial intelligence and health. \u003cem\u003eFrontiers in global women\u0026rsquo;s health\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e, Article 970312. https://doi.org/10.3389/fgwh.2023.970312\u003c/li\u003e\n\u003cli\u003eCecere, G., Jean, C.\u0026dagger;., Le Guel, F., \u0026amp; Manant, M. (2024). Artificial intelligence and algorithmic bias? Field tests on social network with teens. \u003cem\u003eTechnological\u003c/em\u003e\u003cem\u003e \u003c/em\u003e\u003cem\u003eForecasting and Social Change\u003c/em\u003e\u003cem\u003e, 201\u003c/em\u003e, Article 123204. https://doi.org/10.1016/j.techfore.2023.123204\u003c/li\u003e\n\u003cli\u003eChang, X. (2023). Gender Bias in Hiring: An Analysis of the Impact of Amazon\u0026rsquo;s Recruiting Algorithm. Advances in Economics. \u003cem\u003eManagement and Political Sciences\u003c/em\u003e\u003cem\u003e, 23\u003c/em\u003e(1), 134\u0026ndash;140. https://doi.org/10.54254/2754-1169/23/20230367\u003c/li\u003e\n\u003cli\u003eChen, F., Wang, L., Hong, J., Jiang, J., \u0026amp; Zhou, L. (2024). Unmasking bias in artificial intelligence: A systematic review of bias detection and mitigation strategies in electronic health record-based models. \u003cem\u003eJournal of the American Medical Informatics Association\u003c/em\u003e, \u003cem\u003e31\u003c/em\u003e(5), 1172\u0026ndash;1183. https://doi.org/10.1093/jamia/ocae060\u003c/li\u003e\n\u003cli\u003eChen, Y. B., Zhai, Y. J., \u0026amp; Sun, S. J. (2024). The gendered lens of AI: Examining news imagery across digital spaces. \u003cem\u003eJournal of Computer-Mediated Communication\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(1), zmad047. https://doi.org/10.1093/jcmc/zmad047\u003c/li\u003e\n\u003cli\u003eChu, C. H., Donato-Woodger, S., Khan, S. S., Nyrup, R.,\u0026middot;Leslie, K., Lyn, A., Shi, T. Y., Bianchi, A., Rahimi, S. A., \u0026amp; Grenier, A. (2023). Age-related bias and artificial intelligence: A scoping review.\u003cem\u003e \u003c/em\u003e\u003cem\u003eHumanities and Social Sciences Communications\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, Article 510. https://doi.org/10.1057/s41599-023-01999-y\u003c/li\u003e\n\u003cli\u003eChung, H., Park, C., Kang, W. S., \u0026amp; Lee, J. (2021). Gender Bias in Artificial Intelligence: Severity Prediction at an Early Stage of COVID-19. \u003cem\u003eFrontiers in physiology\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, Article 778720. https://doi.org/10.3389/fphys.2021.778720\u003c/li\u003e\n\u003cli\u003eCoiro, P., \u0026amp; Pollak, D. D. (2019). Sex and gender bias in the experimental neurosciences: The case of the maternal immune activation model. \u003cem\u003eTrans\u003c/em\u003e\u003cem\u003eitional\u003c/em\u003e\u003cem\u003e Psychiatry\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e, Article 90. https://doi.org/10.1038/s41398-019-0423-8\u003c/li\u003e\n\u003cli\u003eCorrea, R., Pahwa, K., Patel, B., Vachon, C. M. (2023). Efficient Adversarial debiasing with Concept Activation Vector\u0026mdash;Medical image case-studies. \u003cem\u003eJournal of Biomedical Informatics\u003c/em\u003e, \u003cem\u003e149\u003c/em\u003e(11), Article 104548. https://doi.org/10.1016/j.jbi.2023.104548\u003c/li\u003e\n\u003cli\u003eDaneshjou, R., Smith, M. P., Sun, M. D., Rotemberg, V., \u0026amp; Zou, J. (2021). Lack of Transparency and Potential Bias in Artificial Intelligence Data Sets and Algorithms: A Scoping Review. \u003cem\u003eJAMA dermatology\u003c/em\u003e\u003cem\u003e, 157\u003c/em\u003e(11), 1362\u0026ndash;1369. https://doi.org/10.1001/jamadermatol.2021.3129\u003c/li\u003e\n\u003cli\u003eDelgado, J., de Manuel, A., Parra, I., Moyano, C., Rueda, J., Guersenzvaig, A., Ausin, T., Cruz, M., Casacuberta, D., \u0026amp; Puyol, A. (2022). Bias in algorithms of AI systems developed for COVID-19: A scoping review. \u003cem\u003eJournal of Bioethical Inquiry\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(3), 407\u0026ndash;419. https://doi.org/10.1007/s11673-022-10200-z\u003c/li\u003e\n\u003cli\u003eDiNoia, T., Tintarev, N., Fatourou, P., \u0026amp; Schedl, M. (2022). Recommender systems under European AI regulations. \u003cem\u003eCommunications of the ACM\u003c/em\u003e, \u003cem\u003e65\u003c/em\u003e(4), 69\u0026ndash;73. https://doi.org/10.1145/3512728\u003c/li\u003e\n\u003cli\u003eDonnelly, N., \u0026amp; Stapleton, L. (2021). Digital Enterprise Technologies: Do Enterprise Control and Automation Technologies Reinforce Gender Biases and Marginalization? \u003cem\u003eIFAC-PapersOnLine.\u003c/em\u003e \u003cem\u003e54\u003c/em\u003e(13), 551\u0026ndash;556. https://doi.org/10.1016/j.ifacol.2021.10.507\u003c/li\u003e\n\u003cli\u003eDSouza, K. M., \u0026amp; French, A. M. (2024). Fake news detection using machine learning: An adversarial collaboration approach. \u003cem\u003eInternet Research\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(5), 1664\u0026ndash;1678. https://doi.org/10.1108/INTR-03-2022-0176\u003c/li\u003e\n\u003cli\u003eDur\u0026aacute;n, R. P., Eisenhart, M., Erickson, F. D., Grant, C. A., Green, J. L., Hedges, L. V., Levine, F. J., Moss, P. A., Pellegrino, J. W., \u0026amp; Schneider, B. L. (2006). Standards for Reporting on Empirical Social Science Research in AERA Publications. \u003cem\u003eEducational Researcher\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e, 33\u0026ndash;40. https://doi.org/10.3102/0013189X035006033\u003c/li\u003e\n\u003cli\u003eDwivedi, S., Ghosh, S., \u0026amp; Dwivedi, S. (2023). Breaking the Bias: Gender Fairness in LLMs Using Prompt Engineering and In-Context Learning. \u003cem\u003eRupkatha Journal on Interdisciplinary Studies in Humanities\u003c/em\u003e\u003cem\u003e, 15\u003c/em\u003e, Article 4. https://doi.org/10.21659/rupkatha.v15n4.10\u003c/li\u003e\n\u003cli\u003eDwivedi, Y. K., Sharma, A., Rana, N. P., Giannakis, M., Goel, P., \u0026amp; Dutot, V.(2023). Evolution of artificial intelligence research in technological forecasting and social change: Research topics, trends, and future directions. \u003cem\u003eTechnological Forecasting and Social Change\u003c/em\u003e, \u003cem\u003e192\u003c/em\u003e, Article 122579. https://doi.org/10.1016/j.techfore.2023.122579\u003c/li\u003e\n\u003cli\u003eFabris, A., Purpura, A., Silvello, G., \u0026amp; Susto, G. (2020). Gender Stereotype Reinforcement: Measuring the Gender Bias Conveyed by Ranking Algorithms. \u003cem\u003eInf. Process. Manag.\u003c/em\u003e, \u003cem\u003e57\u003c/em\u003e, Article 102377. https://doi.org/10.1016/j.ipm.2020.102377\u003c/li\u003e\n\u003cli\u003eFang, X., Che, S. K., Mao, M. J., Zhang, H. Z., Zhao. M., \u0026amp; Zhao, X. F. (2024). Bias of AI-generated content: An examination of news produced by large language models. \u003cem\u003eSci. Rep\u003c/em\u003e\u003cem\u003e.,\u003c/em\u003e \u003cem\u003e14\u003c/em\u003e, Article 5224. https://doi.org/10.1038/s41598-024-55686-2\u003c/li\u003e\n\u003cli\u003eFife, S. T., \u0026amp; Gossner, J. D. (2024). Deductive Qualitative Analysis: Evaluating, Expanding, and Refining Theory. \u003cem\u003eInternational Journal of Qualitative Methods\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(6), 1\u0026ndash;12. https://doi.org/10.1177/16094069241244856\u003c/li\u003e\n\u003cli\u003eFranklin, G., Stephens, R., Piracha, M., Tiosano, S., LeHouillier, F., Koppel, R., \u0026amp; Elkin, P. L.(2024). The Sociodemographic Biases in Machine Learning Algorithms: A Biomedical Informatics Perspective. \u003cem\u003eLife-basel\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(6), https://doi.org/10.3390/life14060652\u003c/li\u003e\n\u003cli\u003eGagandeep, Kaur, J., Mathur, S., Kaur, S., Nayyar, A., Singh, S. P., \u0026amp; Mathur, S. (2024). Evaluating and mitigating gender bias in machine learning based resume filtering. \u003cem\u003eMultimed Tools Appl\u003c/em\u003e\u003cem\u003e.,\u003c/em\u003e\u003cem\u003e 83\u003c/em\u003e, 26599\u0026ndash;26619. https://doi.org/10.1007/s11042-023-16552-x\u003c/li\u003e\n\u003cli\u003eGalos, D. R., \u0026amp; Coppock, A. (2023). Gender composition predicts gender bias:A meta-reanalysis of hiring discrimination audit experiments.\u003cem\u003e Sci. Adv.,\u003c/em\u003e \u003cem\u003e9\u003c/em\u003e(18), esde7979. https://doi.org/10.1126/sciadv.ade7979\u003c/li\u003e\n\u003cli\u003eGhassemi, M., Birhane, A., Bilal, M., Kankaria, S., Malone, C., Mollick, E., \u0026amp; Tustumi, F. (2023). ChatGPT one year on: Who is using it, how and why? \u003cem\u003eNature\u003c/em\u003e, \u003cem\u003e624\u003c/em\u003e, 39\u0026ndash;41. https://doi.org/10.1038/d41586-023-03798-6\u003c/li\u003e\n\u003cli\u003eG\u0026oacute;rska, A. M., \u0026amp; Jemielniak, D. (2023). The invisible women: Uncovering gender bias in AI-generated images of professionals. \u003cem\u003eFeminist Media Studies\u003c/em\u003e,\u003cem\u003e 23\u003c/em\u003e, 4370\u0026ndash;4375. https://doi.org/10.1080/14680777.2023.2263659\u003c/li\u003e\n\u003cli\u003eGregory, A., \u0026amp; Denniss, A. R. (2018). An Introduction to Writing Narrative and Systematic Reviews-Tasks, Tips and Traps for Aspiring Authors. \u003cem\u003eHeart, lung \u0026amp; circulation\u003c/em\u003e,\u003cem\u003e 27\u003c/em\u003e(7), 893\u0026ndash;898. https://doi.org/10.1016/j.hlc.2018.03.027\u003c/li\u003e\n\u003cli\u003eGross, N. (2023). What ChatGPT Tells Us about Gender: A Cautionary Tale about Performativity and Gender Biases in AI. \u003cem\u003eSocial Sciences\u003c/em\u003e\u003cem\u003e, 12\u003c/em\u003e(8), Article 435. https://doi.org/10.3390/socsci12080435\u003c/li\u003e\n\u003cli\u003eGuevara-G\u0026oacute;mez, A., Z\u0026aacute;rate-Alcarazo, L. O., \u0026amp; Criado, J. I. (2021). Feminist perspectives to artificial intelligence: Comparing the policy frames of the European Union and Spain. \u003cem\u003eInf. Polity\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(2), 173\u0026ndash;192. https://doi.org/10.3233/IP-200299\u003c/li\u003e\n\u003cli\u003eGupta, M., Parra, C. M., \u0026amp; Dennehy, D. (2022). Questioning Racial and Gender Bias in AI-based Recommendations: Do Espoused National Cultural Values Matter? \u003cem\u003eInformation systems frontiers: a journal of research and innovation\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(5), 1465\u0026ndash;1481. https://doi.org/10.1007/s10796-021-10156-2\u003c/li\u003e\n\u003cli\u003eGupta, A., Maurya, M. K., Dhere, K., \u0026amp; Chaurasiya, V. K. (2024). Privacy-Preserving Hybrid Federated Learning Framework for Mental Healthcare Applications: Clustered and Quantum Approaches. \u003cem\u003eIEEE Access\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 145054\u0026ndash;145068. https://doi.org/10.1109/ACCESS.2024.3464240\u003c/li\u003e\n\u003cli\u003eHall, P., \u0026amp; Ellis, D. (2023). A systematic review of socio-technical gender bias in AI algorithms. \u003cem\u003eOnline Inf. Rev.,\u003c/em\u003e \u003cem\u003e47\u003c/em\u003e, 1264\u0026ndash;1279. https://doi.org/10.1108/oir-08-2021-045\u003c/li\u003e\n\u003cli\u003eHan, M., Zhu, T., \u0026amp; Zhou, W. (2024). Fair Federated Learning with Opposite GAN. \u003cem\u003eKnowl. Based Syst.\u003c/em\u003e, \u003cem\u003e287\u003c/em\u003e, Article 111420. https://doi.org/10.1016/j.knosys.2024.111420\u003c/li\u003e\n\u003cli\u003eHort, M., Moussa, R., \u0026amp; Sarro, F. (2023). Multi-objective search for gender-fair and semantically correct word embeddings. \u003cem\u003eAppl. Soft Comput.,\u003c/em\u003e \u003cem\u003e133\u003c/em\u003e, Article 109916. https://doi.org/10.1016/j.asoc.2022.109916\u003c/li\u003e\n\u003cli\u003eHorv\u0026aacute;t, E. A., \u0026amp; Bail\u0026oacute;n, S. G. (2024). Quantifying gender disparities and bias online: editors\u0026rsquo; introduction to \u0026ldquo;Gender Gaps in Digital Spaces\u0026rdquo; special issue. \u003cem\u003eJournal of Computer-Mediated Communication\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(1), zmad054. https://doi.org/10.1093/jcmc/zmad054\u003c/li\u003e\n\u003cli\u003eHoward, A. M., \u0026amp; Borenstein, J. (2017). The Ugly Truth About Ourselves and Our Robot Creations: The Problem of Bias and Social Inequity. \u003cem\u003eScience and Engineering Ethics\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e, 1521\u0026ndash;1536. https://doi.org/10.1007/s11948-017-9975-2\u003c/li\u003e\n\u003cli\u003eJeon, J., Lee, S., \u0026amp; Choi, S. (2023). A systematic review of research on speech-recognition chatbots for language learning: Implications for future directions in the era of large language models. \u003cem\u003eInteractive Learning Environments\u003c/em\u003e, 1\u0026ndash;19. https://doi.org/10.1080/10494820.2023.2204343\u003c/li\u003e\n\u003cli\u003eJin, E., \u0026amp; Eastin, M. (2023). Gender Bias in Virtual Doctor Interactions: Gender Matching Effects of Chatbots and Users on Communication Satisfactions and Future Intentions to Use the Chatbot.\u003cem\u003e International Journal of Human\u003c/em\u003e\u0026ndash;\u003cem\u003eComputer Interaction\u003c/em\u003e. https://doi.org/10.1080/10447318.2023.2279402\u003c/li\u003e\n\u003cli\u003eKaplan, D. M., Palitsky, R., Arconada Alvarez, S. J., Pozzo, N. S., Greenleaf, M. N., Atkinson, C. A., \u0026amp; Lam, W. A. (2024). What\u0026rsquo;s in a Name? Experimental Evidence of Gender Bias in Recommendation Letters Generated by ChatGPT. \u003cem\u003eJournal of medical Internet research\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e, e51837. https://doi.org/10.2196/51837\u003c/li\u003e\n\u003cli\u003eKarasinski, M., \u0026amp; Candiotto, K. B. B. (2024). AI\u0026rsquo;s black box and the supremacy of standards. \u003cem\u003eFilosofia Unisinos\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(1), 1\u0026ndash;13. https://doi.org/10.4013/fsu.2024.251.13\u003c/li\u003e\n\u003cli\u003eKelley, S., Ovchinnikov, A., Hardoon, D. R., \u0026amp; Heinrich, A. (2022). Antidiscrimination Laws, Artificial Intelligence, and Gender Bias: A Case Study in Nonmortgage Fintech Lending. \u003cem\u003eManufacturing and Service Operations Management\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(6), 2797\u0026ndash;3306. https://doi.org/10.1287/msom.2022.1108\u003c/li\u003e\n\u003cli\u003eKhan, M., \u0026amp; Ewuoso, C. (2024). Epistemic (in)justice, social identity and the Black Box problem in patient care. \u003cem\u003eMed Health Care and Philos.,\u003c/em\u003e \u003cem\u003e27\u003c/em\u003e, 227\u0026ndash;240. https://doi.org/10.1007/s11019-024-10194-y\u003c/li\u003e\n\u003cli\u003eKhosravi, M., Zare, Z., Mojtabaeian, S. M., \u0026amp; Izadi, R. (2024). Artificial Intelligence and Decision-Making in Healthcare: A Thematic Analysis of a Systematic Review of Reviews. \u003cem\u003eHealth Services Research and Managerial Epidemiology\u003c/em\u003e, \u003cem\u003e11, \u003c/em\u003e1\u0026ndash;15. https://doi.org/10.1177/23333928241234863\u003c/li\u003e\n\u003cli\u003eKim, D., Woo, H., \u0026amp; Lee, Y. (2024). Addressing Bias and Fairness Using Fair Federated Learning: A Synthetic Review. \u003cem\u003eElectronics\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(23), Article 4664. https://doi.org/10.3390/electronics13234664\u003c/li\u003e\n\u003cli\u003eKiyasseh, D., Laca, J., Haque, T. F., Otiato, M., Miles, B. J., Wagner, C., Donoho, D. A., Trinh, Q., Anandkumar, A., \u0026amp; Hung, A. J. (2023). Human visual explanations mitigate bias in AI-based assessment of surgeon skills. \u003cem\u003enpj Digit. Med., 6\u003c/em\u003e, Article 54. https://doi.org/10.1038/s41746-023-00766-2\u003c/li\u003e\n\u003cli\u003eKoo, S. (2022). Leading or Cheer-Leading? The Gender Gap in Political Smiles. \u003cem\u003ePolitics \u0026amp; Gender\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(1), 183\u0026ndash;211. https://doi.org/10.1017/S1743923X20000379\u003c/li\u003e\n\u003cli\u003eKurasaki, K. S. (2000). Intercoder Reliability for Validating Conclusions Drawn from Open-Ended Interview Data. \u003cem\u003eField Methods\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(3). https://doi.org/10.1177/1525822X0001200301\u003c/li\u003e\n\u003cli\u003eLarrazabal, A. J., Nieto, N., Peterson, V., Milone, D. H., \u0026amp; Ferrante, E. (2020).Gender imbalance in medical imaging datasets produces biased classifiers for computer-aided diagnosis. \u003cem\u003eProceedings of the National Academy of Sciences of the United States of America\u003c/em\u003e, \u003cem\u003e117\u003c/em\u003e(23), 12592\u0026ndash;12594. https://doi.org/10.1073/pnas.1919012117\u003c/li\u003e\n\u003cli\u003eManasi, A., Panchanadeswaran, S., Sours, E., \u0026amp; Lee, S. J. (2022). Mirroring the bias: Gender and artificial intelligence. Gender, \u003cem\u003eTechnology and Development\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e, 295\u0026ndash;305. https://doi.org/10.1080/09718524.2022.2128254\u003c/li\u003e\n\u003cli\u003eMarcus, E., \u0026amp; Teuwen, J. (2024). Artificial intelligence and explanation: How, why, and when to explain black boxes. \u003cem\u003eEuropean Journal of Radiology\u003c/em\u003e, \u003cem\u003e173\u003c/em\u003e, Article 111393. https://doi.org/10.1016/j.ejrad.2024.111393\u003c/li\u003e\n\u003cli\u003eMehmood, F., Ghani, M. U., Ibrahim, M. A., Shahzadi, R., Mahmood, W., \u0026amp; and Asim, M. N.(2020). A Precisely Xtreme-Multi Channel Hybrid Approach for Roman Urdu Sentiment Analysis. \u003cem\u003eIEEE Access\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, 192740\u0026ndash;192759, https://doi.org/10.1109/ACCESS.2020.3030885. \u003c/li\u003e\n\u003cli\u003eMosteiro, P., Kuiper, J., Masthoff, J., Scheepers, F., \u0026amp; Spruit, M. (2022). Bias Discovery in Machine Learning Models for Mental Health. \u003cem\u003eInformation\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(5), Article 237. https://doi.org/10.3390/info13050237\u003c/li\u003e\n\u003cli\u003eNadeem, A., Marjanovic, O., \u0026amp; Abedin, B. (2022). Gender bias in AI-based decision-making systems: A systematic literature review. \u003cem\u003eAustralas. J. Inf. Syst.,\u003c/em\u003e \u003cem\u003e26\u003c/em\u003e. https://doi.org/10.3127/ajis.v26i0.3835\u003c/li\u003e\n\u003cli\u003eNah, S., Luo, J., Kim, S., Chen, M., Mitson, R., \u0026amp; Joo, J. (2024). Algorithmic Bias or Algorithmic Reconstruction? A Comparative Analysis Between AI News and Human News. \u003cem\u003eInternational Journal of Communication\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e, 700\u0026ndash;729.\u003c/li\u003e\n\u003cli\u003eNewstead, T. P., Eager, B., \u0026amp; Wilson, S. (2023). How AI can perpetuate \u0026ndash; Or help mitigate \u0026ndash;Gender bias in leadership. \u003cem\u003eOrganizational Dynamics\u003c/em\u003e, \u003cem\u003e52\u003c/em\u003e(4), Article 100998. https://doi.org/10.1016/j.orgdyn.2023.100998\u003c/li\u003e\n\u003cli\u003eNoble, S. U. (2018). \u003cem\u003eAlgorithms of Oppression\u003c/em\u003e: \u003cem\u003eHow Search Engines Reinforce Racism\u003c/em\u003e. New York: NYU Press. https://doi.org/10.2307/j.ctt1pwt9w5.\u003c/li\u003e\n\u003cli\u003eNoguero, D. S., Ram\u0026iacute;rez-Cifuentes, D., R\u0026iacute;ssola, E. A., \u0026amp; Freire, A. (2023). Gender Bias When Using Artificial Intelligence to Assess Anorexia Nervosa on Social Media: Data-Driven Study. \u003cem\u003eJournal of medical Internet Research\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e, e45184. https://doi.org/10.2196/45184\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Connor, S., \u0026amp; Liu, H. (2023). Gender bias perpetuation and mitigation in AI technologies: Challenges and opportunities. \u003cem\u003eAI \u0026amp; Soc\u003c/em\u003e\u003cem\u003eiety\u003c/em\u003e, 1\u0026ndash;13. https://doi.org/10.1007/s00146-023-01675-4\u003c/li\u003e\n\u003cli\u003eOca, M. C., Meller, L., Wilson, K., Parikh, A. O., McCoy, A., Chang, J., Sudharshan, R., Gupta, S., \u0026amp; Zhang-Nunes, S. (2023). Bias and Inaccuracy in AI Chatbot Ophthalmologist Recommendations.\u003cem\u003e Cureus\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e. https://doi.org/10.7759/cureus.45911\u003c/li\u003e\n\u003cli\u003ePage, M. J. et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. \u003cem\u003eSystematic Reviews\u003c/em\u003e, \u003cem\u003e2021\u003c/em\u003e(10), 89\u0026ndash;100. https://doi.org/10.1186/s13643-021-01626-4\u003c/li\u003e\n\u003cli\u003ePark, J., Arunachalam, R., Silenzio, V., \u0026amp; Singh, V. K. (2022). Fairness in Mobile Phone-Based Mental Health Assessment Algorithms: Exploratory Study. \u003cem\u003eJMIR formative research\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(6), e34366. https://doi.org/10.2196/34366\u003c/li\u003e\n\u003cli\u003ePaul, S., Maindarkar, M. A., Saxena, S., Saba, L., Turk, M., Kalra, M. S., Krishnan, P. R., \u0026amp; Suri, J. S. (2022). Bias Investigation in Artificial Intelligence Systems for Early Detection of Parkinson\u0026rsquo;s Disease: A Narrative Review. \u003cem\u003eDiagnostics\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(1), Article 166. https://doi.org/10.3390/diagnostics12010166\u003c/li\u003e\n\u003cli\u003ePisanelli, E. (2022). Your resume is your gatekeeper: Automated resume screening as a strategy to reduce gender gaps in hiring. \u003cem\u003eEconomics Letters\u003c/em\u003e, \u003cem\u003e221\u003c/em\u003e, Article 110892. https://doi.org/10.1016/j.econlet.2022.110892\u003c/li\u003e\n\u003cli\u003ePrates, M. O. R., Avelar, P. H., \u0026amp; Lamb, L. C. (2020). Assessing gender bias in machine translation: A case study with Google Translate. \u003cem\u003eNeural Comput \u0026amp; Applic\u003c/em\u003e\u003cem\u003e.,\u003c/em\u003e \u003cem\u003e32\u003c/em\u003e, 6363\u0026ndash;6381. https://doi.org/10.1007/s00521-019-04144-6\u003c/li\u003e\n\u003cli\u003eRatajczak, R., \u0026amp; Cockerill, R. G. (2023). Artificial Intelligence in Violence Risk Assessment: Addressing Racial Bias and Inequity. \u003cem\u003eJournal of Psychiatric Practice\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(3), 239\u0026ndash;245. https://doi.org/10.1097/PRA.0000000000000713\u003c/li\u003e\n\u003cli\u003eRigby, M. J. (2019). Ethical Dimensions of Using Artificial Intelligence in Health Care. \u003cem\u003eAMA Journal of Ethics\u003c/em\u003e,\u003cem\u003e 21\u003c/em\u003e(2), 121\u0026ndash;124. https://doi.org/10.1001/AMAJETHICS.2019.121\u003c/li\u003e\n\u003cli\u003eRizhinashvili, D., Sham, A. H., \u0026amp; Anbarjafari, G. (2022). Gender Neutralisation for Unbiased Speech Synthesising. \u003cem\u003eElectronics\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(10), Article1594. https://doi.org/10.3390/electronics11101594\u003c/li\u003e\n\u003cli\u003eSelbst, A. D., Boyd, D., Friedler, S. A., Venkatasubramanian, S., \u0026amp; Vertesi, J. (2019). Fairness and Abstraction in Sociotechnical Systems. \u003cem\u003eProceedings of the Conference on Fairness, Accountability, and Transparency, \u003c/em\u003e59\u0026ndash;68. https://doi.org/10.1145/3287560.3287598.\u003c/li\u003e\n\u003cli\u003eShrestha, S., \u0026amp; Das, S. (2022). Exploring gender biases in ML and AI academic research through systematic literature review. \u003cem\u003eFrontiers in artificial intelligence\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e, Article 976838. https://doi.org/10.3389/frai.2022.976838\u003c/li\u003e\n\u003cli\u003eSolans N, D., Ram\u0026iacute;rez-Cifuentes, D., R\u0026iacute;ssola, E. A., \u0026amp; Freire, A. (2023). Gender Bias When Using Artificial Intelligence to Assess Anorexia Nervosa on Social Media: Data-Driven Study. \u003cem\u003eJournal of Medical Internet Research\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e, e45184. https://doi.org/10.2196/45184\u003c/li\u003e\n\u003cli\u003eSuen, H., \u0026amp; Hung, K. (2023). Building trust in automatic video interviews using various AI interfaces: Tangibility, immediacy, and transparency.\u003cem\u003e Comput. Hum. Behav.,\u003c/em\u003e \u003cem\u003e143\u003c/em\u003e, Article 107713. https://doi.org/10.1016/j.chb.2023.107713\u003c/li\u003e\n\u003cli\u003eSun, L. H., Wei, M., Sun, Y. B., Suh, Y. J., Shen, L. W., \u0026amp; Yang. S. J. (2023). Smiling Women Pitching Down: Auditing Representational and Presentational Gender Biases in Image Generative AI. \u003cem\u003eJournal of Computer-Mediated Communication\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(1), zmad045. https://doi.org/10.1093/jcmc/zmad045\u003c/li\u003e\n\u003cli\u003eTadimalla, S. Y., \u0026amp; Maher, M. L. (2024). Implications of Identity in AI: Creators, Creations, and Consequences. \u003cem\u003eProceedings of the AAAI Symposium Series\u003c/em\u003e,\u003cem\u003e 3\u003c/em\u003e(1), 528\u0026ndash;535. https://doi.org/10.1609/aaaiss.v3i1.31268\u003c/li\u003e\n\u003cli\u003eTomalin, M., Byrne, B., Concannon, S., Saunders, D., \u0026amp; Ullmann, S. (2021). The practical ethics of bias reduction in machine translation: Why domain adaptation is better than data debiasing. \u003cem\u003eEthics Inf. Technol.\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e, 419\u0026ndash;433. https://doi.org/10.1007/s10676-021-09583-1\u003c/li\u003e\n\u003cli\u003eTubadji, A., Huang, H., \u0026amp; Webber, D. J. (2021). Cultural proximity bias in AI acceptability: The importance of being human. \u003cem\u003eTechnol. Forecast. Soc. Change\u003c/em\u003e,\u003cem\u003e \u003c/em\u003e\u003cem\u003e173\u003c/em\u003e, Article 121100. https://doi.org/10.1016/j.techfore.2021.121100\u003c/li\u003e\n\u003cli\u003eUNESCO. (2020). Artificial intelligence and gender equality: Key findings of UNESCO\u0026rsquo;s Global Dialogue.`https://unesdoc.unesco.org/ark:/48223/pf0000374174. Accessed 20 May 2024\u003c/li\u003e\n\u003cli\u003eVallor, S. (2024). \u003cem\u003eThe AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking\u003c/em\u003e. New York: Oxford University Press. https://doi.org/10.1093/oso/9780197759066.001.0001\u003c/li\u003e\n\u003cli\u003eW\u0026aelig;raas, A. (2022). Thematic Analysis: Making Values Emerge from Texts. In G. Espedal, B. Jelstad L\u0026oslash;vaas, S. Sirris, A. W\u0026aelig;raas (Eds.),\u003cem\u003e Researching Values \u003c/em\u003e(pp. 153\u0026ndash;170). Palgrave Macmillan, Cham. https://doi.org/10.1007/978-3-030-90769-3_9\u003c/li\u003e\n\u003cli\u003eWang, C., Wang, K., Bian, A., Islam, R., Keya, K., Foulds, J. R., \u0026amp; Pan, S. (2023). When Biased Humans Meet Debiased AI: A Case Study in College Major Recommendation. \u003cem\u003eACM Transactions on Interactive Intelligent Systems\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e, 1\u0026ndash;28. https://doi.org/10.1145/3611313\u003c/li\u003e\n\u003cli\u003eWang, L. (2020). The Three Harms of Gendered Technology. \u003cem\u003eAustralasian Journal of Information Systems\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e. https://doi.org/10.3127/ajis.v24i0.2799\u003c/li\u003e\n\u003cli\u003eWang, Y. L., Wang, Q., Zhao, L. C., \u0026amp; Wang, C. (2023) Differential privacy in deep learning: Privacy and beyond. \u003cem\u003eFuture Generation Computer Systems\u003c/em\u003e, \u003cem\u003e148\u003c/em\u003e, 408\u0026ndash;424. https://doi.org/10.1016/j.future.2023.06.010.\u003c/li\u003e\n\u003cli\u003eYang, J. J., \u0026amp; Zhou, C. (2023). Algorithmic black-box problem in medical artificial intelligence: Ethical challenges and solution approach. \u003cem\u003eChinese Science Bulletin\u003c/em\u003e, \u003cem\u003e68\u003c/em\u003e(13), 1604\u0026ndash;1610. https://doi.org/10.1360/TB-2022-1320\u003c/li\u003e\n\u003cli\u003eYang, Q., Huang, A. B., Fan, L. X., Chan, C. S., Lim, J. H., Ng, K. W., Ong, D. S., \u0026amp; Li, B. (2023). Federated Learning with Privacy-preserving and Model IP-right-protection.\u003cem\u003e Mach. Intell. Res.,\u003c/em\u003e \u003cem\u003e20\u003c/em\u003e, 19\u0026ndash;37. https://doi.org/10.1007/s11633-022-1343-2\u003c/li\u003e\n\u003cli\u003eYoder-Himes, D. R., Asif, A., Kinney, K., Brandt, T. J., Cecil, R., Himes, P. R., Cashon, C. H., Hopp, R. M., \u0026amp; Ross, E. (2022). Racial, skin tone, and sex disparities in automated proctoring software. \u003cem\u003eFrontiers in Education\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e, 1\u0026ndash;16. https://doi.org/10.3389/feduc.2022.881449\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-psychology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"discpsy","sideBox":"Learn more about [Discover Psychology](https://www.springer.com/44202)","snPcode":"","submissionUrl":"","title":"Discover Psychology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"AI, Gender bias, Gender stereotype, Algorithmic bias, Human bias","lastPublishedDoi":"10.21203/rs.3.rs-9240627/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9240627/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial intelligence (AI) systems have been applied across various fields and the problems of AI\u0026rsquo;s gender bias become widely recognized, attracting researchers\u0026rsquo; attention. However, there is a lack of review articles that comprehensively consider publication, research design and research focus aspects related to gender bias in AI. Therefore, this study conducted a systematic review of 29 articles based on PRISMA to explore them. The findings revealed an overall increasing research trend of AI\u0026rsquo;s gender bias and identified 10 domains within AI that exhibited gender bias. Besides, it was found that the causes of AI\u0026rsquo;s gender bias could be attributed to data bias, human bias, algorithmic bias and social bias. Solutions were correspondingly proposed from data level, ethical and social level, algorithmic and model level, and assessment and diagnosis. The study also identified that amplification of existing inequalities, gender stereotypes and unequal treatment were three main consequences and the challenges of mitigating gender bias were generated from the performance of models, culture and value, and ethical concerns. Further study could proceed to study users\u0026rsquo; perception of AI\u0026rsquo;s gender bias to provide a comfortable experience environment and promote a healthy development of AI.\u003c/p\u003e","manuscriptTitle":"A systematic review and thematic analysis of gender bias in AI","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-23 09:19:33","doi":"10.21203/rs.3.rs-9240627/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-05-06T15:52:02+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-20T23:19:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"295969910898887471222670283162025333538","date":"2026-04-20T22:31:08+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-18T06:54:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-15T23:21:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"260583177465410335343749421594688581370","date":"2026-04-15T23:19:09+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-15T05:58:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"97147092919532846121377076776354337254","date":"2026-04-15T05:12:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"97585047952162094437555044792113852029","date":"2026-04-15T04:16:53+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-04-15T03:49:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-04-07T02:36:13+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-04-02T12:00:21+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-04-02T07:53:52+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Psychology","date":"2026-04-02T07:30:37+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-psychology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"discpsy","sideBox":"Learn more about [Discover Psychology](https://www.springer.com/44202)","snPcode":"","submissionUrl":"","title":"Discover Psychology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bf5d8098-2838-4fcf-90cb-94782c6d98f0","owner":[],"postedDate":"April 23rd, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Revision requested","date":"2026-05-06T15:52:02+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-05-06T15:54:41+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-23 09:19:33","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9240627","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9240627","identity":"rs-9240627","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}