Perceived Risks and Benefits of Disclosing ADHD to AI-based Educational Technologies: Semi-structured Interviews

Perceived Risks and Benefits of Disclosing ADHD to AI-based Educational Technologies: Semi-structured Interviews | 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 Research Article Perceived Risks and Benefits of Disclosing ADHD to AI-based Educational Technologies: Semi-structured Interviews Oriane Pierrès, Alireza Darvishy, Markus Christen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6106311/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study investigates the perspectives of students with attention deficit and hyperactivity disorder (ADHD) on disclosing their condition in the context of AI-based educational technologies (AI EdTech). Neurodivergent students often face challenges in disclosing their condition. It is unclear whether these difficulties persist in the context of AI EdTech. On the one hand, collecting data on neurodiversity could help ensure that these technologies are inclusive and personalized. Moreover, some students might find it easier to disclose their neurodivergence to an AI where their anonymity is guaranteed, rather than to colleagues or peers who may harbor negative attitudes. On the other hand, gathering and storing disability data might pose privacy risks depending on the technology design, such as the potential for re-identification, and may pressure neurodivergent students to disclose their conditions in order to access services. To better understand how disclosure is perceived in AI EdTech, we conducted 20 semi-structured interviews with students with ADHD. Results suggest that participants perceived AI tutors more positively than early warning systems due to a risk of stigmatization. This concern could be addressed by granting students greater control over their data, especially in deciding which lecturer should have access to their information. Still, participants were generally open to disclosing their ADHD status in AI EdTech, especially those who had already disclosed their ADHD to their universities. Finally, this paper provides reflections for developers and educators to create inclusive adaptive educational technologies that respect students’ privacy. Privacy AI Ethics Higher Education Accessibility Neurodivergent students Figures Figure 1 Figure 2 1 Introduction Students with disabilities and chronic conditions are expected to disclose their disabilities to their higher education institutions (HEI) if they want to receive reasonable accommodations such as additional time for exams, remediation of inaccessible learning materials, or sign language interpreters, to name a few. However, many students decide not to inform their institutions about their disabilities (Grimes et al., 2019; Newman & Madaus, 2015; Roberts et al., 2011). The disclosure process can be particularly difficult for neurodivergent students (Bertilsdotter Rosqvist et al., 2023; Moriña, 2024; Smith et al., 2021). Clouder et al. (2020) define neurodiversity as “an umbrella term, including dyspraxia, dyslexia, attention deficit, hyperactivity disorder, dyscalculia, autistic spectrum and Tourette Syndrome” (p. 757). The term was developed within the autistic activist communities (Botha et al., 2024). It recognizes that “differences in individual brain function and behavioral traits [are a] normal variation in the population” (Clouder et al., 2020, p. 758). In this paper, “neurodivergence” is the preferred expression, although “disability” is also used when referring to the field of disability research. Disability is understood from a socio-medical perspective, acknowledging that it results from both personal and environmental factors (World Health Organisation, 2021). As neurodiversity has no physical traits, students must decide whether to disclose their neurodivergence. Disclosure can be emotionally challenging due to past discrimination, a lack of understanding by academic staff, administrative burdens, or a desire to study "normally" without identifying as disabled (Moriña, 2024). As a result, many researchers advocate for the adoption of the universal design for learning (UDL) framework in higher education, because it reduces the need for disclosure (Mullins & Preyde, 2013; Osborne, 2019). UDL is a framework that acknowledges all students learn differently, emphasizing that failure to learn often lies in the environment (CAST, n.d.-a). It calls for flexible environments that accommodate individual needs by clarifying the purpose of learning, presenting information in diverse ways, and allowing learners to demonstrate their competencies and knowledge through various formats (CAST, n.d.-b). With UDL, learning environments may become inclusive for students with disabilities without requiring them to advocate for special adaptations to study (Osborne, 2019). Still, receiving reasonable accommodations is a student’s right according to the United Nations Convention on the Rights of People with Disabilities (Convention on the Rights of Persons with Disabilities Pledged, Article 24 – Education, 2006). When developing artificial intelligence-based education technologies (AI EdTech), developers and educators must weigh in on how much information they collect from their students. In higher education, AI usually refers to systems using machine learning techniques that employ data to make predictions or solve problems (Baker & Hawn, 2021). For instance, many researchers are trying to leverage big data to identify early students at risk of failing or dropping out in order to provide them with timely support (see e.g. Ciolacu et al. (2018); Ortigosa et al. (2019); S. Tsai et al. (2020)). Research has also been focusing on how to use newer Large Language Models (LLMs) to develop specialized learning assistants (Labadze et al., 2023). Collecting data on neurodiversity in AI EdTech could have several benefits. Firstly, these technologies are often trained on datasets in which disability is underrepresented which can lead to bias and errors for this group (Riazy et al., 2020). With this information, researchers can audit AI systems and ensure that their outcomes do not discriminate against neurodivergent students (Aboulafia et al., 2024). Secondly, integrating diverse learning needs into the design of AI EdTech could help personalize higher education. For example, current recommendation systems suggesting courses or learning materials rarely include accessibility factors in their algorithms (Pierrès, Christen, et al., 2024). Course recommendations could consider teaching practices that are known to support neurodivergent students, such as interactive and discussion-based course formats for students with ADHD (Flowers, 2012). Third, systems could adapt to students' needs without requiring human intervention. Many neurodivergent students choose not to disclose their disabilities due to the associated stigma and lack of awareness among academic staff (Moriña, 2024). For example, students have reported instances where professors would assume that they are faking their conditions or using it as an excuse for laziness (Osborne, 2019). As a result, an adaptive platform guaranteeing anonymity could appeal to some. There are, however, several concerns regarding the collection of disability data in AI EdTech. Legally, disability status is classified as health-related data, which the European Union considers sensitive (European Union, 2016). Due to its nature, this data should be processed with greater restraint and enhanced privacy protection. Integrating personal information on disabilities, chronic conditions, or neurodivergence into AI EdTech can compromise data anonymization, because certain conditions are rare, making re-identification easier (Morris, 2020). Additionally, persons with disabilities may be deprived of their choice to disclose or not. This can happen when a service would be inaccessible without disclosure (Aboulafia, 2024) or when the system can deduce a user’s disability from their data (Morris, 2020) or from their use of assistive technologies (Aboulafia, 2024; Marsh & Milne, 2024). In brief, there is a trade-off between disclosing neurodivergent status in AI EdTech to foster inclusion and protecting sensitive data. Investigating students’ opinions could enable the identification of conditions when disclosing one’s neurodiversity could be acceptable. These conditions can inform developers and higher education staff to understand how to provide personalized technologies while preserving privacy. This study focuses on the perspective of students with ADHD, as the term “neurodivergent” is very broad. Differences in behavior and opinions could diverge due to the diversity of the conditions. The overarching research question is: How do students assess the risk-benefits of disclosing their ADHD in the use of AI-based educational technologies? To answer this question, sub-questions were formulated. The first sub-question relates to the general perception of students with ADHD on the utility of two cases of AI EdTech, early warning systems and AI tutors. This overall opinion is relevant to ask because it may impact students’ readiness to disclose sensitive data. As a result, the following sub-question is raised: How do students with ADHD perceive the utility of AI EdTech? Three sub-questions are specifically connected to the disclosure of ADHD status in AI EdTech, asking for perceived risks and benefits as well as conditions under which it is acceptable to disclose: What are the students’ perceived risks of disclosing their ADHD to AI EdTech? What are the students’ perceived benefits of disclosing their ADHD to AI EdTech? Under which conditions are students willing to disclose their ADHD? 2 Theoretical Background 2.1 Students with ADHD in Higher Education From a medical perspective, ADHD is a neurodevelopmental condition that is usually diagnosed based on the identification of symptoms outlined in international psychiatric standards such as the Diagnostic and Statistical Manual of Mental Disorders 5 th Edition text revision (DSM-5-TR). These symptoms relate to inattention, hyperactivity, and impulsivity (American Psychiatric Association, 2022). From a neurodiverse perspective, ADHD is a variation of the brain with its own strengths and weaknesses (Colombo-Dougovito et al., 2020). The medical perspective often overlooks the positive aspects of ADHD as the diagnosis is based on individuals’ deficits. Yet, in a study, Dutch ADHD adults self-reported the following positive characteristics: creativity, being dynamic, flexibility, socio-affective skills, and higher-order cognitive skills (Schippers et al., 2022). While it is not clear whether these come from living with ADHD or coping mechanisms (Schippers et al., 2022), they highlight that ADHD cannot be reduced to a disorder. Disability research has sought to understand why some students with non-visible disabilities choose not to disclose their conditions and the consequences of not receiving adequate support (Clouder et al., 2020; Moriña, 2024). A literature review notably highlighted the tension between the universities’ requirement to disclose neurodivergence to request support and the uneven awareness of neurodiversity among academic staff (Clouder et al., 2020). At times, lecturers treat neurodivergent students poorly or in a discriminatory manner, do not provide support, and lack flexibility (Clouder et al., 2020). Consequently, students with disabilities engage in a rational complex reflection on whether to disclose their disabilities (Grimes et al., 2019). On the one hand, non-disclosure is a personal choice where students wish to study “normally” or do not view themselves as having a disability (Moriña, 2024). Still, a recurring issue is that learners want to avoid the associated stigma (Clouder et al., 2020; Moriña, 2024). For example, students reported situations when academic staff and peers treated them as lesser persons who did not belong to an HEI (Grimes et al., 2019). Additionally, because their conditions are not visibly apparent, they may face situations where others question the legitimacy of their experiences or diagnoses (Moriña, 2024; Osborne, 2019). Unsurprisingly, some students may feel more comfortable discussing their struggles with chatbots who would not judge them (Pierrès, Darvishy, et al., 2024). On the other hand, universities expect disclosure to grant reasonable accommodations that can eliminate barriers to study. For example, some students get a note-taker to let them concentrate on the course content. Others can also get additional time during an exam and / or can write an exam in a separate room to avoid distractions. However, the effectiveness of these accommodations is inconclusive for students with ADHD (Römhild & Hollederer, 2024). An explanation for this could be that the interventions are ill-fitted to students’ needs (Römhild & Hollederer, 2024). It could also be that causal effects are difficult to quantify as qualitative studies indicate positive effects of disability-related services on student success (Römhild & Hollederer, 2024). According to the literature review by Moriña (2024), not disclosing a disability can result in students feeling like they cannot be themselves or achieve as much as their peers. It can also affect their mental health negatively (Moriña, 2024). Consequently, health promotion interventions, coaching, as well as social and academic integration can become instrumental in ensuring the success of neurodivergent students (Clouder et al., 2020; Römhild & Hollederer, 2024). Clouder et al. (2020) also highlighted that adopting a universal design approach could reduce the need to disclose to receive specific adaptations. In brief, students are likely to weigh on whether disclosing their ADHD will benefit them. Sharing this information with an AI EdTech may be appealing as it could avoid stigma while providing a certain flexibility and personalized support. 2.2 Data privacy and AI EdTech After a review of the literature and to the best of the authors’ knowledge, there is no research focusing on how students with ADHD deal with data in higher education technologies. One study explored privacy and security concerns of students with disabilities in their use of assistive technologies, but it included only one neurodivergent individual (Marsh & Milne, 2024). Within the field of human-computer interaction, research on ADHD and technology often focuses on children and rarely investigates what people want or feel (Spiel et al., 2022). Understanding what students with ADHD in higher education want in AI could ensure that the technologies fulfill their needs. Research on data privacy in higher education was often conducted with the general population of students. This body of knowledge can inform how neurodivergent students think about data sharing. For these studies, researchers often refer to the privacy calculus theory, which explains the choice of disclosing personal information as a rational process that seeks to maximize benefits while avoiding negative consequences (Laufer & Wolfe, 1977). Culnan and Armstrong (1999, p. 106) explained that “in general, individuals are less likely to perceive information collection as privacy-invasive when a) information is collected in the context of an existing relationship, b) they perceive that they have the ability to control future use of the information, c) the information collected or used is relevant to the transaction, and d) they believe the information will be used to draw reliable and valid inferences about themselves.” With this theory, studies have looked for factors and reasons that explain data sharing in HEI. They have often found that students trusted their HEI to handle their data, but they were not aware of data policies (Jones et al., 2020; Soffer & Cohen, 2024; Y.-S. Tsai et al., 2020). However, Jones et al. (2020) argued that this did not mean that students did not care about data management and control. Students have expectations about the reasons for sharing data, the type of data, and who has access to the data. We know that acceptable reasons to share data are for education purposes (e.g. improving learning experience) and altruistic goals (e.g. improving a study program) (Jones et al., 2020; Y.-S. Tsai et al., 2020). Students are more reluctant to share personal data such as demographics (e.g. gender), personal academic records (e.g. grades), or online activities (e.g. time spent on a learning management platform) than pedagogical data (e.g. feedback on an assignment) (Soffer & Cohen, 2024; Y.-S. Tsai et al., 2020). The process of disclosing a disability in an AI EdTech could follow a sort of specific privacy calculus theory. Much like students are willing to share data for education purposes and altruistic goals, learners with disabilities are also willing to share information on their disabilities in assistive technologies if it could improve their or their peers’ access to learning (Marsh & Milne, 2024). A particularity that arises when technologies have assistive functions that the accessibility to a service (e.g. a course) is that students may not have the luxury to refrain from using a tool when privacy conditions are deemed unacceptable (Aboulafia, 2024; Marsh & Milne, 2024). 2.3 Early warning systems and AI tutors This work focuses on two types of AI EdTech due to their prevalence in the field: predictive learning analytics to detect at-risk or drop-out students called early warning systems (EWS), and intelligent tutoring systems (ITS) called AI tutors. Predictive learning analytics in education is a well-researched area. Researchers have used machine-learning models to predict students’ academic performance, risk of failing or dropping out, enrollment chances, engagement, and satisfaction (Sghir et al., 2023). This study focuses on the predictions of students’ risk of failing or dropping out as it is one of the most researched area in the field (Sghir et al., 2023; Zawacki-Richter et al., 2019). These predictions are used to create EWS, whose goal is to identify early students who may be struggling and provide them with adequate support (e.g, through email intervention, counselling) (Pierrès, Christen, et al., 2024). For example, German universities are considering using EWS to address the decline in youth interest in the fields of science, technology, engineering, and mathematics (STEM) (acatech - Deutsche Akademie der Technikwissenschaften & Joachim Herz Stiftung, 2024). To form predictions, researchers often use students’ online behavioral data (e.g. clicks on learning activities, time spent on a learning platform), prior and current academic data (e.g. grades, past courses), demographic and socio-economic information, and sometimes psychological features (Sghir et al., 2023). Although EWS follow a well-meaning goal, the reliance on online behavioral data such as the number of clicks or the time spent on an online learning platform as well as socio-economic information risk discriminate against students with disabilities (Pierrès, Christen, et al., 2024). Due to this risk, it is interesting to gather the opinion of students with ADHD on the use of EWS in higher education. Additionally, the case of EWS is worth discussing with students as the interventions following the predictions may involve lecturers, academic staff, and students differently. Lecturers are often the main decision-makers (Pierrès, Christen, et al., 2024). For example, lecturers are encouraged to contact via email students identified as at-risk (see e.g. Ciolacu et al. (2018) and Monllaó Olivé et al. (2020)). In other cases, counselling advisors are mainly intervening (see S.-C. Tsai et al. (2020)). Students are rarely involved in the decision-making process of EWS (Pierrès, Christen, et al., 2024). Hellings and Haelermans (2022) presented one of rare cases where students could monitor their predicted grade on a dashboard without the involvement of any other academic staff. Depending on the degree of involvement of the various stakeholders, students may be more or less ready to accept EWS and disclose their ADHD status. Along predictive learning analytics, researchers have sought to personalize learning with the development of ITS, i.e. applications that seek to “simulate one-to-one personal tutoring” (Zawacki-Richter et al., 2019, p. 4). The release of ChatGPT by OpenAI in November 2022 has spurred greater interest in ITS, notably due to its progress in providing feedback in the form of a dialogue (Batsaikhan & Correia, 2024). Those systems do not seek to eliminate human tutors, but rather complement them by providing always-available support independent of time and location (Batsaikhan & Correia, 2024). Educational platforms such as Khan Academy and universities are developing AI agents, called AI tutor, AI buddy or virtual assistant, that aim at assisting students along their studies (Baillifard et al., 2024; Bernstein et al., 2024; Johnson, 2019; Khan Academy, n.d.; Sajja et al., 2024). Possible functions include learning support, time management and administrative study organization, learning materials and course recommendations, and networking (see for instance Bernstein et al. (2024), Johnson (2019), and Sajja et al. (2024)). Additionally, the way students use ChatGPT could guide the design of universities’ AI tutors. For example, students with disabilities use ChatGPT to help them with different tasks such as studying (e.g. providing explanations, clarifying instructions, preparing for exams), writing, reading and research assistant, and self-organization (Pierrès, Darvishy, et al., 2024). The future of AI tutors is therefore likely to be unified applications capable of assisting students with various aspects of their studies instead of using multiple tools for specific tasks such as Grammarly for text editing, Co-pilot for programming, and calendar apps to manage meetings. Depending on its functionalities, AI tutors use different data types. Very often, they collect textual data and employ natural language processing techniques to analyze them (Pierrès, Darvishy, et al., 2024). Textual data can include students’ input in a chat interface, learning materials or even automatically transcribed course recording. For example, Sajja et al. (2024) assessed students’ chat request to detect their emotional state and make the AI tutor respond empathetically if necessary. With textual data, there is a critical distinction between what constitutes data and information. Data is a raw piece of text that does not necessarily contain information that can be used (Boisot & Canals, 2004). For instance, a student could describe symptoms of ADHD and mention the condition in their input, but without explicit extraction of this information an AI tutor will not necessarily categorize this student as having ADHD and adapt their recommendations and answers. In comparison, if the AI tutor was specifically designed to detect and analyze mentions of ADHD in a text, it could turn textual data into textual information. Information, although often used as a synonym for data, is structured data that can enhance an agent’s understanding (Boisot & Canals, 2004). AI tutors can for instance collect information on academic schedules, the courses chosen by a student to help them organize their time. Another piece of information is socio-demographic information. It could be imagined that students can inform directly whether they have ADHD or not in the hope that recommendations and interactions will be more ADHD-friendly. For that reason, it is interesting to ask for students with ADHD how confident they would be to disclose this information. 3 Methodology Semi-structured interviews were conducted to investigate whether and how students are willing to disclose their ADHD in AI EdTech. This qualitative approach enables the identification of factors that could not have been uncovered from previous research on privacy and educational technologies focusing on the general population of students. Additionally, participants filled out a 5-minute online survey before starting the interviews. This questionnaire consisted of closed questions regarding demographics, information on their ADHD diagnosis, and their estimated knowledge of AI. This information was mainly used to describe the sample of participants and to check for patterns in the response of subgroups of participants. 3.1 Interview guide An interview guide was developed in German based on the existing literature; it is available as supplementary material. The interview starts with open questions regarding general experience and opinions on ADHD disclosure in higher education. Then, two cases of AI EdTech are presented to the participants: 1) EWS identifying students who could require support (e.g. additional exercises or explanation from an instructor), and 2) AI tutors. After the two cases were presented and participants’ possible clarification questions were answered, interviewees were asked about their opinion on the use cases and whether and how they would disclose their ADHD. The interviews concluded with questions reflecting on the two use cases. 3.2 Sampling Students with ADHD enrolled in Swiss-German HEI were recruited via e-mail. To ensure the survey reached students who had not disclosed their ADHD to the university, information was disseminated through two universities’ survey channels for the general population of students, LinkedIn, and three universities’ “marketplace” where people can post job adverts and the like. We also sent an email to students previously involved in the authors’ research studies. One university's disability-related services also informed their students about the study. Additionally, we used the online survey data to balance the number of students who disclosed their ADHD status at their HEI and those who did not as well as their gender. 3.3 Data collection Before filling out the short online survey, students had to read and accept a consent form providing information on the objectives of the study, a general overview of the questions, data usage, and storage. Participants were invited to ask questions if they had any. Then, before the interviews, participants were reminded about study participation conditions, in particular, that the exchange would be recorded and transcribed automatically with Microsoft Teams. The transcripts were corrected by the first author and a research assistant following an intelligent verbatim style, i.e., the corrector deleted repetitions or verbal fillers such as “umm” (McMullin, 2023 ) because only the content of the answers was analyzed. During the interviews, the first author kept a diary of notes and reflections occurring during the phase. The 20 interviews were conducted in July 2024. 3.4 Data analysis Two researchers conducted a content analysis of the interview transcripts, following Mayring’s method (2014) on deductive and inductive category assignment. First, the two coders worked independently using an initial coding table based on existing literature and aligned with the interview questionnaire structure. New codes were also created inductively. To ensure consistency in coding with the initial table, the two coders met after analyzing two transcripts to discuss their approaches before proceeding with further independent coding. After coding all transcripts independently, the researchers fully reviewed and discussed three other transcripts to harmonize their codes. The discussions revealed only unsubstantial differences in their coding decisions. Subsequently, the first coder compared all the codes and restructured them into bigger categories. The second coder then reanalyzed the transcripts using this revised code structure. When both coders had analyzed all transcripts, the two researchers discussed any remaining discrepancies and finalized the codes. Finally, the transcript codes that were exclusive were combined with the survey data and analyzed visually with R to identify patterns. 3.5 Ethics committee approval Due to the involvement of human participants and the necessity to record the interviews for analysis, the study design was presented to and approved by the ethical committee of the authors’ university. 4 Results 4.1 Sample description The sample was composed of 10 female and 10 male participants. Participants were aged between 22 and 50 years old, with a median of 26. One person did not disclose their birth year. Six interviewees indicated having a migration background. Most students (17) were enrolled in a bachelor's program, two in a master's, and one in a PhD. A majority of them (15) studied social and business sciences, two natural sciences, and one engineering. Two other students were enrolled in interdisciplinary programs, one combining computer science and linguistics, and the other medicine and social sciences. Most participants (11) assessed their knowledge of AI at the user level. Six indicated that they were interested in the subject but had no technical skills and three had already coded at least one small AI model. A large majority of the interviewees (13) found out they had ADHD after their 18th birthday, i.e. during adulthood. All but one had a formal medical diagnosis. Half of the sample received reasonable accommodations whereas the other half did not. Eight identified as a person with disabilities. For most participants (18), the ease of talking about ADHD depended on the context or the person they were conversing with. Very often, they would explain that some people are open and interested while others would simply not understand. As a result, eleven students mentioned situations when they find it easy to talk about ADHD such as with friends, with fellow students with ADHD, or in social or psychological studies. Fourteen participants reported not disclosing their ADHD in some situations, often to avoid negative reactions or because they did not see any benefit in sharing that information. Two also explained that they preferred to adapt and tried to solve issues by themselves. Overall, past experiences with HEI staff and other students were positive. 4.2 Perceived utility of two cases of AI EdTech A majority of participants preferred the use of AI tutors over EWS. While students saw the support possibilities with AI tutors, they expressed more concerns (e.g. discrimination) with the adoption of EWS in HEI. 4.2.1 Use case: Early warning systems (EWS) In the interviews, participants were invited to imagine an EWS that could provide early support to students predicted to require help. Three scenarios were presented: Intervention without humans : an automatic system contacts students via e-mail directly to inform them that they were identified as potentially requiring support. The e-mail recommends contacting a relevant person (e.g. lecturer) or reading additional learning materials. Intervention with lecturers : the list of students requiring help is sent to lecturers who then proactively seek to support them by offering their help or starting a discussion. Structural intervention for future students : the information is used at the group level to change the university structurally (e.g. curricula, provision of support offers targeting specific groups). In general, participants saw the added value of offering proactive support to students. However, an often-recurring concern was that such a system could be discriminatory, especially in the case of interventions one and two. In total, 12 students explained that predictions could be inaccurate. They reflected on the fact that previous grades in secondary education would not necessarily mean they would fail a class in higher education. Six explicitly mentioned that ADHD characteristics could influence predictions, such as the tendency to start studying later in the semester or the need to move while learning. For example, Participant 4 explained this in the following manner: You mentioned the activities on learning platforms. Now, for example, when you say that my clicks are somehow looked at, how often I do the exercises and if I don't do anything for a long time, I might get a warning, an early warning. Well, as a person who always does everything at the last second, I would get a lot of messages saying: Hey, you haven't done anything on our exercises yet, why don't you do it? That would stress me out even more. Concerns for inaccurate predictions were particularly named when considering intervention without humans. This also led eight students to say that receiving an automatic e-mail would trigger negative emotions such as stress, demotivation, uncertainty about why they would be assessed in this manner or feeling patronized. Additionally, half of the interviewees believed that an automatic system without human intervention could be inefficient or useless. In particular, seven participants reckoned that learners already know when they are struggling, and that support offers already exist. Others thought that an e-mail was easy to ignore and would not necessarily motivate them to ask for support. On the other hand, seven students highlighted the benefits of proactive support in situations when students may be unaware of their struggle or of existing solutions or they are too shy to ask for help. In the case of the intervention with a lecturer, six participants highlighted how a lecturer can truly provide proactive and individualized support. Participant 16 notably emphasized that human intervention could be more thoughtful (“taktgefühl” in German) and would allow him to explain himself. Nevertheless, seven expressed strong concerns related to the risk of having lecturers stereotype and label students which could lead to unfair grading. For example, Participant 2 said the following: I think it's really bad, I don't think it's a good thing at all. I think it leads to a lot more stigmatization of people based on stereotypical assumptions about who needs help and who doesn't. And especially when the list goes to lecturers, we know from studies that if you tell teachers in advance, “This student is very good and this one is very bad”, they will grade accordingly, regardless of intelligence level. So, I think it leads to stigmatization and that people are then treated according to the bias they have. Compared to the first two interventions, the third intervention for structural measures was positively perceived. Reasons for this were that the focus was less on the individual and that this could encourage universities to review their course structures, as explained by Participant 4: I prefer [the intervention with structural measures] best because it is not so individualized and because I think it would force the university to make structural changes to a course that isn't working so well with different groups of people, for example, and not simply put the blame on the individuals, in the sense of: Ah, you're not getting through, that means you need extra support. Instead, we say: No, something about the course is not right. We're looking at how we can reach more people. That's what I like better. Participants were also invited to reflect on how the design of EWS could be improved. Eight interviewees emphasized they would like to remain in control, i.e., receiving a prediction and following the support advice should be voluntary. Six also wished for detailed information on their performance, showing both strengths and weaknesses. Participant 4 argued that a positive and supportive formulation would be important for her to reduce additional stress. In the quote below, she emphasized that this aspect is crucial for individuals with ADHD: I think the question made me realize in general that, if [early warning systems] happen at some point, it has to be done with a lot of sensitivity, because even if it's AI, it's still people who are affected with their feelings. Especially with ADHD, along with our comorbidities, where maybe you get depressed faster or feel more anxious about rejection, especially in such cases, you have to choose your words very carefully when something like that comes up. Additionally, five mentioned that the intervention could be improved by focusing on how to support at-risk students (e.g. providing resources or exercises, sending task reminders). Two students suggested integrating human oversight into the system. 4.2.2 Use case: AI Tutor The other use case described an AI tutor that could support six functions: Studying and preparing for an exam : The AI could prepare students for an exam through discussion (as a sparring partner). It could also answer questions about the lesson content. Explanations of exercises and instructions : The AI could explain the instructions of an exercise without giving the answer. Editing : The AI could edit text and suggest improvements. Advice on courses and learning materials : The AI could recommend courses and learning materials. For example, if a student prefers to learn with videos, the AI could suggest learning videos instead of articles. Support for time management and organization : The AI could, for example, create a learning plan and send reminders for important deadlines at the university. Support networking among students : For example, the AI could recommend groups of students with common interests. In general, students viewed an AI tutor as a useful tool, though certain functions garnered more interest than others. The most valuable feature would be its support for time management and organization. Among the 16 participants who mentioned benefits, 10 linked this utility to challenges related to ADHD symptoms or personal difficulties. Often, students mentioned the difficulty of creating a realistic study plan and stick to it. However, four persons doubted an AI could be helpful because they believed the study plan would be too strict for them and could cause more stress than relief. Participant 14 highlighted how it could be better to help him assess the amount of work rather than providing a fixed plan: “With ADHD and procrastination and so on, it's quite difficult with deadlines and systematics and so on. Systematic in the sense of a regular habit. For example, if it told me: “You have to spend an hour every week on this subject, an hour on this subject, half an hour on this subject.” For me, that's much less useful than when it structures me like this: “Here are the topics, this is the scope of topics here and there. This is the expected amount of time and work.” Because then it's not so much telling me what I have to do, but what I should do. And this distinction is really important for me when it comes to my own time management, because I can't follow my own time management very well either way.” Eleven interviewees positively perceived the use of an AI Tutor to help with studying and preparing for an exam. Among them, five participants highlighted that it would enable them or other students to ask questions they would not dare ask in front of others or the lecturer. Still, two persons expressed concerns that the function would not help them learn effectively and one mentioned the risk of losing human contact. Additionally, Participant 9 emphasized the importance of the tool being optimized for the course, explaining that in his math and physics courses, it is essential for the problem-solving methods to align with those taught in class. Half of the participants valued the text editor function, notably to help detect careless mistakes (“Flüchtigkeitsfehler” in German) in writing often associated with ADHD. Nevertheless, four participants raised concerns about potential text standardization, diminished development of writing skills, or the risk of text appearing plagiarized. Two students were also skeptical it could help them as they needed more support with the content of the text rather than spelling. Two others also explained that they already had good editing tools. Seven participants reported benefits for the feature “explanation of exercises and instructions”. In particular, four mentioned that they had previously struggled to understand instructions which had hindered their learning. Two also added that this feature would make them less dependent on other students. Nonetheless, three interviewees argued that the feature might reduce learning if students did not spend enough time trying to understand the task on their own. Opinions on the feature “advice on courses and learning materials” were more mitigated. On the one hand, eight persons saw possibilities to personalize learning or to facilitate course selection. On the other hand, six believed that this feature would not be useful because they already knew how to select their courses or that they believed that this was a lecturer’s task. In comparison to other features, networking support was perceived as less useful. Students would not necessarily see why an AI could help with this task as networking possibilities occur by themselves, especially in small study programs. Others also found it too personal or did not like learning with others. Participants were also asked about their preference between using an editing tool from a private company (e.g. Grammarly) and their university. The large majority of interviewees perceived positively that universities provide such tools, notably due to the belief that their data would be better protected. Four also mentioned that the tool could be better optimized for their academic needs. At the same time, four participants questioned whether universities have the capacity to provide tools as good as those from private companies. Additionally, two interviewees raised concerns that university staff could misuse the information. For example, lecturers could evaluate the original version of a text before it was modified with AI, thus making the recourse to AI-based text editing useless. 4.3 Willingness and reasons to disclose ADHD in AI use cases Figure 1 illustrates the reported willingness to share ADHD status in the two AI use cases among the students who disclosed their ADHD at their HEI and those who did not. Those who had disclosed their ADHD to their HEI were in general more willing to share their neurodivergence with an AI-based system, regardless of the use case. Participants who had not informed their HEI that they have ADHD were more reluctant, especially with EWS. In EWS, 13 participants would share their ADHD status hoping to increase accuracy or improve intervention. For example, Participant 19 argued that interventions would need to take into account whether the person has ADHD or not: Speaking again from my own experience: at the beginning of my studies, the problem for me was not that I didn't fundamentally understand the subject matter, but simply the quantity, how to deal with the material “how do I learn efficiently? How do I prioritize, how do I create a structure?” That's why I simply have the feeling that these measures, if they are based on this fictitious system, can actually be tailored quite differently for ADHD. So I see the possibilities of providing better interventions with ADHD. Two participants also mentioned they would share their ADHD data because it could help others receive adequate support. In total, 7 interviewees said they would not disclose their ADHD in an EWS, even if in two cases, the students reckoned that it could help intervene according to neurodivergent students’ needs. Reasons for not sharing were mainly due to bias risk and a belief that sharing that information would not be useful. Fourteen students said they would disclose their ADHD to an AI tutor mainly to enable greater personalization. Among them, eight explained that time management could become more flexible and send more reminders. Two participants also imagined that the system could thus recommend learning resources that are more helpful for individuals with ADHD. Additionally, Participant 20 mentioned that the tool could be specifically designed to enhance concentration thanks to shortened paragraphs, color choice or a reminder to take short breaks. Two persons said that it could help students with ADHD connect with one another. Four participants answered that they would not share their ADHD with an AI tutor. They explained that everybody learns differently, ADHD affects people uniquely, and having ADHD would not significantly impact their learning experience with an AI tutor. One person also elucidated that they would rather disclose their ADHD to a human. Apart from the two use cases, participants were invited to talk about where they found information about their condition and whether they already disclosed their ADHD in an AI system. While mental health professionals such as therapists remain a primary source of information, 18 interviewees mentioned looking up the internet. For 11 students, social media was a source of information. In some cases, this is how they came to think they might have ADHD. Among them, three explained that they did not actively search for information, but that content was suggested in their timelines. Additionally, although only one person indicated that they wrote in an AI system that they have ADHD, four other persons mentioned asking questions related to ADHD in ChatGPT. 4.4 Conditions to share data The 20 participants were asked “Who should have access to your data?” when discussing each of the presented AI use case (Fig. 2 ). As the question was open, the total number of responses is not equal for each stakeholder. In the question, the term “data” was not explicitely defined. However, the interviewer mentioned online interaction data (e.g. click on activities on a learning platform or time spent) and previous grades when describing the EWS. For the AI tutor, the use of textual data was implied. When participants asked for clarification, the interviewer named these types of data. Additionally, the question followed one asking whether students would share information about their ADHD and the interviewer reminded them that this was also a type of information that could be shared. In both cases, access for lecturers was more controversial, notably due to stigmatization risk or negative consequences on grading. In the case of the AI tutor, some students pointed out that it would change their interaction with the tool. Still, three interviewees explained they would grant access to lecturers in an EWS because instructors are more apt to intervene adequately than in an automatic system. Another person mentioned that they can control for errors. Two persons emphasized how important it is for them to remain in control to decide which lecturer gets access to data. For example, Participant 10 drew a parallel with the procedure for notifying lecturers about reasonable accommodations: instructors are informed of the accommodations but not the students’ specific conditions, allowing students to choose whether to discuss the subject with lecturers individually. In the case of AI tutors, seven participants mentioned that lecturers could have access to their data to improve the course in the future. Three of them emphasized that data should be anonymized for this. For example, Participant 1 suggested showing only the most frequently asked questions. Access for faculty staff was in both use cases more acceptable, notably because participants imagined a more anonymous use where the university seeks to improve course quality for future students rather than intervene individually. In general, anonymity was central in AI tutors as 75% of participants would prefer data to be stored anonymously. Six participants first answered that data were to remain between them and the AI tool. Eleven interviewees would convene that IT team should have access to data to improve the tool. In comparison, in an EWS, despite that 13 would prefer anonymized data, nine also would accept logging with their name as this enables them to receive support. As a result, eight students highlighted the importance for them to be able to decide whether they want to use such a tool, who has access to information and whether they want to follow through support recommendations. This preference for greater involvement in decision-making was evident as five individuals expressed a desire for detailed information on how the prediction was made. 5 Discussion The interviews with students with ADHD have highlighted a clear difference between AI EdTech-supporting lecturers’ tasks and those assisting students. AI tutors were perceived positively and seen as a tool to complement learning and overcome personal difficulties. While participants recognized the value of providing proactive support with EWS, their concerns about labeling and stereotyping called for greater student control. As highlighted by Culnan and Armstrong ( 1999 ), when individuals believe inferences drawn from their data to be incorrect, they experience the data collection as an invasion of privacy. In comparison, the unwillingness to disclose ADHD to AI tutors related to the lack of relevance. The general student population is also concerned about the risk of surveillance associated with the increased use of data to optimize HEIs (Jones et al., 2020 ). However, focusing on the perspective of students with ADHD or students who face systemic barriers due to their race, gender, or disabilities, has value because 1) they show concerns that could increase existing inequalities that HEIs are committed to eliminating, and 2) ensuring that systems do not disadvantage these students is likely to benefit all students. In this study, interviewees regularly connected their concerns or opinions (positive and negative) with their experience as an individual with ADHD. In particular, the concern for stigmatization and labeling cannot be ignored in light of the existing research showing that neurodivergent students still report discrimination in tertiary education (Clouder et al., 2020 ; Grimes et al., 2019 ; Moriña, 2024 ; Osborne, 2019 ). 5.1 Considerations on human involvement data policies in AI EdTech Interestingly, students in this research did not necessarily oppose humans to the supposed anonymity of technology. Several students emphasized that human bias is replicated in AI-based systems. An important factor was who had access to their data, naming those with greater influence on their study path and career (i.e. lecturers) as more critical. Li et al. ( 2022 ) found that students’ willingness to consent to leaning analytics depends on their comfort with instructors using their data for learning engagement. They argued that lecturers build relationships with students, fostering trust and increasing acceptance of data sharing compared to requests from administrative staff (Li et al., 2022 ). However, our study suggests that students with ADHD may view non-grading academic staff as less intimidating, as learners can remain anonymous and do not risk negative academic consequences. This should encourage developers to identify the most acceptable stakeholders to intervene with AI-based systems. Kizilcec ( 2024 ) argued for more research on how lecturers perceive AI EdTech because they are the final decision-makers. However, Despande and Sharp (2022, p. 233) explained that “users of the system are the most relevant stakeholders when considering who is likely to be impacted by responsible AI systems” and continued by identifying those underrepresented in datasets as most likely to be affected by AI systems. While educators are important stakeholders, they are not likely to be as impacted as students by AI-based decisions. A focus on students with disabilities is essential due to the fact that this group is often underrepresented in datasets (Riazy et al., 2020 ). Therefore, like Marsh et al. (2024), we call for greater student agency in AI EdTech. In their study, Marsh & Milne ( 2024 ) found that students were more likely to share disability information with lecturers and peers later in their studies as they developed trusting relationships. Similarly, in our study, some participants wished to control which instructors could access their data because they knew some were more understanding than others. This suggests that data access should be easy to modify over time. However, the necessity of a trusting relationship between students and HEI staff could mean that EWS may not be as effective as intended considering that such tools typically target students who recently started their studies. Student agency could also translate into allowing students to decide on the intervention following a prediction from an EWS. For example, Han et al. ( 2025 , p. 17) encouraged AI EdTech developers to let students “adjust the frequency, tone, and type of feedback” they receive. This recommendation aligns with our findings where some participants preferred human support, while others favored insights into their strengths and weaknesses to reflect on their learning progress. Providing such control over an EWS could also accommodate those who feel anxious about receiving feedback on their performance. Additionally, Marsh et al. (2024, p. 17) called for greater data transparency and argued that “any technology which offers accessibility options, including features such as the ability to change text size, turn on closed captions, enable read-aloud, adjust contrast and similar, should consider the status of those settings to be a potential privacy issue”. The development of accessible and inclusive AI tutors is likely to include such accessibility options. Many users may not realize that this technology implicitly shares data linked to neurodivergence or disability. Similarly, online searches and text input in chatbots become data containing information on disabilities. Some of our study participants were aware that their online searches could influence their social media feed which suggested them content on ADHD. Social media platforms often employ online user behavior and textual data to form profiles intended to provide users with interesting content and persons to connect (Gilbert et al., 2023 ; Ricci et al., 2022 ). Nevertheless, these may also be misused to target people for political purposes as the Cambridge Analytica case revealed when Facebook user data were used to influence US American voters (Cadwalladr & Graham-Harrison, 2018; Gilbert et al., 2023 ). From our study, it appears that some students did not realize that asking ChatGPT about ADHD generate data that could hint to their ADHD status if the information were to be extracted. At the moment and to the best of the authors’ knowledge, ChatGPT interactions are not used to create user profiles and textual data is not turned into meaningful information. Still, to guarantee privacy rights, HEI are encouraged to raise awareness about how students’ online behavior may provide sensitive information. Moreover, HEI need to consider this risk when using student data for research purposes, designing new applications, or acquiring AI-based applications from third parties. 5.2 Collecting ADHD information to design inclusive AI EdTech In general, students were open to sharing their data, even information on ADHD, especially in cases where the goal is to improve the university, as opposed to individual interventions. This is in line with research that found that the general student population trusts their university and is willing to share data for altruistic goals (Jones et al., 2020 ; Y.-S. Tsai et al., 2020 ). However, an experiment with the general student population indicated that 92% of participants were reluctant to share medical information (Ifenthaler & Schumacher, 2016 ). Our study nuances this claim as students were willing to share their ADHD status to help other students and improve the accuracy of AI EdTech. This willingness is particularly evident among those who had disclosed their status to their HEI to receive reasonable accommodation for example. This finding may be due to the fact that these participants faced barriers in their studies and had positive experiences with the adaptations, making them more committed to enhancing inclusion measures in HEI. This result suggests that researchers and developers can collaborate with students with ADHD who are ready to share their lived experience and data to improve accessibility. Furthermore, it implies that optimizing AI tutors for students with ADHD would primarily benefit those with an official diagnosis who feel legitimate or comfortable seeking support. This also raises the question of the roles of different technological actors in creating responsible and inclusive AI. Students did not necessarily expect their universities to provide an all-rounding tool, acknowledging that universities may not have the resources to commit to the provision of such tools. Considering the trust towards HEI, researchers may be well-positioned to explore specific features that could benefit students with ADHD. These functions could then be integrated into existing tools, following a universal design approach. For example, an AI tutor could include features to support time management based on preferences and habits without labeling the option as “ADHD-friendly”. For EWS, researchers could focus on how to present information positively without triggering negative emotions that can particularly affect those prone to anxiety and depression. 6 Conclusion This study investigates ADHD disclosure in AI EdTech with 20 semi-structured interviews in the German-speaking regions of Switzerland. This work aimed to encourage researchers to investigate how to design inclusive AI EdTech, bearing in mind that a universal design approach would ensure that everyone benefits from it. Results indicate that students with ADHD generally perceive AI tutors as more useful than EWS due to the risk of discrimination by AI systems and academic staff. Still, students were more open to the use of EWS if it aimed at structural changes. Moreover, acceptance of EWS could be increased by allowing students to opt in or out of such tools, offering them detailed information over how predictions are formulated, giving them control over which lecturer has access to their data and letting them decide on their preferred intervention type. Participants also showed interest in an AI tutor that could flexibly support them with time management. There is a certain openness among students to disclose their ADHD to AI EdTech, especially from those who have shared their conditions with their HEI. Perceived benefits of sharing this information with EWS include improving prediction accuracy and intervention. For AI tutors, participants saw an opportunity to get personalized resource recommendations and support with challenging tasks such as time management. Reasons not to disclose ADHD in the two uses comprised a perceived lack of relevance to share this information and a bias risk. Guaranteeing student control (e.g. letting them decide flexibly which lecturer gets access to their data) and anonymity is critical to preserve the privacy of students with ADHD. 6.1 Limitations One potential limitation of this study is the possibility of self-selection bias due to the chosen methodology. Although semi-structured interviews enable a deep understanding of participants’ opinions on a topic, they require individuals who feel comfortable discussing the given subject. Despite efforts in our recruitment strategy to minimize this effect, we recognize that the openness towards disclosing ADHD could be more represented than those who prefer not to talk about it. Declarations Competing Interests The authors have no relevant financial or non-financial interests to disclose. Funding This work was financed through two swissuniversities projects: P7 – Accessible teaching in higher education and P8—Swiss digital skills academy: accessible open educational resources. Author Contribution This study is part of Oriane Pierrès’s doctoral thesis under the supervision of Prof. Dr. Alireza Darvishy and PD Dr. Markus Christen. All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Oriane Pierrès. The first draft of the manuscript was written by Oriane Pierrès and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Acknowledgement We thank Albiona Hajdari for her transcription of the interviews and her support in this work as a second coder of the interviews. We would also like to thank the following persons for sharing their expertise: Benjamin Börner, Nico Ebert, Corinna Hertweck, Sebastian Wäscher, and Holger Baumann. We also thank Juliet Manning for proofreading this work. Data Availability To protect study participant privacy, interview transcripts are not made available. Although transcripts were anonymized by deleting information such as name, university, or location, participants shared personal information in the interviews, such as their ADHD status and specific life examples. This could potentially allow reidentification. However, the anonymized data can be made available to reviewers upon request to the corresponding author. References Aboulafia, A. (2024, January 19). Internet privacy is a disability rights issues. Tech Policy . https://www.techpolicy.press/internet-privacy-is-a-disability-rights-issue/ Aboulafia, A., Bogen, M., & Swenor, B. (2024). To Reduce Disability Bias in Technology, Start With Disability Data (p. 25). Center for democracy and technology (CDT). https://cdt.org/wp-content/uploads/2024/07/2024-07-23-Data-Disability-report-final.pdf acatech - Deutsche Akademie der Technikwissenschaften & Joachim Herz Stiftung. (2024). MINT Nachwuchsbarometer 2024 (pp. 1–17) [Application/pdf]. Acatech. https://doi.org/10.48669/ACA_2024-8 American Psychiatric Association (Ed.). (2022). Diagnostic and statistical manual of mental disorders: DSM-5-TR TM (Fifth edition, text revision). American Psychiatric Association Publishing. Baillifard, A., Gabella, M., Lavenex, P. B., & Martarelli, C. S. (2024). Effective learning with a personal AI tutor: A case study. Education and Information Technologies . https://doi.org/10.1007/s10639-024-12888-5 Baker, R. S., & Hawn, A. (2021). Algorithmic Bias in Education. International Journal of Artificial Intelligence in Education . https://doi.org/10.1007/s40593-021-00285-9 Batsaikhan, B. (Zack), & Correia, A.-P. (2024). The effects of Generative Artificial Intelligence on Intelligent Tutoring Systems in higher education: A systematic review. Studies in Technology Enhanced Learning , 4 (1). https://doi.org/10.21428/8c225f6e.33570bb1 Bernstein, A., Cordoba, A., Witt, C., Siegers, D., Wessels, D., Thouvenin, F., Hastings, J., Hatt, J.-M., Buhmann, J., Martinez-Moreno, J., Schawinski, K., Christen, M., Coors, M., Denner, M., Hunger, P., von Thiessen, R., Fabrikant, S., Friemel, T., Neupert, T., & Sachs, U. (2024). AI in Education, Research, and Innovation (pp. 1–11) [Position Paper]. Universität Zürich. https://www.dsi.uzh.ch/dam/jcr:c1b5b525-39af-400f-9215-c29f25bc1c71/dsi-strategy-lab-23-positionspapier-en.pdf Bertilsdotter Rosqvist, H., Hultman, L., Österborg Wiklund, S., Nygren, A., Storm, P., & Sandberg, G. (2023). ADHD in higher education and academia. Canadian Journal on Disability Studies , 12 (3), 1–26. Boisot, M., & Canals, A. (2004). Data, information and knowledge: Have we got it right? Journal of Evolutionary Economics , 14 (1), 43–67. https://doi.org/10.1007/s00191-003-0181-9 Botha, M., Chapman, R., Giwa Onaiwu, M., Kapp, S. K., Stannard Ashley, A., & Walker, N. (2024). The neurodiversity concept was developed collectively: An overdue correction on the origins of neurodiversity theory. Autism , 28 (6), 1591–1594. https://doi.org/10.1177/13623613241237871 Cadwalladr, C., & Graham-Harrison, E. (2018, March 17). Revealed: 50 million Facebook profiles harvested for Cambridge Analytica in major data breach. The Guardian . https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election CAST. (n.d.-a). Universal Design for Learning . Retrieved 24 January 2025, from https://www.cast.org/what-we-do/universal-design-for-learning/ CAST. (n.d.-b). Universal Design for Learning Guidelines Version 2.2 . Retrieved 2 December 2024, from http://udlguidelines.cast.org/ Ciolacu, M., Tehrani, A. F., Binder, L., & Svasta, P. M. (2018). Education 4.0 - Artificial Intelligence Assisted Higher Education: Early recognition System with Machine Learning to support Students’ Success. 2018 IEEE 24th International Symposium for Design and Technology in Electronic Packaging (SIITME) , 23–30. https://doi.org/10.1109/siitme.2018.8599203 Clouder, L., Karakus, M., Cinotti, A., Ferreyra, M. V., Fierros, G. A., & Rojo, P. (2020). Neurodiversity in higher education: A narrative synthesis. Higher Education , 80 (4), 757–778. https://doi.org/10.1007/s10734-020-00513-6 Colombo-Dougovito, A. M., Dillon, S. R., & Mpofu, E. (2020). The Wellbeing of People with Neurodiverse Conditions. In E. Mpofu (Ed.), Sustainable Community Health: Systems and Practices in Diverse Settings (pp. 499–535). Springer International Publishing. https://doi.org/10.1007/978-3-030-59687-3_15 Convention on the Rights of Persons with Disabilities Pledged, Article 24 – Education (2006). https://www.un.org/development/desa/disabilities/convention-on-the-rights-of-persons-with-disabilities.html Culnan, M. J., & Armstrong, P. K. (1999). Information Privacy Concerns, Procedural Fairness, and Impersonal Trust: An Empirical Investigation. Organization Science , 10 (1), 104–115. JSTOR. Deshpande, A., & Sharp, H. (2022). Responsible AI Systems: Who are the Stakeholders? Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society , 227–236. https://doi.org/10.1145/3514094.3534187 European Union. (2016, April 27). REGULATION (EU) 2016/679 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation) (Text with EEA relevance) . https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32016R0679&from=EN#d1e1374-1-1 Flowers, L. (2012). Navigating and accessing higher education: The experiences of community college students with attention deficit hyperactivity disorder [Doctoral dissertation]. https://kortschakcenter.usc.edu/wp-content/uploads/2015/04/Flowers-Community-College-ADHD-Access.pdf Gilbert, J., Hamid, S., Hashem, I. A. T., Ghani, N. A., & Boluwatife, F. F. (2023). The rise of user profiling in social media: Review, challenges and future direction. Social Network Analysis and Mining , 13 (1), 137. https://doi.org/10.1007/s13278-023-01146-0 Grimes, S., Southgate, E., Scevak, J., & Buchanan, R. (2019). University student perspectives on institutional non-disclosure of disability and learning challenges: Reasons for staying invisible. International Journal of Inclusive Education , 23 (6), 639–655. https://doi.org/10.1080/13603116.2018.1442507 Han, B., Nawaz, S., Buchanan, G., & McKay, D. (2025). Students’ Perceptions: Exploring the Interplay of Ethical and Pedagogical Impacts for Adopting AI in Higher Education. International Journal of Artificial Intelligence in Education . https://doi.org/10.1007/s40593-024-00456-4 Hellings, J., & Haelermans, C. (2022). The effect of providing learning analytics on student behaviour and performance in programming: A randomised controlled experiment. Higher Education , 83 (1), 1–18. https://doi.org/10.1007/s10734-020-00560-z Ifenthaler, D., & Schumacher, C. (2016). Student perceptions of privacy principles for learning analytics. Educational Technology Research and Development , 64 (5), 923–938. https://doi.org/10.1007/s11423-016-9477-y Johnson, D. (2019). OCLC Research Works in Progress Webinar: Deakin Genie . https://www.oclc.org/content/dam/research/presentations/2019/Deakin-Genie.pdf Jones, K. M. L., Asher, A., Goben, A., Perry, M. R., Salo, D., Briney, K. A., & Robertshaw, M. B. (2020). “We’re being tracked at all times”: Student perspectives of their privacy in relation to learning analytics in higher education. Journal of the Association for Information Science and Technology , 71 (9), 1044–1059. https://doi.org/10.1002/asi.24358 Khan Academy. (n.d.). Khanmigo by Khan Academy . Retrieved 14 June 2024, from https://www.khanmigo.ai/ Kizilcec, R. F. (2024). To Advance AI Use in Education, Focus on Understanding Educators. International Journal of Artificial Intelligence in Education , 34 (1), 12–19. https://doi.org/10.1007/s40593-023-00351-4 Labadze, L., Grigolia, M., & Machaidze, L. (2023). Role of AI chatbots in education: Systematic literature review. International Journal of Educational Technology in Higher Education , 20 (1), 56. https://doi.org/10.1186/s41239-023-00426-1 Laufer, R. S., & Wolfe, M. (1977). Privacy as a Concept and a Social Issue: A Multidimensional Developmental Theory. Journal of Social Issues , 33 (3), 22–42. https://doi.org/10.1111/j.1540-4560.1977.tb01880.x Li, W., Sun, K., Schaub, F., & Brooks, C. (2022). Disparities in Students’ Propensity to Consent to Learning Analytics. International Journal of Artificial Intelligence in Education , 32 (3), 564–608. https://doi.org/10.1007/s40593-021-00254-2 Marsh, A., & Milne, L. R. (2024). I Don’t Want to Sound Rude, but It’s None of Their Business: Exploring Security and Privacy Concerns around Assistive Technology Use in Educational Settings. ACM Trans. Access. Comput. , 17 (2). https://doi.org/10.1145/3670690 Mayring, P. (2014). Qualitative content analysis: Theoretical foundation, basic procedures and software solution . https://www.ssoar.info/ssoar/handle/document/39517 McMullin, C. (2023). Transcription and Qualitative Methods: Implications for Third Sector Research. VOLUNTAS: International Journal of Voluntary and Nonprofit Organizations , 34 (1), 140–153. https://doi.org/10.1007/s11266-021-00400-3 Monllaó Olivé, D., Huynh, D. Q., Reynolds, M., Dougiamas, M., & Wiese, D. (2020). A supervised learning framework: Using assessment to identify students at risk of dropping out of a MOOC. Journal of Computing in Higher Education , 32 (1), 9–26. https://doi.org/10.1007/s12528-019-09230-1 Moriña, A. (2024). When what is unseen does not exist: Disclosure, barriers and supports for students with invisible disabilities in higher education. Disability & Society , 39 (4), 914–932. https://doi.org/10.1080/09687599.2022.2113038 Morris, M. R. (2020). AI and Accessibility. Commun. ACM , 63 (6), 35–37. https://doi.org/10.1145/3356727 Mullins, L., & Preyde, M. (2013). The lived experience of students with an invisible disability at a Canadian university. Disability & Society , 28 (2), 147–160. https://doi.org/10.1080/09687599.2012.752127 Newman, L. A., & Madaus, J. W. (2015). Reported Accommodations and Supports Provided to Secondary and Postsecondary Students With Disabilities: National Perspective. Career Development and Transition for Exceptional Individuals , 38 (3), 173–181. https://doi.org/10.1177/2165143413518235 Ortigosa, A., Carro, R. M., Bravo-Agapito, J., Lizcano, D., Alcolea, J. J., & Blanco, O. (2019). From Lab to Production: Lessons Learnt and Real-Life Challenges of an Early Student-Dropout Prevention System. IEEE Transactions on Learning Technologies , 12 (2), 264–277. https://doi.org/10.1109/TLT.2019.2911608 Osborne, T. (2019). Not lazy, not faking: Teaching and learning experiences of university students with disabilities. Disability & Society , 34 (2), 228–252. https://doi.org/10.1080/09687599.2018.1515724 Pierrès, O., Christen, M., Schmitt-Koopmann, F. M., & Darvishy, A. (2024). Could the Use of AI in Higher Education Hinder Students With Disabilities? A Scoping Review. IEEE Access , 12 , 27810–27828. https://doi.org/10.1109/ACCESS.2024.3365368 Pierrès, O., Darvishy, A., & Christen, M. (2024). Exploring the role of generative AI in higher education: Semi-structured interviews with students with disabilities. Education and Information Technologies . https://doi.org/10.1007/s10639-024-13134-8 Riazy, S., Simbeck, K., & Schreck, V. (2020). Fairness in Learning Analytics: Student At-risk Prediction in Virtual Learning Environments: Proceedings of the 12th International Conference on Computer Supported Education , 15–25. https://doi.org/10.5220/0009324100150025 Ricci, F., Rokach, L., & Shapira, B. (2022). Recommender Systems: Techniques, Applications, and Challenges. In F. Ricci, L. Rokach, & B. Shapira (Eds.), Recommender Systems Handbook (pp. 1–35). Springer US. https://doi.org/10.1007/978-1-0716-2197-4_1 Roberts, J. B., Crittenden, L. A., & Crittenden, J. C. (2011). Students with disabilities and online learning: A cross-institutional study of perceived satisfaction with accessibility compliance and services. The Internet and Higher Education , 14 (4), 242–250. https://doi.org/10.1016/j.iheduc.2011.05.004 Römhild, A., & Hollederer, A. (2024). Effects of disability-related services, accommodations, and integration on academic success of students with disabilities in higher education. A scoping review. European Journal of Special Needs Education , 39 (1), 143–166. https://doi.org/10.1080/08856257.2023.2195074 Sajja, R., Sermet, Y., Cikmaz, M., Cwiertny, D., & Demir, I. (2024). Artificial Intelligence-Enabled Intelligent Assistant for Personalized and Adaptive Learning in Higher Education. Information , 15 (10). https://doi.org/10.3390/info15100596 Schippers, L. M., Horstman, L. I., Velde, H. V. D., Pereira, R. R., Zinkstok, J., Mostert, J. C., Greven, C. U., & Hoogman, M. (2022). A qualitative and quantitative study of self-reported positive characteristics of individuals with ADHD. Frontiers in Psychiatry , 13 , 922788. https://doi.org/10.3389/fpsyt.2022.922788 Sghir, N., Adadi, A., & Lahmer, M. (2023). Recent advances in Predictive Learning Analytics: A decade systematic review (2012–2022). Education and Information Technologies , 28 (7), 8299–8333. https://doi.org/10.1007/s10639-022-11536-0 Smith, S. A., Woodhead, E., & Chin-Newman, C. (2021). Disclosing accommodation needs: Exploring experiences of higher education students with disabilities. International Journal of Inclusive Education , 25 (12), 1358–1374. https://doi.org/10.1080/13603116.2019.1610087 Soffer, T., & Cohen, A. (2024). Privacy versus pedagogy–students’ perceptions of using learning analytics in higher education. Australasian Journal of Educational Technology . Spiel, K., Hornecker, E., Williams, R. M., & Good, J. (2022). ADHD and Technology Research – Investigated by Neurodivergent Readers. Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems . https://doi.org/10.1145/3491102.3517592 Tsai, S., Chen, C., Shiao, Y., Ciou, J., & Wu, T. (2020). Precision education with statistical learning and deep learning: A case study in Taiwan. INTERNATIONAL JOURNAL OF EDUCATIONAL TECHNOLOGY IN HIGHER EDUCATION , 17 (1). https://doi.org/10.1186/s41239-020-00186-2 Tsai, S.-C., Chen, C.-H., Shiao, Y.-T., Ciou, J.-S., & Wu, T.-N. (2020). Precision education with statistical learning and deep learning: A case study in Taiwan. International Journal of Educational Technology in Higher Education , 17 (1). https://doi.org/10.1186/s41239-020-00186-2 Tsai, Y.-S., Whitelock-Wainwright, A., & Gašević, D. (2020). The privacy paradox and its implications for learning analytics. Proceedings of the Tenth International Conference on Learning Analytics & Knowledge , 230–239. https://doi.org/10.1145/3375462.3375536 World Health Organisation. (2021, November 4). Disability and Health . https://www.who.int/news-room/fact-sheets/detail/disability-and-health#:~:text=Disability%20refers%20to%20the%20interaction,%2C%20and%20limited%20social%20supports). Zawacki-Richter, O., Marín, V. I., Bond, M., & Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education – where are the educators? International Journal of Educational Technology in Higher Education , 16 (1). Scopus. https://doi.org/10.1186/s41239-019-0171-0 Additional Declarations No competing interests reported. Supplementary Files researchprotocolwcodingtabledisclosingADHDAIeducationaltechnologiesFINAL.docx DEInterviewsleitfadenDiscloseDisabilityv7final.docx Cite Share Download PDF Status: Posted Version 1 posted 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-6106311","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":422240673,"identity":"b413c493-5757-426f-bbfc-e743833c8c39","order_by":0,"name":"Oriane Pierrès","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABDklEQVRIiWNgGAWjYFACxgYwxQbh2fAYQBgHGBgkiNOSRowWVHCYgaAWfunDjZ8LKhjy+Nh7n3348Oe8jLnY4QPMPH/uMPDPbsCqRbIvsVl6xhmGYjae48YzZ7bd5rGcnZbAzMPzjEHizgGsWgzOMDZI87YxJLZJpDEz8zbc5jG4nWPAzCMBdKFEAi4tzb95/wG1yD9jZv7z5xxQS/4HZh4DvFrapHkbQLawMTMzsB0A2cLAzJOAW4tkD2ObNc8xCaBf0pgZe9uSQX4xODjnwGEeiRvYtfDzsD++zVNjkyfffoyZ4ccfO3tz6eSHD978OSzHPwO7FihAc8MBIObBpx4E8Bo4CkbBKBgFIxwAABOSVEBuVZeCAAAAAElFTkSuQmCC","orcid":"","institution":"University of Zurich (UZH)","correspondingAuthor":true,"prefix":"","firstName":"Oriane","middleName":"","lastName":"Pierrès","suffix":""},{"id":422240674,"identity":"8a7a566f-c411-48cb-bf0c-a49bc9e649ff","order_by":1,"name":"Alireza Darvishy","email":"","orcid":"","institution":"Zurich University for Applied Sciences (ZHAW)","correspondingAuthor":false,"prefix":"","firstName":"Alireza","middleName":"","lastName":"Darvishy","suffix":""},{"id":422240675,"identity":"88b848a0-a5a5-4585-a358-eb5baf289ee1","order_by":2,"name":"Markus Christen","email":"","orcid":"","institution":"University of Zurich (UZH)","correspondingAuthor":false,"prefix":"","firstName":"Markus","middleName":"","lastName":"Christen","suffix":""}],"badges":[],"createdAt":"2025-02-25 14:53:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6106311/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6106311/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":77684353,"identity":"c2a9f85b-850e-46b0-9739-9cca5f12a61c","added_by":"auto","created_at":"2025-03-04 08:58:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":41206,"visible":true,"origin":"","legend":"\u003cp\u003eWillingness to share ADHD in an EWS and an AI tutor among students who disclosed their ADHD at their HEI and those who did not\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6106311/v1/926946fbf4ea71c3d80ccaf8.png"},{"id":77684358,"identity":"4975a6aa-4af6-4842-abbf-0554c021d015","added_by":"auto","created_at":"2025-03-04 08:58:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":47362,"visible":true,"origin":"","legend":"\u003cp\u003eDistributions of answers on who should have access to data in an EWS (left graph) and an AI tutor (right graph) according to 20 participants. During the interview,\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6106311/v1/e7178ad13e0cc44b0ef58def.png"},{"id":81648139,"identity":"e9e91465-2c60-48f7-80f1-78905eb2c9e9","added_by":"auto","created_at":"2025-04-29 15:16:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":969934,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6106311/v1/48daf256-6d86-470c-80b3-fc8ce3070fbb.pdf"},{"id":77684359,"identity":"fbe5428b-2c58-4814-9fd6-0a30f5000c27","added_by":"auto","created_at":"2025-03-04 08:58:05","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":80827,"visible":true,"origin":"","legend":"","description":"","filename":"researchprotocolwcodingtabledisclosingADHDAIeducationaltechnologiesFINAL.docx","url":"https://assets-eu.researchsquare.com/files/rs-6106311/v1/a187ec32ffdf40d1f0990f8e.docx"},{"id":77686005,"identity":"407106a2-dba6-4bc7-aa48-6e06a8976566","added_by":"auto","created_at":"2025-03-04 09:06:05","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":34331,"visible":true,"origin":"","legend":"","description":"","filename":"DEInterviewsleitfadenDiscloseDisabilityv7final.docx","url":"https://assets-eu.researchsquare.com/files/rs-6106311/v1/a81d02630bb4b43198462a9e.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Perceived Risks and Benefits of Disclosing ADHD to AI-based Educational Technologies: Semi-structured Interviews","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eStudents with disabilities and chronic conditions are expected to disclose their disabilities to their higher education institutions (HEI) if they want to receive reasonable accommodations such as additional time for exams, remediation of inaccessible learning materials, or sign language interpreters, to name a few. However, many students decide not to inform their institutions about their disabilities\u0026nbsp;(Grimes et al., 2019; Newman \u0026amp; Madaus, 2015; Roberts et al., 2011). The disclosure process can be particularly difficult for neurodivergent students (Bertilsdotter Rosqvist et al., 2023; Mori\u0026ntilde;a, 2024; Smith et al., 2021). Clouder et al. (2020) define neurodiversity as \u0026ldquo;an umbrella term, including dyspraxia, dyslexia, attention deficit, hyperactivity disorder, dyscalculia, autistic spectrum and Tourette Syndrome\u0026rdquo; (p. 757). The term was developed within the autistic activist communities (Botha et al., 2024). It recognizes that \u0026ldquo;differences in individual brain function and behavioral traits [are a] normal variation in the population\u0026rdquo;\u0026nbsp;(Clouder et al., 2020, p. 758). In this paper, \u0026ldquo;neurodivergence\u0026rdquo; is the preferred expression, although \u0026ldquo;disability\u0026rdquo; is also used when referring to the field of disability research. Disability is understood from a socio-medical perspective, acknowledging that it results from both personal and environmental factors\u0026nbsp;(World Health Organisation, 2021).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs neurodiversity has no physical traits, students must decide whether to disclose their neurodivergence. Disclosure can be emotionally challenging due to past discrimination, a lack of understanding by academic staff, administrative burdens, or a desire to study \u0026quot;normally\u0026quot; without identifying as disabled (Mori\u0026ntilde;a, 2024). As a result, many researchers advocate for the adoption of the universal design for learning (UDL) framework in higher education, because it reduces the need for disclosure (Mullins \u0026amp; Preyde, 2013; Osborne, 2019). UDL is a framework that acknowledges all students learn differently, emphasizing that failure to learn often lies in the environment (CAST, n.d.-a). It calls for flexible environments that accommodate individual needs by clarifying the purpose of learning, presenting information in diverse ways, and allowing learners to demonstrate their competencies and knowledge through various formats (CAST, n.d.-b). With UDL, learning environments may become inclusive for students with disabilities without requiring them to advocate for special adaptations to study (Osborne, 2019). Still, receiving reasonable accommodations is a student\u0026rsquo;s right according to the United Nations Convention on the Rights of People with Disabilities (Convention on the Rights of Persons with Disabilities Pledged, Article 24 \u0026ndash; Education, 2006).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWhen developing artificial intelligence-based education technologies (AI EdTech), developers and educators must weigh in on how much information they collect from their students. In higher education, AI usually refers to systems using machine learning techniques that employ data to make predictions or solve problems (Baker \u0026amp; Hawn, 2021). For instance, many researchers are trying to leverage big data to identify early students at risk of failing or dropping out in order to provide them with timely support (see e.g.\u0026nbsp;Ciolacu et al. (2018); Ortigosa et al. (2019); S. Tsai et al. (2020)).\u0026nbsp;Research has also been focusing on how to use newer Large Language Models (LLMs) to develop specialized learning assistants\u0026nbsp;(Labadze et al., 2023). Collecting data on neurodiversity in AI EdTech could have several benefits. Firstly, these technologies are often trained on datasets in which disability is underrepresented which can lead to bias and errors for this group\u0026nbsp;(Riazy et al., 2020). With this information, researchers can audit AI systems and ensure that their outcomes do not discriminate against neurodivergent students\u0026nbsp;(Aboulafia et al., 2024). Secondly, integrating diverse learning needs into the design of AI EdTech could help personalize higher education. For example, current recommendation systems suggesting courses or learning materials rarely \u0026nbsp;include accessibility factors in their algorithms\u0026nbsp;(Pierr\u0026egrave;s, Christen, et al., 2024). Course recommendations could consider teaching practices that are known to support neurodivergent students, such as interactive and discussion-based course formats for students with ADHD\u0026nbsp;(Flowers, 2012). \u0026nbsp;Third, systems could adapt to students\u0026apos; needs without requiring human intervention. Many neurodivergent students choose not to disclose their disabilities due to the associated stigma and lack of awareness among academic staff\u0026nbsp;(Mori\u0026ntilde;a, 2024). For example, students have reported instances where professors would assume that they are faking their conditions or using it as an excuse for laziness\u0026nbsp;(Osborne, 2019). As a result, an adaptive platform guaranteeing anonymity could appeal to some.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThere are, however, several concerns regarding the collection of disability data in AI EdTech. Legally, disability status is classified as health-related data, which the European Union considers sensitive (European Union, 2016). Due to its nature, this data should be processed with greater restraint and enhanced privacy protection. Integrating personal information on disabilities, chronic conditions, or neurodivergence into AI EdTech can compromise data anonymization, because certain conditions are rare, making re-identification easier (Morris, 2020). Additionally, persons with disabilities may be deprived of their choice to disclose or not. This can happen when a service \u0026nbsp;would be inaccessible without disclosure (Aboulafia, 2024) or when the system can deduce a user\u0026rsquo;s disability from their data (Morris, 2020) or from their use of assistive technologies (Aboulafia, 2024; Marsh \u0026amp; Milne, 2024).\u003c/p\u003e\n\u003cp\u003eIn brief, there is a trade-off between disclosing neurodivergent status in AI EdTech to foster inclusion and protecting sensitive data. Investigating students\u0026rsquo; opinions could enable the identification of conditions when disclosing one\u0026rsquo;s neurodiversity could be acceptable. These conditions can inform developers and higher education staff to understand how to provide personalized technologies while preserving privacy. This study focuses on the perspective of students with ADHD, as the term \u0026ldquo;neurodivergent\u0026rdquo; is very broad. Differences in behavior and opinions could diverge due to the diversity of the conditions.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe overarching research question is: How do students assess the risk-benefits of disclosing their ADHD in the use of AI-based educational technologies? To answer this question, sub-questions were formulated. The first sub-question relates to the general perception of students with ADHD on the utility of two cases of AI EdTech, early warning systems and AI tutors. This overall opinion is relevant to ask because it may impact students\u0026rsquo; readiness to disclose sensitive data. As a result, the following sub-question is raised: \u0026nbsp;\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003eHow do students with ADHD perceive the utility of AI EdTech?\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eThree sub-questions are specifically connected to the disclosure of ADHD status in AI EdTech, asking for perceived risks and benefits as well as conditions under which it is acceptable to disclose:\u0026nbsp;\u003c/p\u003e\n\u003col start=\"2\"\u003e\n \u003cli\u003eWhat are the students\u0026rsquo; perceived risks of disclosing their ADHD to AI EdTech?\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eWhat are the students\u0026rsquo; perceived benefits of disclosing their ADHD to AI EdTech?\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eUnder which conditions are students willing to disclose their ADHD?\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"2\tTheoretical Background ","content":"\u003ch3\u003e2.1\u0026nbsp;\u0026nbsp;Students with ADHD in Higher Education\u003c/h3\u003e\n\u003cp\u003eFrom a medical perspective, ADHD is a neurodevelopmental condition that is usually diagnosed based on the identification of symptoms outlined in international psychiatric standards such as the Diagnostic and Statistical Manual of Mental Disorders 5\u003csup\u003eth\u003c/sup\u003e Edition text revision (DSM-5-TR). These symptoms relate to inattention, hyperactivity, \u0026nbsp;and impulsivity (American Psychiatric Association, 2022). \u0026nbsp;From a neurodiverse perspective, ADHD is a variation of the brain with its own strengths and weaknesses (Colombo-Dougovito et al., 2020). The medical perspective often overlooks the positive aspects of ADHD as the diagnosis is based on individuals’ deficits. Yet, in a study, Dutch ADHD adults self-reported the following positive characteristics: creativity, being dynamic, flexibility, socio-affective skills, and higher-order cognitive skills (Schippers et al., 2022). While it is not clear whether these come from living with ADHD or coping mechanisms (Schippers et al., 2022), they highlight that ADHD cannot be reduced to a disorder.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDisability research has sought to understand why some students with non-visible disabilities choose not to disclose their conditions and the consequences of not receiving adequate support (Clouder et al., 2020; Moriña, 2024). A literature review notably highlighted the tension between the universities’ requirement to disclose neurodivergence to request support and the uneven awareness of neurodiversity among academic staff (Clouder et al., 2020). At times, lecturers treat neurodivergent students poorly or in a discriminatory manner, do not provide support, and lack flexibility (Clouder et al., 2020). Consequently, students with disabilities engage in a rational complex reflection on whether to disclose their disabilities\u0026nbsp;(Grimes et al., 2019). On the one hand, non-disclosure is a personal choice where students wish to study “normally” or do not view themselves as having a disability\u0026nbsp;(Moriña, 2024). Still, a recurring issue is that learners want to avoid the associated stigma\u0026nbsp;(Clouder et al., 2020; Moriña, 2024). For example, students reported situations when academic staff and peers treated them as lesser persons who did not belong to an HEI\u0026nbsp;(Grimes et al., 2019). Additionally, because their conditions are not visibly apparent, they may face situations where others question the legitimacy of their experiences or diagnoses\u0026nbsp;(Moriña, 2024; Osborne, 2019). Unsurprisingly, some students may feel more comfortable discussing their struggles with chatbots who would not judge them\u0026nbsp;(Pierrès, Darvishy, et al., 2024).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOn the other hand, universities expect disclosure to grant reasonable accommodations that can eliminate barriers to study. For example, some students get a note-taker to let them concentrate on the course content. Others can also get additional time during an exam and / or can write an exam in a separate room to avoid distractions. However, the effectiveness of these accommodations is inconclusive for students with ADHD (Römhild \u0026amp; Hollederer, 2024). An explanation for this could be that the interventions are ill-fitted to students’ needs (Römhild \u0026amp; Hollederer, 2024). It could also be that causal effects are difficult to quantify as qualitative studies indicate positive effects of disability-related services on student success (Römhild \u0026amp; Hollederer, 2024). According to the literature review by Moriña (2024), not disclosing a disability can result in students feeling like they cannot be themselves or achieve as much as their peers. It can also affect their mental health negatively \u0026nbsp;(Moriña, 2024). Consequently, health promotion interventions, coaching, as well as social and academic integration can become instrumental in ensuring the success of neurodivergent students (Clouder et al., 2020; Römhild \u0026amp; Hollederer, 2024). Clouder et al. (2020) \u0026nbsp;also highlighted that adopting a universal design approach could reduce the need to disclose to receive specific adaptations. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn brief, students are likely to weigh on whether disclosing their ADHD will benefit them. Sharing this information with an AI EdTech may be appealing as it could avoid stigma while providing a certain flexibility and personalized support.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e2.2\u0026nbsp;\u0026nbsp;Data privacy and AI EdTech\u003c/h3\u003e\n\u003cp\u003eAfter a review of the literature and to the best of the authors’ knowledge, there is no research focusing on how students with ADHD deal with data in higher education technologies. One study explored privacy and security concerns of students with disabilities in their use of assistive technologies, but it included only one neurodivergent individual (Marsh \u0026amp; Milne, 2024). Within the field of human-computer interaction, research on ADHD and technology often focuses on children and rarely investigates what people want or feel (Spiel et al., 2022). Understanding what students with ADHD in higher education want in AI could ensure that the technologies fulfill their needs.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eResearch on data privacy in higher education was often conducted with the general population of students. This body of knowledge can inform how neurodivergent students think about data sharing. For these studies, researchers often refer to the privacy calculus theory, which explains the choice of disclosing personal information as a rational process that seeks to maximize benefits while avoiding \u0026nbsp; negative consequences (Laufer \u0026amp; Wolfe, 1977). Culnan and Armstrong (1999, p. 106) explained that “in general, individuals are less likely to perceive information collection as privacy-invasive when a) information is collected in the context of an existing relationship, b) they perceive that they have the ability to control future use of the information, c) the information collected or used is relevant to the transaction, and d) they believe the information will be used to draw reliable and valid inferences about themselves.”\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWith this theory, studies have looked for factors and reasons that explain data sharing in HEI. They have often found that students trusted their HEI to handle their data, but they were not aware of data policies (Jones et al., 2020; Soffer \u0026amp; Cohen, 2024; Y.-S. Tsai et al., 2020). However, Jones et al. (2020) argued that this did not mean that students did not care about data management and control. Students have expectations about the reasons for sharing data, the type of data, and who has access to the data. We know that acceptable reasons to share data are for education purposes (e.g. improving learning experience) and altruistic goals (e.g. improving a study program) (Jones et al., 2020; Y.-S. Tsai et al., 2020). \u0026nbsp;Students are more reluctant to share personal data such as demographics (e.g. gender), personal academic records (e.g. grades), or online activities (e.g. time spent on a learning management platform) than pedagogical data (e.g. feedback on an assignment) \u0026nbsp;(Soffer \u0026amp; Cohen, 2024; Y.-S. Tsai et al., 2020).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe process of disclosing a disability in an AI EdTech could follow a sort of specific privacy calculus theory. Much like students are willing to share data for education purposes and altruistic goals, learners with disabilities are also willing to share information on their disabilities in assistive technologies if it could improve their or their peers’ access to learning (Marsh \u0026amp; Milne, 2024). A particularity that arises when technologies have assistive functions that the accessibility to a service (e.g. a course) is that students may not have the luxury to refrain from using a tool when privacy conditions are deemed unacceptable (Aboulafia, 2024; Marsh \u0026amp; Milne, 2024).\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e2.3\u0026nbsp;\u0026nbsp;Early warning systems and AI tutors\u003c/h3\u003e\n\u003cp\u003eThis work focuses on two types of AI EdTech due to their prevalence in the field: predictive learning analytics to detect at-risk or drop-out students called early warning systems (EWS), and intelligent tutoring systems (ITS) called AI tutors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePredictive learning analytics in education is a well-researched area. Researchers have used machine-learning models to predict students’ academic performance, risk of failing or dropping out, enrollment chances, engagement, and satisfaction (Sghir et al., 2023). This study focuses on the predictions of students’ risk of failing or dropping out as it is one of the most researched area in the field\u0026nbsp;(Sghir et al., 2023; Zawacki-Richter et al., 2019). These predictions are used to create EWS, whose goal is to identify early students who may be struggling and provide them with adequate support (e.g, through email intervention, counselling)\u0026nbsp;(Pierrès, Christen, et al., 2024). For example, German universities are considering using EWS to address the decline in youth interest in the fields of science, technology, engineering, and mathematics (STEM)\u0026nbsp;(acatech - Deutsche Akademie der Technikwissenschaften \u0026amp; Joachim Herz Stiftung, 2024). To form predictions, researchers often use students’ online behavioral data (e.g. clicks on learning activities, time spent on a learning platform), prior and current academic data (e.g. grades, past courses), demographic and socio-economic information, and sometimes psychological features \u0026nbsp;(Sghir et al., 2023). Although EWS follow a well-meaning goal, the reliance on online behavioral data such as the number of clicks or the time spent on an online learning platform as well as socio-economic information risk discriminate against students with disabilities\u0026nbsp;(Pierrès, Christen, et al., 2024). Due to this risk, it is interesting to gather the opinion of students with ADHD on the use of EWS in higher education.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAdditionally, the case of EWS is worth discussing with students as the interventions following the predictions may involve lecturers, academic staff, and students differently. Lecturers are often the main decision-makers\u0026nbsp;(Pierrès, Christen, et al., 2024). For example, lecturers are encouraged to contact via email students identified as at-risk (see e.g.\u0026nbsp;Ciolacu et al. (2018) and Monllaó Olivé et al. (2020)). In other cases, counselling advisors are mainly intervening (see S.-C. Tsai et al. (2020)). Students are rarely involved in the decision-making process of EWS\u0026nbsp;(Pierrès, Christen, et al., 2024). Hellings and Haelermans\u0026nbsp;(2022)\u0026nbsp;presented one of rare cases where students could monitor their predicted grade on a dashboard without the involvement of any other academic staff. \u0026nbsp;Depending on the degree of involvement of the various stakeholders, students may be more or less ready to accept EWS and disclose their ADHD status.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAlong predictive learning analytics, researchers have sought to personalize learning with the development of ITS, i.e. applications that seek to “simulate one-to-one personal tutoring”\u0026nbsp;(Zawacki-Richter et al., 2019, p. 4). The release of ChatGPT by OpenAI in November 2022 has spurred greater interest in ITS, notably due to its progress in providing feedback in the form of a dialogue (Batsaikhan \u0026amp; Correia, 2024). Those systems do not seek to eliminate human tutors, but rather complement them by providing always-available support independent of time and location (Batsaikhan \u0026amp; Correia, 2024). Educational platforms such as Khan Academy and universities are developing AI agents, called AI tutor, AI buddy or virtual assistant, that aim at assisting students along their studies (Baillifard et al., 2024; Bernstein et al., 2024; Johnson, 2019; Khan Academy, n.d.; Sajja et al., 2024). Possible functions include learning support, time management and administrative study organization, learning materials and course recommendations, and networking (see for instance Bernstein et al. (2024), Johnson (2019), and Sajja et al. (2024)). Additionally, the way students use ChatGPT could guide the design of universities’ AI tutors. For example, students with disabilities use ChatGPT to help them with different tasks such as studying (e.g. providing explanations, clarifying instructions, preparing for exams), writing, reading and research assistant, and self-organization\u0026nbsp;(Pierrès, Darvishy, et al., 2024). The future of AI tutors is therefore likely to be unified applications capable of assisting students with various aspects of their studies instead of using multiple tools for specific tasks such as Grammarly for text editing, Co-pilot for programming, and calendar apps to manage meetings.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDepending on its functionalities, AI tutors use different data types. Very often, they collect textual data and employ natural language processing techniques to analyze them (Pierrès, Darvishy, et al., 2024). Textual data can include students’ input in a chat interface, learning materials or even automatically transcribed course recording. For example, Sajja et al. (2024) assessed students’ chat request to detect their emotional state and make the AI tutor respond empathetically if necessary. With textual data, there is a critical distinction between what constitutes data and information. Data is a raw piece of text that does not necessarily contain information that can be used (Boisot \u0026amp; Canals, 2004). For instance, a student could describe symptoms of ADHD and mention the condition in their input, but without explicit extraction of this information an AI tutor will not necessarily categorize this student as having ADHD and adapt their recommendations and answers. In comparison, if the AI tutor was specifically designed to detect and analyze mentions of ADHD in a text, it could turn textual data into textual information. Information, although often used as a synonym for data, is structured data that can enhance an agent’s understanding (Boisot \u0026amp; Canals, 2004). AI tutors can for instance collect information on academic schedules, the courses chosen by a student to help them organize their time. Another piece of information is socio-demographic information. It could be imagined that students can inform directly whether they have ADHD or not in the hope that recommendations and interactions will be more ADHD-friendly. For that reason, it is interesting to ask for students with ADHD how confident they would be to disclose this information.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eSemi-structured interviews were conducted to investigate whether and how students are willing to disclose their ADHD in AI EdTech. This qualitative approach enables the identification of factors that could not have been uncovered from previous research on privacy and educational technologies focusing on the general population of students. Additionally, participants filled out a 5-minute online survey before starting the interviews. This questionnaire consisted of closed questions regarding demographics, information on their ADHD diagnosis, and their estimated knowledge of AI. This information was mainly used to describe the sample of participants and to check for patterns in the response of subgroups of participants.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Interview guide\u003c/h2\u003e \u003cp\u003eAn interview guide was developed in German based on the existing literature; it is available as supplementary material. The interview starts with open questions regarding general experience and opinions on ADHD disclosure in higher education. Then, two cases of AI EdTech are presented to the participants: 1) EWS identifying students who could require support (e.g. additional exercises or explanation from an instructor), and 2) AI tutors. After the two cases were presented and participants\u0026rsquo; possible clarification questions were answered, interviewees were asked about their opinion on the use cases and whether and how they would disclose their ADHD. The interviews concluded with questions reflecting on the two use cases.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Sampling\u003c/h2\u003e \u003cp\u003eStudents with ADHD enrolled in Swiss-German HEI were recruited via e-mail. To ensure the survey reached students who had not disclosed their ADHD to the university, information was disseminated through two universities\u0026rsquo; survey channels for the general population of students, LinkedIn, and three universities\u0026rsquo; \u0026ldquo;marketplace\u0026rdquo; where people can post job adverts and the like. We also sent an email to students previously involved in the authors\u0026rsquo; research studies. One university's disability-related services also informed their students about the study. Additionally, we used the online survey data to balance the number of students who disclosed their ADHD status at their HEI and those who did not as well as their gender.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Data collection\u003c/h2\u003e \u003cp\u003eBefore filling out the short online survey, students had to read and accept a consent form providing information on the objectives of the study, a general overview of the questions, data usage, and storage. Participants were invited to ask questions if they had any. Then, before the interviews, participants were reminded about study participation conditions, in particular, that the exchange would be recorded and transcribed automatically with Microsoft Teams. The transcripts were corrected by the first author and a research assistant following an intelligent verbatim style, i.e., the corrector deleted repetitions or verbal fillers such as \u0026ldquo;umm\u0026rdquo; (McMullin, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) because only the content of the answers was analyzed. During the interviews, the first author kept a diary of notes and reflections occurring during the phase.\u003c/p\u003e \u003cp\u003eThe 20 interviews were conducted in July 2024.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Data analysis\u003c/h2\u003e \u003cp\u003eTwo researchers conducted a content analysis of the interview transcripts, following Mayring\u0026rsquo;s method (2014) on deductive and inductive category assignment. First, the two coders worked independently using an initial coding table based on existing literature and aligned with the interview questionnaire structure. New codes were also created inductively. To ensure consistency in coding with the initial table, the two coders met after analyzing two transcripts to discuss their approaches before proceeding with further independent coding. After coding all transcripts independently, the researchers fully reviewed and discussed three other transcripts to harmonize their codes. The discussions revealed only unsubstantial differences in their coding decisions. Subsequently, the first coder compared all the codes and restructured them into bigger categories. The second coder then reanalyzed the transcripts using this revised code structure. When both coders had analyzed all transcripts, the two researchers discussed any remaining discrepancies and finalized the codes. Finally, the transcript codes that were exclusive were combined with the survey data and analyzed visually with R to identify patterns.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.5 Ethics committee approval\u003c/h2\u003e \u003cp\u003eDue to the involvement of human participants and the necessity to record the interviews for analysis, the study design was presented to and approved by the ethical committee of the authors\u0026rsquo; university.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Sample description\u003c/h2\u003e \u003cp\u003eThe sample was composed of 10 female and 10 male participants. Participants were aged between 22 and 50 years old, with a median of 26. One person did not disclose their birth year. Six interviewees indicated having a migration background.\u003c/p\u003e \u003cp\u003eMost students (17) were enrolled in a bachelor's program, two in a master's, and one in a PhD. A majority of them (15) studied social and business sciences, two natural sciences, and one engineering. Two other students were enrolled in interdisciplinary programs, one combining computer science and linguistics, and the other medicine and social sciences. Most participants (11) assessed their knowledge of AI at the user level. Six indicated that they were interested in the subject but had no technical skills and three had already coded at least one small AI model.\u003c/p\u003e \u003cp\u003eA large majority of the interviewees (13) found out they had ADHD after their 18th birthday, i.e. during adulthood. All but one had a formal medical diagnosis. Half of the sample received reasonable accommodations whereas the other half did not. Eight identified as a person with disabilities.\u003c/p\u003e \u003cp\u003e For most participants (18), the ease of talking about ADHD depended on the context or the person they were conversing with. Very often, they would explain that some people are open and interested while others would simply not understand. As a result, eleven students mentioned situations when they find it easy to talk about ADHD such as with friends, with fellow students with ADHD, or in social or psychological studies. Fourteen participants reported not disclosing their ADHD in some situations, often to avoid negative reactions or because they did not see any benefit in sharing that information. Two also explained that they preferred to adapt and tried to solve issues by themselves. Overall, past experiences with HEI staff and other students were positive.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Perceived utility of two cases of AI EdTech\u003c/h2\u003e \u003cp\u003eA majority of participants preferred the use of AI tutors over EWS. While students saw the support possibilities with AI tutors, they expressed more concerns (e.g. discrimination) with the adoption of EWS in HEI.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section3\"\u003e \u003ch2\u003e4.2.1 Use case: Early warning systems (EWS)\u003c/h2\u003e \u003cp\u003eIn the interviews, participants were invited to imagine an EWS that could provide early support to students predicted to require help. Three scenarios were presented:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eIntervention without humans\u003c/b\u003e: an automatic system contacts students via e-mail directly to inform them that they were identified as potentially requiring support. The e-mail recommends contacting a relevant person (e.g. lecturer) or reading additional learning materials.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eIntervention with lecturers\u003c/b\u003e: the list of students requiring help is sent to lecturers who then proactively seek to support them by offering their help or starting a discussion.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eStructural intervention for future students\u003c/b\u003e: the information is used at the group level to change the university structurally (e.g. curricula, provision of support offers targeting specific groups).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eIn general, participants saw the added value of offering proactive support to students. However, an often-recurring concern was that such a system could be discriminatory, especially in the case of interventions one and two. In total, 12 students explained that predictions could be inaccurate. They reflected on the fact that previous grades in secondary education would not necessarily mean they would fail a class in higher education. Six explicitly mentioned that ADHD characteristics could influence predictions, such as the tendency to start studying later in the semester or the need to move while learning. For example, Participant 4 explained this in the following manner:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eYou mentioned the activities on learning platforms. Now, for example, when you say that my clicks are somehow looked at, how often I do the exercises and if I don't do anything for a long time, I might get a warning, an early warning. Well, as a person who always does everything at the last second, I would get a lot of messages saying: Hey, you haven't done anything on our exercises yet, why don't you do it? That would stress me out even more.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eConcerns for inaccurate predictions were particularly named when considering intervention without humans. This also led eight students to say that receiving an automatic e-mail would trigger negative emotions such as stress, demotivation, uncertainty about why they would be assessed in this manner or feeling patronized.\u003c/p\u003e \u003cp\u003eAdditionally, half of the interviewees believed that an automatic system without human intervention could be inefficient or useless. In particular, seven participants reckoned that learners already know when they are struggling, and that support offers already exist. Others thought that an e-mail was easy to ignore and would not necessarily motivate them to ask for support. On the other hand, seven students highlighted the benefits of proactive support in situations when students may be unaware of their struggle or of existing solutions or they are too shy to ask for help.\u003c/p\u003e \u003cp\u003e In the case of the intervention with a lecturer, six participants highlighted how a lecturer can truly provide proactive and individualized support. Participant 16 notably emphasized that human intervention could be more thoughtful (\u0026ldquo;taktgef\u0026uuml;hl\u0026rdquo; in German) and would allow him to explain himself. Nevertheless, seven expressed strong concerns related to the risk of having lecturers stereotype and label students which could lead to unfair grading. For example, Participant 2 said the following:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eI think it's really bad, I don't think it's a good thing at all. I think it leads to a lot more stigmatization of people based on stereotypical assumptions about who needs help and who doesn't. And especially when the list goes to lecturers, we know from studies that if you tell teachers in advance, \u0026ldquo;This student is very good and this one is very bad\u0026rdquo;, they will grade accordingly, regardless of intelligence level. So, I think it leads to stigmatization and that people are then treated according to the bias they have.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eCompared to the first two interventions, the third intervention for structural measures was positively perceived. Reasons for this were that the focus was less on the individual and that this could encourage universities to review their course structures, as explained by Participant 4:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eI prefer [the intervention with structural measures] best because it is not so individualized and because I think it would force the university to make structural changes to a course that isn't working so well with different groups of people, for example, and not simply put the blame on the individuals, in the sense of: Ah, you're not getting through, that means you need extra support. Instead, we say: No, something about the course is not right. We're looking at how we can reach more people. That's what I like better.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eParticipants were also invited to reflect on how the design of EWS could be improved. Eight interviewees emphasized they would like to remain in control, i.e., receiving a prediction and following the support advice should be voluntary. Six also wished for detailed information on their performance, showing both strengths and weaknesses. Participant 4 argued that a positive and supportive formulation would be important for her to reduce additional stress. In the quote below, she emphasized that this aspect is crucial for individuals with ADHD:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eI think the question made me realize in general that, if [early warning systems] happen at some point, it has to be done with a lot of sensitivity, because even if it's AI, it's still people who are affected with their feelings. Especially with ADHD, along with our comorbidities, where maybe you get depressed faster or feel more anxious about rejection, especially in such cases, you have to choose your words very carefully when something like that comes up.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eAdditionally, five mentioned that the intervention could be improved by focusing on how to support at-risk students (e.g. providing resources or exercises, sending task reminders). Two students suggested integrating human oversight into the system.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section3\"\u003e \u003ch2\u003e4.2.2 Use case: AI Tutor\u003c/h2\u003e \u003cp\u003eThe other use case described an AI tutor that could support six functions:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eStudying and preparing for an exam\u003c/b\u003e: The AI could prepare students for an exam through discussion (as a sparring partner). It could also answer questions about the lesson content.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eExplanations of exercises and instructions\u003c/b\u003e: The AI could explain the instructions of an exercise without giving the answer.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eEditing\u003c/b\u003e: The AI could edit text and suggest improvements.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eAdvice on courses and learning materials\u003c/b\u003e: The AI could recommend courses and learning materials. For example, if a student prefers to learn with videos, the AI could suggest learning videos instead of articles.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eSupport for time management and organization\u003c/b\u003e: The AI could, for example, create a learning plan and send reminders for important deadlines at the university.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eSupport networking among students\u003c/b\u003e: For example, the AI could recommend groups of students with common interests.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eIn general, students viewed an AI tutor as a useful tool, though certain functions garnered more interest than others. The most valuable feature would be its support for time management and organization. Among the 16 participants who mentioned benefits, 10 linked this utility to challenges related to ADHD symptoms or personal difficulties. Often, students mentioned the difficulty of creating a realistic study plan and stick to it. However, four persons doubted an AI could be helpful because they believed the study plan would be too strict for them and could cause more stress than relief. Participant 14 highlighted how it could be better to help him assess the amount of work rather than providing a fixed plan:\u003c/p\u003e \u003cp\u003e\u0026ldquo;With ADHD and procrastination and so on, it's quite difficult with deadlines and systematics and so on. Systematic in the sense of a regular habit. For example, if it told me: \u0026ldquo;You have to spend an hour every week on this subject, an hour on this subject, half an hour on this subject.\u0026rdquo; For me, that's much less useful than when it structures me like this: \u0026ldquo;Here are the topics, this is the scope of topics here and there. This is the expected amount of time and work.\u0026rdquo; Because then it's not so much telling me what I have to do, but what I should do. And this distinction is really important for me when it comes to my own time management, because I can't follow my own time management very well either way.\u0026rdquo;\u003c/p\u003e \u003cp\u003eEleven interviewees positively perceived the use of an AI Tutor to help with studying and preparing for an exam. Among them, five participants highlighted that it would enable them or other students to ask questions they would not dare ask in front of others or the lecturer. Still, two persons expressed concerns that the function would not help them learn effectively and one mentioned the risk of losing human contact. Additionally, Participant 9 emphasized the importance of the tool being optimized for the course, explaining that in his math and physics courses, it is essential for the problem-solving methods to align with those taught in class.\u003c/p\u003e \u003cp\u003eHalf of the participants valued the text editor function, notably to help detect careless mistakes (\u0026ldquo;Fl\u0026uuml;chtigkeitsfehler\u0026rdquo; in German) in writing often associated with ADHD. Nevertheless, four participants raised concerns about potential text standardization, diminished development of writing skills, or the risk of text appearing plagiarized. Two students were also skeptical it could help them as they needed more support with the content of the text rather than spelling. Two others also explained that they already had good editing tools.\u003c/p\u003e \u003cp\u003eSeven participants reported benefits for the feature \u0026ldquo;explanation of exercises and instructions\u0026rdquo;. In particular, four mentioned that they had previously struggled to understand instructions which had hindered their learning. Two also added that this feature would make them less dependent on other students. Nonetheless, three interviewees argued that the feature might reduce learning if students did not spend enough time trying to understand the task on their own.\u003c/p\u003e \u003cp\u003eOpinions on the feature \u0026ldquo;advice on courses and learning materials\u0026rdquo; were more mitigated. On the one hand, eight persons saw possibilities to personalize learning or to facilitate course selection. On the other hand, six believed that this feature would not be useful because they already knew how to select their courses or that they believed that this was a lecturer\u0026rsquo;s task.\u003c/p\u003e \u003cp\u003eIn comparison to other features, networking support was perceived as less useful. Students would not necessarily see why an AI could help with this task as networking possibilities occur by themselves, especially in small study programs. Others also found it too personal or did not like learning with others.\u003c/p\u003e \u003cp\u003eParticipants were also asked about their preference between using an editing tool from a private company (e.g. Grammarly) and their university. The large majority of interviewees perceived positively that universities provide such tools, notably due to the belief that their data would be better protected. Four also mentioned that the tool could be better optimized for their academic needs. At the same time, four participants questioned whether universities have the capacity to provide tools as good as those from private companies. Additionally, two interviewees raised concerns that university staff could misuse the information. For example, lecturers could evaluate the original version of a text before it was modified with AI, thus making the recourse to AI-based text editing useless.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Willingness and reasons to disclose ADHD in AI use cases\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the reported willingness to share ADHD status in the two AI use cases among the students who disclosed their ADHD at their HEI and those who did not. Those who had disclosed their ADHD to their HEI were in general more willing to share their neurodivergence with an AI-based system, regardless of the use case. Participants who had not informed their HEI that they have ADHD were more reluctant, especially with EWS.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn EWS, 13 participants would share their ADHD status hoping to increase accuracy or improve intervention. For example, Participant 19 argued that interventions would need to take into account whether the person has ADHD or not:\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eSpeaking again from my own experience: at the beginning of my studies, the problem for me was not that I didn't fundamentally understand the subject matter, but simply the quantity, how to deal with the material \u0026ldquo;how do I learn efficiently? How do I prioritize, how do I create a structure?\u0026rdquo; That's why I simply have the feeling that these measures, if they are based on this fictitious system, can actually be tailored quite differently for ADHD. So I see the possibilities of providing better interventions with ADHD.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eTwo participants also mentioned they would share their ADHD data because it could help others receive adequate support.\u003c/p\u003e \u003cp\u003eIn total, 7 interviewees said they would not disclose their ADHD in an EWS, even if in two cases, the students reckoned that it could help intervene according to neurodivergent students\u0026rsquo; needs. Reasons for not sharing were mainly due to bias risk and a belief that sharing that information would not be useful.\u003c/p\u003e \u003cp\u003eFourteen students said they would disclose their ADHD to an AI tutor mainly to enable greater personalization. Among them, eight explained that time management could become more flexible and send more reminders. Two participants also imagined that the system could thus recommend learning resources that are more helpful for individuals with ADHD. Additionally, Participant 20 mentioned that the tool could be specifically designed to enhance concentration thanks to shortened paragraphs, color choice or a reminder to take short breaks. Two persons said that it could help students with ADHD connect with one another.\u003c/p\u003e \u003cp\u003eFour participants answered that they would not share their ADHD with an AI tutor. They explained that everybody learns differently, ADHD affects people uniquely, and having ADHD would not significantly impact their learning experience with an AI tutor. One person also elucidated that they would rather disclose their ADHD to a human.\u003c/p\u003e \u003cp\u003e Apart from the two use cases, participants were invited to talk about where they found information about their condition and whether they already disclosed their ADHD in an AI system. While mental health professionals such as therapists remain a primary source of information, 18 interviewees mentioned looking up the internet. For 11 students, social media was a source of information. In some cases, this is how they came to think they might have ADHD. Among them, three explained that they did not actively search for information, but that content was suggested in their timelines. Additionally, although only one person indicated that they wrote in an AI system that they have ADHD, four other persons mentioned asking questions related to ADHD in ChatGPT.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.4 Conditions to share data\u003c/h2\u003e \u003cp\u003eThe 20 participants were asked \u0026ldquo;Who should have access to your data?\u0026rdquo; when discussing each of the presented AI use case (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). As the question was open, the total number of responses is not equal for each stakeholder. In the question, the term \u0026ldquo;data\u0026rdquo; was not explicitely defined. However, the interviewer mentioned online interaction data (e.g. click on activities on a learning platform or time spent) and previous grades when describing the EWS. For the AI tutor, the use of textual data was implied. When participants asked for clarification, the interviewer named these types of data. Additionally, the question followed one asking whether students would share information about their ADHD and the interviewer reminded them that this was also a type of information that could be shared.\u003c/p\u003e \u003cp\u003eIn both cases, access for lecturers was more controversial, notably due to stigmatization risk or negative consequences on grading. In the case of the AI tutor, some students pointed out that it would change their interaction with the tool. Still, three interviewees explained they would grant access to lecturers in an EWS because instructors are more apt to intervene adequately than in an automatic system. Another person mentioned that they can control for errors. Two persons emphasized how important it is for them to remain in control to decide which lecturer gets access to data. For example, Participant 10 drew a parallel with the procedure for notifying lecturers about reasonable accommodations: instructors are informed of the accommodations but not the students\u0026rsquo; specific conditions, allowing students to choose whether to discuss the subject with lecturers individually.\u003c/p\u003e \u003cp\u003eIn the case of AI tutors, seven participants mentioned that lecturers could have access to their data to improve the course in the future. Three of them emphasized that data should be anonymized for this. For example, Participant 1 suggested showing only the most frequently asked questions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAccess for faculty staff was in both use cases more acceptable, notably because participants imagined a more anonymous use where the university seeks to improve course quality for future students rather than intervene individually. In general, anonymity was central in AI tutors as 75% of participants would prefer data to be stored anonymously. Six participants first answered that data were to remain between them and the AI tool. Eleven interviewees would convene that IT team should have access to data to improve the tool.\u003c/p\u003e \u003cp\u003eIn comparison, in an EWS, despite that 13 would prefer anonymized data, nine also would accept logging with their name as this enables them to receive support. As a result, eight students highlighted the importance for them to be able to decide whether they want to use such a tool, who has access to information and whether they want to follow through support recommendations. This preference for greater involvement in decision-making was evident as five individuals expressed a desire for detailed information on how the prediction was made.\u003c/p\u003e \u003c/div\u003e"},{"header":"5 Discussion","content":"\u003cp\u003eThe interviews with students with ADHD have highlighted a clear difference between AI EdTech-supporting lecturers\u0026rsquo; tasks and those assisting students. AI tutors were perceived positively and seen as a tool to complement learning and overcome personal difficulties. While participants recognized the value of providing proactive support with EWS, their concerns about labeling and stereotyping called for greater student control. As highlighted by Culnan and Armstrong (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e1999\u003c/span\u003e), when individuals believe inferences drawn from their data to be incorrect, they experience the data collection as an invasion of privacy. In comparison, the unwillingness to disclose ADHD to AI tutors related to the lack of relevance.\u003c/p\u003e \u003cp\u003eThe general student population is also concerned about the risk of surveillance associated with the increased use of data to optimize HEIs (Jones et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). However, focusing on the perspective of students with ADHD or students who face systemic barriers due to their race, gender, or disabilities, has value because 1) they show concerns that could increase existing inequalities that HEIs are committed to eliminating, and 2) ensuring that systems do not disadvantage these students is likely to benefit all students. In this study, interviewees regularly connected their concerns or opinions (positive and negative) with their experience as an individual with ADHD. In particular, the concern for stigmatization and labeling cannot be ignored in light of the existing research showing that neurodivergent students still report discrimination in tertiary education (Clouder et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Grimes et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Mori\u0026ntilde;a, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Osborne, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Considerations on human involvement data policies in AI EdTech\u003c/h2\u003e \u003cp\u003eInterestingly, students in this research did not necessarily oppose humans to the supposed anonymity of technology. Several students emphasized that human bias is replicated in AI-based systems. An important factor was who had access to their data, naming those with greater influence on their study path and career (i.e. lecturers) as more critical. Li et al. (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) found that students\u0026rsquo; willingness to consent to leaning analytics depends on their comfort with instructors using their data for learning engagement. They argued that lecturers build relationships with students, fostering trust and increasing acceptance of data sharing compared to requests from administrative staff (Li et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). However, our study suggests that students with ADHD may view non-grading academic staff as less intimidating, as learners can remain anonymous and do not risk negative academic consequences. This should encourage developers to identify the most acceptable stakeholders to intervene with AI-based systems.\u003c/p\u003e \u003cp\u003eKizilcec (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) argued for more research on how lecturers perceive AI EdTech because they are the final decision-makers. However, Despande and Sharp (2022, p. 233) explained that \u0026ldquo;users of the system are the most relevant stakeholders when considering who is likely to be impacted by responsible AI systems\u0026rdquo; and continued by identifying those underrepresented in datasets as most likely to be affected by AI systems. While educators are important stakeholders, they are not likely to be as impacted as students by AI-based decisions. A focus on students with disabilities is essential due to the fact that this group is often underrepresented in datasets (Riazy et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eTherefore, like Marsh et al. (2024), we call for greater student agency in AI EdTech. In their study, Marsh \u0026amp; Milne (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) found that students were more likely to share disability information with lecturers and peers later in their studies as they developed trusting relationships. Similarly, in our study, some participants wished to control which instructors could access their data because they knew some were more understanding than others. This suggests that data access should be easy to modify over time. However, the necessity of a trusting relationship between students and HEI staff could mean that EWS may not be as effective as intended considering that such tools typically target students who recently started their studies.\u003c/p\u003e \u003cp\u003eStudent agency could also translate into allowing students to decide on the intervention following a prediction from an EWS. For example, Han et al. (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2025\u003c/span\u003e, p. 17) encouraged AI EdTech developers to let students \u0026ldquo;adjust the frequency, tone, and type of feedback\u0026rdquo; they receive. This recommendation aligns with our findings where some participants preferred human support, while others favored insights into their strengths and weaknesses to reflect on their learning progress. Providing such control over an EWS could also accommodate those who feel anxious about receiving feedback on their performance.\u003c/p\u003e \u003cp\u003eAdditionally, Marsh et al. (2024, p. 17) called for greater data transparency and argued that \u0026ldquo;any technology which offers accessibility options, including features such as the ability to change text size, turn on closed captions, enable read-aloud, adjust contrast and similar, should consider the status of those settings to be a potential privacy issue\u0026rdquo;. The development of accessible and inclusive AI tutors is likely to include such accessibility options. Many users may not realize that this technology implicitly shares data linked to neurodivergence or disability. Similarly, online searches and text input in chatbots become data containing information on disabilities. Some of our study participants were aware that their online searches could influence their social media feed which suggested them content on ADHD. Social media platforms often employ online user behavior and textual data to form profiles intended to provide users with interesting content and persons to connect (Gilbert et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Ricci et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Nevertheless, these may also be misused to target people for political purposes as the Cambridge Analytica case revealed when Facebook user data were used to influence US American voters (Cadwalladr \u0026amp; Graham-Harrison, 2018; Gilbert et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). From our study, it appears that some students did not realize that asking ChatGPT about ADHD generate data that could hint to their ADHD status if the information were to be extracted. At the moment and to the best of the authors\u0026rsquo; knowledge, ChatGPT interactions are not used to create user profiles and textual data is not turned into meaningful information. Still, to guarantee privacy rights, HEI are encouraged to raise awareness about how students\u0026rsquo; online behavior may provide sensitive information. Moreover, HEI need to consider this risk when using student data for research purposes, designing new applications, or acquiring AI-based applications from third parties.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e5.2 Collecting ADHD information to design inclusive AI EdTech\u003c/h2\u003e \u003cp\u003eIn general, students were open to sharing their data, even information on ADHD, especially in cases where the goal is to improve the university, as opposed to individual interventions. This is in line with research that found that the general student population trusts their university and is willing to share data for altruistic goals (Jones et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Y.-S. Tsai et al., \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). However, an experiment with the general student population indicated that 92% of participants were reluctant to share medical information (Ifenthaler \u0026amp; Schumacher, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Our study nuances this claim as students were willing to share their ADHD status to help other students and improve the accuracy of AI EdTech. This willingness is particularly evident among those who had disclosed their status to their HEI to receive reasonable accommodation for example. This finding may be due to the fact that these participants faced barriers in their studies and had positive experiences with the adaptations, making them more committed to enhancing inclusion measures in HEI. This result suggests that researchers and developers can collaborate with students with ADHD who are ready to share their lived experience and data to improve accessibility. Furthermore, it implies that optimizing AI tutors for students with ADHD would primarily benefit those with an official diagnosis who feel legitimate or comfortable seeking support.\u003c/p\u003e \u003cp\u003eThis also raises the question of the roles of different technological actors in creating responsible and inclusive AI. Students did not necessarily expect their universities to provide an all-rounding tool, acknowledging that universities may not have the resources to commit to the provision of such tools. Considering the trust towards HEI, researchers may be well-positioned to explore specific features that could benefit students with ADHD. These functions could then be integrated into existing tools, following a universal design approach. For example, an AI tutor could include features to support time management based on preferences and habits without labeling the option as \u0026ldquo;ADHD-friendly\u0026rdquo;. For EWS, researchers could focus on how to present information positively without triggering negative emotions that can particularly affect those prone to anxiety and depression.\u003c/p\u003e \u003c/div\u003e"},{"header":"6 Conclusion","content":"\u003cp\u003eThis study investigates ADHD disclosure in AI EdTech with 20 semi-structured interviews in the German-speaking regions of Switzerland. This work aimed to encourage researchers to investigate how to design inclusive AI EdTech, bearing in mind that a universal design approach would ensure that everyone benefits from it.\u003c/p\u003e \u003cp\u003eResults indicate that students with ADHD generally perceive AI tutors as more useful than EWS due to the risk of discrimination by AI systems and academic staff. Still, students were more open to the use of EWS if it aimed at structural changes. Moreover, acceptance of EWS could be increased by allowing students to opt in or out of such tools, offering them detailed information over how predictions are formulated, giving them control over which lecturer has access to their data and letting them decide on their preferred intervention type. Participants also showed interest in an AI tutor that could flexibly support them with time management.\u003c/p\u003e \u003cp\u003eThere is a certain openness among students to disclose their ADHD to AI EdTech, especially from those who have shared their conditions with their HEI. Perceived benefits of sharing this information with EWS include improving prediction accuracy and intervention. For AI tutors, participants saw an opportunity to get personalized resource recommendations and support with challenging tasks such as time management. Reasons not to disclose ADHD in the two uses comprised a perceived lack of relevance to share this information and a bias risk. Guaranteeing student control (e.g. letting them decide flexibly which lecturer gets access to their data) and anonymity is critical to preserve the privacy of students with ADHD.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e6.1 Limitations\u003c/h2\u003e \u003cp\u003eOne potential limitation of this study is the possibility of self-selection bias due to the chosen methodology. Although semi-structured interviews enable a deep understanding of participants\u0026rsquo; opinions on a topic, they require individuals who feel comfortable discussing the given subject. Despite efforts in our recruitment strategy to minimize this effect, we recognize that the openness towards disclosing ADHD could be more represented than those who prefer not to talk about it.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting Interests\u003c/h2\u003e \u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was financed through two swissuniversities projects: P7 \u0026ndash; Accessible teaching in higher education and P8\u0026mdash;Swiss digital skills academy: accessible open educational resources.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eThis study is part of Oriane Pierr\u0026egrave;s\u0026rsquo;s doctoral thesis under the supervision of Prof. Dr. Alireza Darvishy and PD Dr. Markus Christen. All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Oriane Pierr\u0026egrave;s. The first draft of the manuscript was written by Oriane Pierr\u0026egrave;s and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe thank Albiona Hajdari for her transcription of the interviews and her support in this work as a second coder of the interviews. We would also like to thank the following persons for sharing their expertise: Benjamin B\u0026ouml;rner, Nico Ebert, Corinna Hertweck, Sebastian W\u0026auml;scher, and Holger Baumann. We also thank Juliet Manning for proofreading this work.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eTo protect study participant privacy, interview transcripts are not made available. Although transcripts were anonymized by deleting information such as name, university, or location, participants shared personal information in the interviews, such as their ADHD status and specific life examples. This could potentially allow reidentification. However, the anonymized data can be made available to reviewers upon request to the corresponding author.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAboulafia, A. (2024, January 19). Internet privacy is a disability rights issues. \u003cem\u003eTech Policy\u003c/em\u003e. https://www.techpolicy.press/internet-privacy-is-a-disability-rights-issue/\u003c/li\u003e\n\u003cli\u003eAboulafia, A., Bogen, M., \u0026amp; Swenor, B. (2024). \u003cem\u003eTo Reduce Disability Bias in Technology, Start With Disability Data\u003c/em\u003e (p. 25). Center for democracy and technology (CDT). https://cdt.org/wp-content/uploads/2024/07/2024-07-23-Data-Disability-report-final.pdf\u003c/li\u003e\n\u003cli\u003eacatech - Deutsche Akademie der Technikwissenschaften \u0026amp; Joachim Herz Stiftung. (2024). \u003cem\u003eMINT Nachwuchsbarometer 2024\u003c/em\u003e (pp. 1\u0026ndash;17) [Application/pdf]. Acatech. https://doi.org/10.48669/ACA_2024-8\u003c/li\u003e\n\u003cli\u003eAmerican Psychiatric Association (Ed.). (2022). \u003cem\u003eDiagnostic and statistical manual of mental disorders: DSM-5-TR\u003csup\u003eTM\u003c/sup\u003e\u003c/em\u003e (Fifth edition, text revision). American Psychiatric Association Publishing.\u003c/li\u003e\n\u003cli\u003eBaillifard, A., Gabella, M., Lavenex, P. B., \u0026amp; Martarelli, C. S. (2024). Effective learning with a personal AI tutor: A case study. \u003cem\u003eEducation and Information Technologies\u003c/em\u003e. https://doi.org/10.1007/s10639-024-12888-5\u003c/li\u003e\n\u003cli\u003eBaker, R. S., \u0026amp; Hawn, A. (2021). Algorithmic Bias in Education. \u003cem\u003eInternational Journal of Artificial Intelligence in Education\u003c/em\u003e. https://doi.org/10.1007/s40593-021-00285-9\u003c/li\u003e\n\u003cli\u003eBatsaikhan, B. (Zack), \u0026amp; Correia, A.-P. (2024). The effects of Generative Artificial Intelligence on Intelligent Tutoring Systems in higher education: A systematic review. \u003cem\u003eStudies in Technology Enhanced Learning\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e(1). https://doi.org/10.21428/8c225f6e.33570bb1\u003c/li\u003e\n\u003cli\u003eBernstein, A., Cordoba, A., Witt, C., Siegers, D., Wessels, D., Thouvenin, F., Hastings, J., Hatt, J.-M., Buhmann, J., Martinez-Moreno, J., Schawinski, K., Christen, M., Coors, M., Denner, M., Hunger, P., von Thiessen, R., Fabrikant, S., Friemel, T., Neupert, T., \u0026amp; Sachs, U. (2024). \u003cem\u003eAI in Education, Research, and Innovation\u003c/em\u003e (pp. 1\u0026ndash;11) [Position Paper]. Universit\u0026auml;t Z\u0026uuml;rich. https://www.dsi.uzh.ch/dam/jcr:c1b5b525-39af-400f-9215-c29f25bc1c71/dsi-strategy-lab-23-positionspapier-en.pdf\u003c/li\u003e\n\u003cli\u003eBertilsdotter Rosqvist, H., Hultman, L., \u0026Ouml;sterborg Wiklund, S., Nygren, A., Storm, P., \u0026amp; Sandberg, G. (2023). ADHD in higher education and academia. \u003cem\u003eCanadian Journal on Disability Studies\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(3), 1\u0026ndash;26.\u003c/li\u003e\n\u003cli\u003eBoisot, M., \u0026amp; Canals, A. (2004). Data, information and knowledge: Have we got it right? \u003cem\u003eJournal of Evolutionary Economics\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(1), 43\u0026ndash;67. https://doi.org/10.1007/s00191-003-0181-9\u003c/li\u003e\n\u003cli\u003eBotha, M., Chapman, R., Giwa Onaiwu, M., Kapp, S. K., Stannard Ashley, A., \u0026amp; Walker, N. (2024). The neurodiversity concept was developed collectively: An overdue correction on the origins of neurodiversity theory. \u003cem\u003eAutism\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(6), 1591\u0026ndash;1594. https://doi.org/10.1177/13623613241237871\u003c/li\u003e\n\u003cli\u003eCadwalladr, C., \u0026amp; Graham-Harrison, E. (2018, March 17). Revealed: 50 million Facebook profiles harvested for Cambridge Analytica in major data breach. \u003cem\u003eThe Guardian\u003c/em\u003e. https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence-us-election\u003c/li\u003e\n\u003cli\u003eCAST. (n.d.-a). \u003cem\u003eUniversal Design for Learning\u003c/em\u003e. Retrieved 24 January 2025, from https://www.cast.org/what-we-do/universal-design-for-learning/\u003c/li\u003e\n\u003cli\u003eCAST. (n.d.-b). \u003cem\u003eUniversal Design for Learning Guidelines Version 2.2\u003c/em\u003e. Retrieved 2 December 2024, from http://udlguidelines.cast.org/\u003c/li\u003e\n\u003cli\u003eCiolacu, M., Tehrani, A. F., Binder, L., \u0026amp; Svasta, P. M. (2018). Education 4.0 - Artificial Intelligence Assisted Higher Education: Early recognition System with Machine Learning to support Students\u0026rsquo; Success. \u003cem\u003e2018 IEEE 24th International Symposium for Design and Technology in Electronic Packaging (SIITME)\u003c/em\u003e, 23\u0026ndash;30. https://doi.org/10.1109/siitme.2018.8599203\u003c/li\u003e\n\u003cli\u003eClouder, L., Karakus, M., Cinotti, A., Ferreyra, M. V., Fierros, G. A., \u0026amp; Rojo, P. (2020). Neurodiversity in higher education: A narrative synthesis. \u003cem\u003eHigher Education\u003c/em\u003e, \u003cem\u003e80\u003c/em\u003e(4), 757\u0026ndash;778. https://doi.org/10.1007/s10734-020-00513-6\u003c/li\u003e\n\u003cli\u003eColombo-Dougovito, A. M., Dillon, S. R., \u0026amp; Mpofu, E. (2020). The Wellbeing of People with Neurodiverse Conditions. In E. Mpofu (Ed.), \u003cem\u003eSustainable Community Health: Systems and Practices in Diverse Settings\u003c/em\u003e (pp. 499\u0026ndash;535). Springer International Publishing. https://doi.org/10.1007/978-3-030-59687-3_15\u003c/li\u003e\n\u003cli\u003eConvention on the Rights of Persons with Disabilities Pledged, Article 24 \u0026ndash; Education (2006). https://www.un.org/development/desa/disabilities/convention-on-the-rights-of-persons-with-disabilities.html\u003c/li\u003e\n\u003cli\u003eCulnan, M. J., \u0026amp; Armstrong, P. K. (1999). Information Privacy Concerns, Procedural Fairness, and Impersonal Trust: An Empirical Investigation. \u003cem\u003eOrganization Science\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(1), 104\u0026ndash;115. JSTOR.\u003c/li\u003e\n\u003cli\u003eDeshpande, A., \u0026amp; Sharp, H. (2022). Responsible AI Systems: Who are the Stakeholders? \u003cem\u003eProceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society\u003c/em\u003e, 227\u0026ndash;236. https://doi.org/10.1145/3514094.3534187\u003c/li\u003e\n\u003cli\u003eEuropean Union. (2016, April 27). \u003cem\u003eREGULATION (EU) 2016/679 OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation) (Text with EEA relevance)\u003c/em\u003e. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32016R0679\u0026amp;from=EN#d1e1374-1-1\u003c/li\u003e\n\u003cli\u003eFlowers, L. (2012). \u003cem\u003eNavigating and accessing higher education: The experiences of community college students with attention deficit hyperactivity disorder\u003c/em\u003e [Doctoral dissertation]. https://kortschakcenter.usc.edu/wp-content/uploads/2015/04/Flowers-Community-College-ADHD-Access.pdf\u003c/li\u003e\n\u003cli\u003eGilbert, J., Hamid, S., Hashem, I. A. T., Ghani, N. A., \u0026amp; Boluwatife, F. F. (2023). The rise of user profiling in social media: Review, challenges and future direction. \u003cem\u003eSocial Network Analysis and Mining\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(1), 137. https://doi.org/10.1007/s13278-023-01146-0\u003c/li\u003e\n\u003cli\u003eGrimes, S., Southgate, E., Scevak, J., \u0026amp; Buchanan, R. (2019). University student perspectives on institutional non-disclosure of disability and learning challenges: Reasons for staying invisible. \u003cem\u003eInternational Journal of Inclusive Education\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(6), 639\u0026ndash;655. https://doi.org/10.1080/13603116.2018.1442507\u003c/li\u003e\n\u003cli\u003eHan, B., Nawaz, S., Buchanan, G., \u0026amp; McKay, D. (2025). Students\u0026rsquo; Perceptions: Exploring the Interplay of Ethical and Pedagogical Impacts for Adopting AI in Higher Education. \u003cem\u003eInternational Journal of Artificial Intelligence in Education\u003c/em\u003e. https://doi.org/10.1007/s40593-024-00456-4\u003c/li\u003e\n\u003cli\u003eHellings, J., \u0026amp; Haelermans, C. (2022). The effect of providing learning analytics on student behaviour and performance in programming: A randomised controlled experiment. \u003cem\u003eHigher Education\u003c/em\u003e, \u003cem\u003e83\u003c/em\u003e(1), 1\u0026ndash;18. https://doi.org/10.1007/s10734-020-00560-z\u003c/li\u003e\n\u003cli\u003eIfenthaler, D., \u0026amp; Schumacher, C. (2016). Student perceptions of privacy principles for learning analytics. \u003cem\u003eEducational Technology Research and Development\u003c/em\u003e, \u003cem\u003e64\u003c/em\u003e(5), 923\u0026ndash;938. https://doi.org/10.1007/s11423-016-9477-y\u003c/li\u003e\n\u003cli\u003eJohnson, D. (2019). \u003cem\u003eOCLC Research Works in Progress Webinar: Deakin Genie\u003c/em\u003e. https://www.oclc.org/content/dam/research/presentations/2019/Deakin-Genie.pdf\u003c/li\u003e\n\u003cli\u003eJones, K. M. L., Asher, A., Goben, A., Perry, M. R., Salo, D., Briney, K. A., \u0026amp; Robertshaw, M. B. (2020). \u0026ldquo;We\u0026rsquo;re being tracked at all times\u0026rdquo;: Student perspectives of their privacy in relation to learning analytics in higher education. \u003cem\u003eJournal of the Association for Information Science and Technology\u003c/em\u003e, \u003cem\u003e71\u003c/em\u003e(9), 1044\u0026ndash;1059. https://doi.org/10.1002/asi.24358\u003c/li\u003e\n\u003cli\u003eKhan Academy. (n.d.). \u003cem\u003eKhanmigo by Khan Academy\u003c/em\u003e. Retrieved 14 June 2024, from https://www.khanmigo.ai/\u003c/li\u003e\n\u003cli\u003eKizilcec, R. F. (2024). To Advance AI Use in Education, Focus on Understanding Educators. \u003cem\u003eInternational Journal of Artificial Intelligence in Education\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(1), 12\u0026ndash;19. https://doi.org/10.1007/s40593-023-00351-4\u003c/li\u003e\n\u003cli\u003eLabadze, L., Grigolia, M., \u0026amp; Machaidze, L. (2023). Role of AI chatbots in education: Systematic literature review. \u003cem\u003eInternational Journal of Educational Technology in Higher Education\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 56. https://doi.org/10.1186/s41239-023-00426-1\u003c/li\u003e\n\u003cli\u003eLaufer, R. S., \u0026amp; Wolfe, M. (1977). Privacy as a Concept and a Social Issue: A Multidimensional Developmental Theory. \u003cem\u003eJournal of Social Issues\u003c/em\u003e, \u003cem\u003e33\u003c/em\u003e(3), 22\u0026ndash;42. https://doi.org/10.1111/j.1540-4560.1977.tb01880.x\u003c/li\u003e\n\u003cli\u003eLi, W., Sun, K., Schaub, F., \u0026amp; Brooks, C. (2022). Disparities in Students\u0026rsquo; Propensity to Consent to Learning Analytics. \u003cem\u003eInternational Journal of Artificial Intelligence in Education\u003c/em\u003e, \u003cem\u003e32\u003c/em\u003e(3), 564\u0026ndash;608. https://doi.org/10.1007/s40593-021-00254-2\u003c/li\u003e\n\u003cli\u003eMarsh, A., \u0026amp; Milne, L. R. (2024). I Don\u0026rsquo;t Want to Sound Rude, but It\u0026rsquo;s None of Their Business: Exploring Security and Privacy Concerns around Assistive Technology Use in Educational Settings. \u003cem\u003eACM Trans. Access. Comput.\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(2). https://doi.org/10.1145/3670690\u003c/li\u003e\n\u003cli\u003eMayring, P. (2014). \u003cem\u003eQualitative content analysis: Theoretical foundation, basic procedures and software solution\u003c/em\u003e. https://www.ssoar.info/ssoar/handle/document/39517\u003c/li\u003e\n\u003cli\u003eMcMullin, C. (2023). Transcription and Qualitative Methods: Implications for Third Sector Research. \u003cem\u003eVOLUNTAS: International Journal of Voluntary and Nonprofit Organizations\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(1), 140\u0026ndash;153. https://doi.org/10.1007/s11266-021-00400-3\u003c/li\u003e\n\u003cli\u003eMonlla\u0026oacute; Oliv\u0026eacute;, D., Huynh, D. Q., Reynolds, M., Dougiamas, M., \u0026amp; Wiese, D. (2020). A supervised learning framework: Using assessment to identify students at risk of dropping out of a MOOC. \u003cem\u003eJournal of Computing in Higher Education\u003c/em\u003e, \u003cem\u003e32\u003c/em\u003e(1), 9\u0026ndash;26. https://doi.org/10.1007/s12528-019-09230-1\u003c/li\u003e\n\u003cli\u003eMori\u0026ntilde;a, A. (2024). When what is unseen does not exist: Disclosure, barriers and supports for students with invisible disabilities in higher education. \u003cem\u003eDisability \u0026amp; Society\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(4), 914\u0026ndash;932. https://doi.org/10.1080/09687599.2022.2113038\u003c/li\u003e\n\u003cli\u003eMorris, M. R. (2020). AI and Accessibility. \u003cem\u003eCommun. ACM\u003c/em\u003e, \u003cem\u003e63\u003c/em\u003e(6), 35\u0026ndash;37. https://doi.org/10.1145/3356727\u003c/li\u003e\n\u003cli\u003eMullins, L., \u0026amp; Preyde, M. (2013). The lived experience of students with an invisible disability at a Canadian university. \u003cem\u003eDisability \u0026amp; Society\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(2), 147\u0026ndash;160. https://doi.org/10.1080/09687599.2012.752127\u003c/li\u003e\n\u003cli\u003eNewman, L. A., \u0026amp; Madaus, J. W. (2015). Reported Accommodations and Supports Provided to Secondary and Postsecondary Students With Disabilities: National Perspective. \u003cem\u003eCareer Development and Transition for Exceptional Individuals\u003c/em\u003e, \u003cem\u003e38\u003c/em\u003e(3), 173\u0026ndash;181. https://doi.org/10.1177/2165143413518235\u003c/li\u003e\n\u003cli\u003eOrtigosa, A., Carro, R. M., Bravo-Agapito, J., Lizcano, D., Alcolea, J. J., \u0026amp; Blanco, O. (2019). From Lab to Production: Lessons Learnt and Real-Life Challenges of an Early Student-Dropout Prevention System. \u003cem\u003eIEEE Transactions on Learning Technologies\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(2), 264\u0026ndash;277. https://doi.org/10.1109/TLT.2019.2911608\u003c/li\u003e\n\u003cli\u003eOsborne, T. (2019). Not lazy, not faking: Teaching and learning experiences of university students with disabilities. \u003cem\u003eDisability \u0026amp; Society\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(2), 228\u0026ndash;252. https://doi.org/10.1080/09687599.2018.1515724\u003c/li\u003e\n\u003cli\u003ePierr\u0026egrave;s, O., Christen, M., Schmitt-Koopmann, F. M., \u0026amp; Darvishy, A. (2024). Could the Use of AI in Higher Education Hinder Students With Disabilities? A Scoping Review. \u003cem\u003eIEEE Access\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, 27810\u0026ndash;27828. https://doi.org/10.1109/ACCESS.2024.3365368\u003c/li\u003e\n\u003cli\u003ePierr\u0026egrave;s, O., Darvishy, A., \u0026amp; Christen, M. (2024). Exploring the role of generative AI in higher education: Semi-structured interviews with students with disabilities. \u003cem\u003eEducation and Information Technologies\u003c/em\u003e. https://doi.org/10.1007/s10639-024-13134-8\u003c/li\u003e\n\u003cli\u003eRiazy, S., Simbeck, K., \u0026amp; Schreck, V. (2020). Fairness in Learning Analytics: Student At-risk Prediction in Virtual Learning Environments: \u003cem\u003eProceedings of the 12th International Conference on Computer Supported Education\u003c/em\u003e, 15\u0026ndash;25. https://doi.org/10.5220/0009324100150025\u003c/li\u003e\n\u003cli\u003eRicci, F., Rokach, L., \u0026amp; Shapira, B. (2022). Recommender Systems: Techniques, Applications, and Challenges. In F. Ricci, L. Rokach, \u0026amp; B. Shapira (Eds.), \u003cem\u003eRecommender Systems Handbook\u003c/em\u003e (pp. 1\u0026ndash;35). Springer US. https://doi.org/10.1007/978-1-0716-2197-4_1\u003c/li\u003e\n\u003cli\u003eRoberts, J. B., Crittenden, L. A., \u0026amp; Crittenden, J. C. (2011). Students with disabilities and online learning: A cross-institutional study of perceived satisfaction with accessibility compliance and services. \u003cem\u003eThe Internet and Higher Education\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(4), 242\u0026ndash;250. https://doi.org/10.1016/j.iheduc.2011.05.004\u003c/li\u003e\n\u003cli\u003eR\u0026ouml;mhild, A., \u0026amp; Hollederer, A. (2024). Effects of disability-related services, accommodations, and integration on academic success of students with disabilities in higher education. A scoping review. \u003cem\u003eEuropean Journal of Special Needs Education\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(1), 143\u0026ndash;166. https://doi.org/10.1080/08856257.2023.2195074\u003c/li\u003e\n\u003cli\u003eSajja, R., Sermet, Y., Cikmaz, M., Cwiertny, D., \u0026amp; Demir, I. (2024). Artificial Intelligence-Enabled Intelligent Assistant for Personalized and Adaptive Learning in Higher Education. \u003cem\u003eInformation\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e(10). https://doi.org/10.3390/info15100596\u003c/li\u003e\n\u003cli\u003eSchippers, L. M., Horstman, L. I., Velde, H. V. D., Pereira, R. R., Zinkstok, J., Mostert, J. C., Greven, C. U., \u0026amp; Hoogman, M. (2022). A qualitative and quantitative study of self-reported positive characteristics of individuals with ADHD. \u003cem\u003eFrontiers in Psychiatry\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e, 922788. https://doi.org/10.3389/fpsyt.2022.922788\u003c/li\u003e\n\u003cli\u003eSghir, N., Adadi, A., \u0026amp; Lahmer, M. (2023). Recent advances in Predictive Learning Analytics: A decade systematic review (2012\u0026ndash;2022). \u003cem\u003eEducation and Information Technologies\u003c/em\u003e, \u003cem\u003e28\u003c/em\u003e(7), 8299\u0026ndash;8333. https://doi.org/10.1007/s10639-022-11536-0\u003c/li\u003e\n\u003cli\u003eSmith, S. A., Woodhead, E., \u0026amp; Chin-Newman, C. (2021). Disclosing accommodation needs: Exploring experiences of higher education students with disabilities. \u003cem\u003eInternational Journal of Inclusive Education\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(12), 1358\u0026ndash;1374. https://doi.org/10.1080/13603116.2019.1610087\u003c/li\u003e\n\u003cli\u003eSoffer, T., \u0026amp; Cohen, A. (2024). Privacy versus pedagogy\u0026ndash;students\u0026rsquo; perceptions of using learning analytics in higher education. \u003cem\u003eAustralasian Journal of Educational Technology\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eSpiel, K., Hornecker, E., Williams, R. M., \u0026amp; Good, J. (2022). ADHD and Technology Research \u0026ndash; Investigated by Neurodivergent Readers. \u003cem\u003eProceedings of the 2022 CHI Conference on Human Factors in Computing Systems\u003c/em\u003e. https://doi.org/10.1145/3491102.3517592\u003c/li\u003e\n\u003cli\u003eTsai, S., Chen, C., Shiao, Y., Ciou, J., \u0026amp; Wu, T. (2020). Precision education with statistical learning and deep learning: A case study in Taiwan. \u003cem\u003eINTERNATIONAL JOURNAL OF EDUCATIONAL TECHNOLOGY IN HIGHER EDUCATION\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(1). https://doi.org/10.1186/s41239-020-00186-2\u003c/li\u003e\n\u003cli\u003eTsai, S.-C., Chen, C.-H., Shiao, Y.-T., Ciou, J.-S., \u0026amp; Wu, T.-N. (2020). Precision education with statistical learning and deep learning: A case study in Taiwan. \u003cem\u003eInternational Journal of Educational Technology in Higher Education\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(1). https://doi.org/10.1186/s41239-020-00186-2\u003c/li\u003e\n\u003cli\u003eTsai, Y.-S., Whitelock-Wainwright, A., \u0026amp; Ga\u0026scaron;ević, D. (2020). The privacy paradox and its implications for learning analytics. \u003cem\u003eProceedings of the Tenth International Conference on Learning Analytics \u0026amp; Knowledge\u003c/em\u003e, 230\u0026ndash;239. https://doi.org/10.1145/3375462.3375536\u003c/li\u003e\n\u003cli\u003eWorld Health Organisation. (2021, November 4). \u003cem\u003eDisability and Health\u003c/em\u003e. https://www.who.int/news-room/fact-sheets/detail/disability-and-health#:~:text=Disability%20refers%20to%20the%20interaction,%2C%20and%20limited%20social%20supports).\u003c/li\u003e\n\u003cli\u003eZawacki-Richter, O., Mar\u0026iacute;n, V. I., Bond, M., \u0026amp; Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education \u0026ndash; where are the educators? \u003cem\u003eInternational Journal of Educational Technology in Higher Education\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(1). Scopus. https://doi.org/10.1186/s41239-019-0171-0\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"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":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Privacy, AI Ethics, Higher Education, Accessibility, Neurodivergent students","lastPublishedDoi":"10.21203/rs.3.rs-6106311/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6106311/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study investigates the perspectives of students with attention deficit and hyperactivity disorder (ADHD) on disclosing their condition in the context of AI-based educational technologies (AI EdTech). Neurodivergent students often face challenges in disclosing their condition. It is unclear whether these difficulties persist in the context of AI EdTech. On the one hand, collecting data on neurodiversity could help ensure that these technologies are inclusive and personalized. Moreover, some students might find it easier to disclose their neurodivergence to an AI where their anonymity is guaranteed, rather than to colleagues or peers who may harbor negative attitudes. On the other hand, gathering and storing disability data might pose privacy risks depending on the technology design, such as the potential for re-identification, and may pressure neurodivergent students to disclose their conditions in order to access services. To better understand how disclosure is perceived in AI EdTech, we conducted 20 semi-structured interviews with students with ADHD. Results suggest that participants perceived AI tutors more positively than early warning systems due to a risk of stigmatization. This concern could be addressed by granting students greater control over their data, especially in deciding which lecturer should have access to their information. Still, participants were generally open to disclosing their ADHD status in AI EdTech, especially those who had already disclosed their ADHD to their universities. Finally, this paper provides reflections for developers and educators to create inclusive adaptive educational technologies that respect students\u0026rsquo; privacy.\u003c/p\u003e","manuscriptTitle":"Perceived Risks and Benefits of Disclosing ADHD to AI-based Educational Technologies: Semi-structured Interviews","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-04 08:58:00","doi":"10.21203/rs.3.rs-6106311/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"176228a0-fc8f-4de5-ba14-6044dd1ac2aa","owner":[],"postedDate":"March 4th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-04-29T15:08:45+00:00","versionOfRecord":[],"versionCreatedAt":"2025-03-04 08:58:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6106311","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6106311","identity":"rs-6106311","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}