Empowering Neurodiversity - Analyzing the Potential of Visual Data Annotation as Employment for Autistic Individuals

Empowering Neurodiversity - Analyzing the Potential of Visual Data Annotation as Employment for Autistic Individuals | 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 Empowering Neurodiversity - Analyzing the Potential of Visual Data Annotation as Employment for Autistic Individuals Dominik Laister, Kristina Köck, Lukas Fischer, Bernhard Schenkenfelder, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6646902/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Artificial Intelligence (AI) systems, which are currently reshaping the global employment landscape, require data to be annotated for training and evaluation – a task performed mainly by annotation workers located in the Global South. That AI-related companies are located in Western countries, but annotation work is performed in regions that are culturally and socioeconomically different implies low-quality, potentially culturally biased annotation under unacceptable working conditions. It would thus be desirable to move annotation work to Western countries. This study investigated whether annotation could be a promising occupation for individuals on the autism spectrum (AS), an underprivileged group in the labor market in Western countries. Our participatory research project included adults on the AS (experimental group, EG; N = 22) and a neurotypical comparison group (CG; N = 17). Job-related self-perceptions were collected from all participants, neurocognitive parameters (nonverbal IQ, reading comprehension, visual attention) were assessed for EG participants, and correlations with annotation outcomes were analyzed. Findings confirmed the hypothesized potential of autistic individuals to perform high quality data annotation. Good reading comprehension supported annotation outcomes, while higher IQ scores (> 110) were related to a tendency towards lower annotation quality, which might be explained by annotation tasks being insufficiently demanding, although neither correlation was significant in the limited study sample. Annotation performance might be supported by particular working conditions (i.e., working in small groups, minimal distraction, visual guidelines). Future studies should explore long-term trajectories (i.e., staying in work, income development, well-being) for annotation workers on the AS to confirm the reality of opportunities for this population. Visual Data Annotation Crowd Work Autism Spectrum Disorder Labor Market Inclusive Jobs Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Introduction 1.1 Artificial Intelligence and Annotation The digital revolution, driven by Artificial Intelligence (AI) and automation, is reshaping the employment landscape around the world, bringing both challenges and opportunities. While this transformation means that people are displaced from certain routine work and manual jobs, it is also creating opportunities in highly skilled roles such as data scientist, AI specialist, and other tech-focused positions (Milanez, 2023 ; World Economic Forum, 2023 ). ChatGPT has been reported to have a great impact on the labor market, with two-thirds of occupations being fully or partially affected, and thus reskilling and upskilling will be required to fight long-term unemployment (Zarifhonarvar, 2023 ). For instance, a report by the Boston Consulting Group ( 2021 ) highlighted that by 2030 the United States and Germany could face labor shortages of up to 12.5 million and up to 2.5 million people, respectively, particularly in tech-related fields. Data annotation is essential to the training, tuning, and evaluation of AI systems, and while research is underway to increase the level of automation, it is currently a manual task. Visual annotation is the process of determining the location and/or content of objects in pictures for an AI to learn and later recognize automatically. Drawing a bounding box on a traffic image defines, for example, the boundaries of a particular road user, and assigning a label to that box defines its class (i.e., bicycle, car, or truck). In the case of medical images, the label can also be applied to the entire image to indicate, for instance, whether a brain tumor is malignant. Regardless of the details, it is only by annotating images, videos, text, audio, or other data that the foundation for AI pipelines can be built. Despite the importance of data annotation to Western countries, where most AI research is conducted, the Global South, where most annotation work is done, experiences a vast recognition gap (Smart et al., 2024 ). This imbalance also manifests as exploitation through unfair salaries and as dehumanization through ghost work (Gray & Suri, 2019 ; Perrigo, 2023 ; Tan & Cabato, 2023 ; Wang et al., 2022 ). While it would be desirable for annotated data to be value-free, neutral and therefore generalizable, various studies have shown that some data are biased due to annotators‘ cultural, social, or geographical backgrounds. Whether their backgrounds are relevant depends on the type of data (Gray & Suri, 2019 ; Perrigo, 2023 ; Smart et al., 2024 ; Wang et al., 2022 ). The importance of such a context depends on the type of data. For example, road traffic data is more useful when annotated by annotators in this particular geographical region (Mark Díaz et al., 2022 ). In a recent workshop paper, Schenkenfelder et al ( 2024 ) shared promising results from a data annotation training initiative with neurodivergent individuals which led to formation of a successful, well-integrated team within an IT company. Supervisors of this inclusive team stated that “resilience to monotony” and “average reading comprehension skills” might be essential skills for trainees. 1.2 Autism Spectrum Condition and the Labor Market To address the problems laid out above, it would be desirable to move annotation work to the countries in which AI systems are developed and researched. As demand for (local) data annotation workers is increasing in these countries, opportunities to participate in the job market might open up for marginalized groups that are underrepresented in the labor market. Individuals on the Autism Spectrum (AS) with IQ scores in the normal range have been identified as being part of this marginalized societal group (Davies et al., 2024 ; Hara & Bigham, 2017 ; Solomon, 2020 ). According to the eleventh revision of the International Statistical Classification of Diseases and Related Health Problems (World Health Organization, 2022 ), Autism Spectrum Disorder (ASD) is classified as a neurodevelopmental condition characterized by persistent differences in social communication and interaction, alongside restricted and repetitive and inflexible patterns of behavior, interests, or activities. Autistic individuals show characteristic features or have difficulties in social reciprocity, nonverbal communication, or in developing and maintaining relationships. Individuals may also exhibit a strong preference for routines and show intense focus on specific topics or interests. Characteristics include atypical sensory responses, such as heightened sensitivity to light, sounds, textures, or smells. Individuals with ASD have varying intellectual abilities, ranging from intellectual disability to average and above-average intelligence. Severity varies widely, and the condition may coexist with other medical or psychiatric conditions (World Health Organization, 2022 ). Despite their ability and desire to work, the employment rate for individuals with ASD in the United Kingdom is 29% (Davies et al., 2024 ), and the average employment rate in Europe is as low as 10% (Baranger, 2019 ). While the UN Convention on the Rights of Persons with Disabilities (Hendriks, 2007) acknowledges the importance of professional participation and earned income for full societal inclusion, there remains a significant shortage of suitable employment opportunities for individuals with ASD within the broader workforce (Perkowski et al., 2024 ). Where access to appropriate jobs in the general labor market is difficult, people with ASD must overcome substantial barriers to find employment, regardless of their level of education (Espelöer et al., 2023 ; Lorenz et al., 2016 ; Zimmermann & Falkner, 2018 ). The reasons for these difficulties are manifold, ranging from barriers in finding appropriate jobs to lack of employer awareness of the strengths and potential of this group. Entering the labor market is challenging, often because insufficient training and on-the-job support for autistic individuals and their working environment is provided (Lorenz et al., 2016 ). Furthermore, higher stress levels during application, training and employment have been demonstrated (Davies et al., 2023 ). For instance, a supportive approach to job training, developed by Adam et al. ( 2017 ), proposes design considerations for an adaptive enterprise system which "assesses users' individual stress levels continuously and unobtrusively". This would allow enterprises to automatically adapt human–computer interfaces (e.g., information filtering, changes in screen color, information presentation mode) based on physiological measures (e.g., pupil dilation, heart rate, mouse pressure, muscle tension, pulse transit time, and skin conductance). 1.3 Autistic features as strengths in digital job tasks? Despite – or because of – their condition, individuals on the AS may possess skills and unique characteristics that make them particularly suited to annotation work, as suggested by a small but growing body of research (Garrison et al., 2023 ). Firstly, individuals diagnosed with ASD often demonstrate an exceptional ability to focus on specific details, which may enable them to excel at tasks that require meticulous attention (Baron-Cohen, 2009 ; Baron-Cohen et al., 2009 ; Jolliffe & Baron-Cohen, 2001 ). The frequently described ability to identify even subtle differences or inconsistencies within data sets makes them well-suited to labeling and categorizing complex data with accuracy (Baron-Cohen, 2009 ). Secondly, autistic individuals often demonstrate an ability to recognize patterns, which may allow them to identify and follow consistent patterns faster – a key competence in tasks such as categorizing data, tagging images, and annotating text for machine learning (Baron-Cohen, 2009 ). Many people on the AS prefer structured environments with clear rules and routines. Data annotation typically involves well-defined tasks and guidelines, which provides a structured framework. Furthermore, the focus on routine and precision in autistic individuals can result in consistent high-quality work when performing tasks such as reviewing and labeling datasets, where accuracy is crucial. Lastly and importantly, many autistic individuals have a high tolerance for repetitive tasks that others may find tedious or under-stimulating(Bury et al., 2021 ) . 1.4 Aims of this study In summary, there is a growing need for highly accurate annotation in the emerging field of AI, and a concurrent disadvantaged employment situation for individuals with ASD with promising qualities for digital tasks. This study analyzed how individuals on the autism spectrum perform in simple visual data-annotation tasks and sought to gain a deeper understanding of putative intraindividual characteristics, which may be associated with visual data-annotation performance. Research Question 1 : How do autistic participants perform in visual data annotation in terms of correctness (labeling tasks) and accuracy (bounding boxes) compared to a neurotypical comparison group? Research Question 2 : Is annotation performance in autistic participants associated with a) nonverbal IQ, b) visual attention performance and c) reading comprehension skills? 2 Methods 2.1 Participatory research design This study was designed and implemented as a participatory research project that included two investigators, one study-assistant and two autistic co-researchers and followed guidelines for participant-friendly research (Nicolaidis et al., 2019 ; Pisani et al., 2022 ). A focus group of six autistic individuals supported the final selection and adaptation of the measures implemented. Four consecutive meetings were co-led by an autistic co-researcher and implementation staff. All pre-selected measures suggested by the investigators were reflected critically in each focus group session to gather and apply specific feedback on aspects of feasibility (length, comprehension, quality of instruction, suitability). 2.2 Participants The total sample consisted of 39 participants, including individuals with a formal autism spectrum diagnosis (experimental group (EG); n = 22) and a neurotypical comparison group (CG; n = 17) without any reported neurodevelopmental condition. The CG consisted of a group of high-potential students from an information technology college in Upper Austria. Individuals with an autism diagnosis were recruited via a labor-integration project for autistic people. No technical affinity or prior knowledge was required. The only requirements were an interest in participating in the study and a commitment to being present at the research institute on two separate assessment days. All participants had to be adults (> 18 years). After exclusion of three participants from the EG – due to a technical error on one of the administration days, their respective annotation-parameter data sets had been lost – the final EG sample comprised 19 participants. Participant demographic information (age, gender, education and occupation) was obtained using a demographic questionnaire. The EG included 5 cisgender females, 11 cisgender males and 3 nonbinary individuals. The mean age of the EG was 24.32 (SD = 6.93). The CG consisted of 3 cisgender females and 14 cisgender males, with a mean age of 19.44 (SD = 1.41). Based on the demographic questionnaire, the participants were assigned an occupational and an educational status. The descriptors for occupational status were “in work”, “in training/qualification program” and “unemployed”. The levels for educational status were “educational level below high-school diploma” (all individuals in the CG were in in the final grade (grade 13) of a technical college of higher vocational education (i.e., an upper level secondary technical and vocational school that awards the equivalent of a high-school diploma upon graduation, which qualifies for university entrance in Austria. Demographic information and group descriptions are presented in Table 1 . Table 1 Descriptive statistics of experimental and comparison groups Group EG CG n % within group n % within group total 19 17 gender male 11 57.89 14 82.35 female 5 26.32 3 17.65 nonbinary 3 15.79 0 0.00 occupational status in work 5 26.30 0 0.00 in training/qualification 5 26.30 17 100.00 unemployed 9 47.40 0 0.00 educational status below high-school diploma 9 47.37 17 100.00 high-school diploma or higher 8 42.11 0 0.00 missing 2 10.52 0 0.00 age* (mean; SD) 24.32 (6.93) 19.44 (1.41) Notes : *…. p-value ≤ .05 (independent sample t-test) 2.3 Procedures Data acquisition differed between the two study groups because study settings were adapted to make them autism-friendly for the experimental group. For the EG, written information about the research (purpose, details about the process, general conditions, anticipated time and travel commitments, planned tests and measurements) was provided in advance via email in accordance with recommendations from the “Participant-Friendly Clinical Trials in Autism Guidebook” (Pisani et al., 2022 ). Participants on the AS were welcomed individually and introduced to the testing environment. They were told that aids such as noise-canceling headphones and sunglasses could be used whenever needed, adjustment of light and temperature in the assessment room were offered, and the IT setup was adapted to individual preferences (e.g., screen adjustments, mouse speed). All instructions were presented in writing and had been tested for clarity by an autistic focus group. Short breaks outside the assessment environment (e.g., common room, garden) were possible on request. It was made explicit to the participants that they could end participation without giving a reason at any point. At the first appointment, an additional verbal introduction to the study with information on study participation and payment was given. The assessment was carried out in small groups of three or four participants on the premises of the research institute in the course of two different appointments, each of which lasted about 4½ hours. Individual neuropsychological assessments took place in a separate room. The duration of the individual assessments and questionnaires for the experimental group (EG) was on average about two hours per appointment. CG data was collected within one appointment with the whole group in their college classroom, which took about three hours. The CG was given a detailed explanation of the research purpose and the conditions for participation, and the data collected were limited to informed consent, demographic and feedback questionnaires and the annotation parameters. The visual data-annotation tasks were assessed using CVAT (Computer Vision Annotation Tool, https://www.cvat.ai/ ), a professional web-based image-annotation tool. Each annotation task (Level 1 to 6) included a certain amount of images. Scores were calculated using the “ground truth” provided by expert annotators. Figure 1 illustrates the CVAT software used. 2.4 Measures The study’s measures can be subdivided into three categories: a) visual data-annotation parameters, b) neurocognitive performance and c) self-perception in daily life. Figure 2 summarizes all measures used. 2.4.1 Visual data-annotation parameters The annotation parameters were obtained using an image data annotation software that was developed specifically for job-training purposes. Three types of annotation tasks had to be performed: 1) bounding-box annotation, 2) labeling and 3) a combination thereof. Bounding-box annotation was achieved by manually drawing rectangles around the object(s) of interest at the pixel level, and the Intersection over Union (IoU) score was used to quantify annotation accuracy. The IoU score is useful for evaluating how well a predicted area of interest (e.g., a tumor in an MRI scan) aligns with the ground truth (the actual tumor boundary marked by experts). Labeling was done by defining a textual label (e.g., car, truck) for a whole image, and accuracy was measured by calculating the F1 score . This score combines Precision and Recall into a single value to give a balanced measure of a test’s accuracy; it is also used to evaluate the accuracy of diagnostic tests in which it is crucial to strike a balance between correctly identifying patients with a particular condition (recall) and not mislabeling healthy patients as sick (precision). The third annotation task combined labeling and bounding-box annotation. Here, multiple objects had to be annotated using bounding boxes, and to each object of interest a dedicated label had to be assigned. Annotation performance was computed per object and using F1 and IoU scores together for the whole image to ensure comprehensive evaluation. This combination of scores is particularly valuable for images that contain multiple objects, where both precise positioning and a balanced detection rate are crucial. Figure 3 illustrates the scores for the three annotation tasks used in this study. 2.4.2 Neurocognitive performance Non-verbal IQ and short-term memory were assessed using the Reynolds Intellectual Assessment Scales and Screening (RIAS; (Taylor et al., 2006 ). The test is normed for the age range 3;0 to 99;11, has an internal consistency of .81 to .95, and can be completed in approximately 30 minutes. This was the only test procedure that was carried out in a one-on-one test setting in a separate room. The other tests were carried out with groups of 4 to 6 participants. The following test instruments were used: Visual attention performance (accurate and quick discrimination of similar visual symbols) was tested with the Frankfurter Aufmerksamkeits-Inventar 2 (English: Frankfurt Attention Inventory; FAIR-2; (Petermann, 2011 )), which is normed for 9 to 85 years, with a split-half reliability of over .90. It takes approximately 10 minutes to administer. Reading comprehension was assessed using Ein Leseverständnistest für Erst- bis Siebtklässler 2 (English: A reading comprehension test for first- to seventh-graders; ELFE II;(Lenhard et al., 2017 )) at the word, sentence and text levels. This test is normed from the first to the seventh grade, and the most recent standard value tables were used in this study. The odd-even-split reliability for this procedure is within the range from r = .87 to .98. The test takes approximately 20 minutes to complete. Self-perceptions in everyday life (questionnaires; see next section) 2.4.3 Self-perceptions Subjectively perceived stress levels in everyday life were assessed by the Perceived Stress Scale (PSS-10; (Schneider et al., 2020 )) with ten items related to the frequency of stress-associated conditions, using a 5-point scale from 1 (never) to 5 (very often). This procedure is intended for 18- to 65-year-olds and has – according to the test authors – good reliability (α = .85). Self-perceived state of disability in life was assessed with the WHO Disability Assessment Schedule 2.0 (WHODAS 2.0; (Federici et al., 2017 )) using a 5-point scale from "none" to "very severe/not possible". All 36 items focus on everyday functionality and have good reliability scores of α = 0.70–0.97. We created (but did not validate) a questionnaire for the self-assessment of job-related personality skills . This questionnaire comprises 20 items and focuses on the following skills: memory, resilience to monotony, on-task accuracy, detail orientation, experience in IT, endurance, decision-making, need for structure, searching for information, and conscientiousness. All items were rated on a 4-point scale from 0 (not true) to 3 (absolutely true) (see supplementary materials). Additionally, we constructed a questionnaire that comprises 13 items on a 4-point scale (see supplementary materials), which was handed out to gain individual feedback on the study procedures and annotation tasks from participants on the autism spectrum. 2.5 Ethical considerations and data security This study was approved by the Johannes Kepler University Linz Ethics Committee. Participants received written information prior to participation and gave written informed consent. To pseudonymize the data, identifiers were removed and replaced with participant identification numbers (e.g., AS01). Data obtained from the neuropsychological measures were stored securely at the Research Institute for Developmental Medicine and on a secure server at the Hospital of St. John of God in Linz in accordance with data protection law. The annotation data was saved anonymously on a server of the Software Competence Center Hagenberg GmbH. The researchers conducting the assessments were sensitive to the needs of individuals on the AS and, throughout the assessment process, tailored the environment to these participants’ requirements (as far as possible). Following the assessments, participants received individual feedback on their annotation performance and the results of the neuropsychological assessments. 2.6 Statistical analysis This study used a between-subject design to measure differences between the experimental and comparison groups. All measures of interest (IoU and F1) were either assessed within one session (CG) or on two assessment days within one calendar week (EG). All data analyses were conducted with SPSS 29 and JAMOVI statistical software. Descriptive values were calculated and reported for all demographic, neurocognitive and questionnaire measures. Paired sample tests were performed to separately analyze differences in participants’ dimensions. Bivariate Pearson correlations between individual characteristics and annotation performance measures were derived. Cohen’s d values were used to report on effect sizes. Scatter-plot and bar diagrams are provided. 3 Results 3.1 Visual Data Annotation Independent sample t-tests revealed no significant differences in visual annotation tasks between EG and CG according to the F1 (t(34) = 1.59; p = .12) and IoU (t(34) = .95; p = .35) scores. However, a trend towards a moderately higher > F1 score annotation performance related to Cohen’s d (ES = .53) was found for the EG (see Table 2). The correlation between F1 and IoU scores was highly significant (r = .73; p < .001) Table 2 Annotation performances for EG and CG. EG n = 19 CG n = 17 t-Test Cohen’s d Mean SD Mean SD p-value ES F1 score .71 .14 .62 .22 .12 .53 IoU score .79 .05 .77 .07 .35 .32 Figure 4. Visualized annotation parameters for EG and CG. The left diagram illustrates F1- and right diagram IoU-annotation mean scores for both groups with no significant differences. 3.2 EG characteristics and correlations with annotation performances Our analysis investigated correlations between EG characteristics and F1 annotation scores. Pearson correlations revealed a non-significant negative correlation between non-verbal IQ and F1 annotation score (r=-.38; p = .105) and a non-significant positive correlation between reading skills and F1 annotation score (r = .37; p = .115). No other statistically relevant correlations between EG characteristics and F1 annotation score were found. EG mean characteristics and correlations with F1 score are listed in Table 3. Figure 5 illustrates individual associations between participant characteristics and F1 score. Table 3 EG characteristics and correlations with annotation performance EG Correlation F1 score M SD Pearson r p Non-verbal IQ (quotient-score) 1 105.63 7.78 − .38 .105 Visual attention (standardized test score) 2 60.84 27.73 .07 .773 Reading comprehension (raw score) 3 23.47 3.32 .37 .115 Stress (raw score) 4 33.20 7.38 − .21 .378 Disability in life (test score) 5 2.18 0.76 − .30 .207 Notes: 1 … RIAS; 2 … FAIR-2 Quality-score; 3 … ELFE-II text-level; 4 … PSS-10; 5 … WHODAS 2.0 3.3 Job-related attitudes of EG and CG Exploratory analysis showed that resilience to monotony was associated between all participants (EG and CG) and annotation performances F1 and IoU (r = .34, p = .044; r = .30, p = .077, respectively; “Monotonous or repetitive tasks are not a problem for me.” ). In addition, a trend towards a positive association between conscientiousness and annotation performances F1 and IoU was found (r = .30, p = .081; r = .28, p = .105, respectively; “When I finish something, I consider what to do next or I ask relevant people.” ). Welch’s t-tests showed significant differences in self-perceived job-related attitudes between EG and CG: EG higher task accuracy (t(32.2) = 3.52, p = .001, ES = 1.17; “I enjoy completing tasks in an accurate manner.” ), CG higher degree of decisiveness (t(34)=-3.39, p = .002, ES=-1.14; ”I find it easy to make decisions.” ) and more experience in IT (t(29.6)=-2.80, p = .009, ES=-.92; “I have extensive experience with digital technologies (PCs, internet, software, etc.)” ). Non-significant trends of differences between EG and CG were found: EG greater need for structure (t(33.4) = 1.80, p = .081, ES = .60; “When completing a multi-step task, a structuring aid (e.g., visual, from other people) would help me.” ), EG likes searching for information more (t(31.8) = 1.91, p = .066, ES = .64; “I enjoy searching for new information and explanations.” ). 3.4 Qualitative feedback on annotation experiences After test administration, all participants from the EG were invited to give feedback on their experiences of the study and of annotation (N = 17). Overall, the participants felt very comfortable and well-informed about the study and reported that the general conditions during the tests were very accommodating to their needs. Wording and procedures were interpreted as clear, and the majority of participants (88%) would participate again. Twelve out of 17 participants wished to “… learn more about annotation in the future.” and 10 participants “…could imagine working in the field of annotation.”. 4 Discussion As the digital revolution progresses (Boston Consulting Group, 2021 ), the field of AI is beginning to exert a significant influence on the global job market (Milanez, 2023 ). This development also offers specific groups – particularly those who face challenges in finding and maintaining suitable employment – novel opportunities for employment (Davies et al., 2024 ). This study aimed to investigate the extent to which this applies to autistic individuals. Over the past decade, data annotation – a prerequisite for the training, tuning and evaluation of AI systems – has emerged as a new line of work, which is expected to grow in the coming years (World Economic Forum, 2023 ). Data scientists and IT enterprises pursue AI research in Western countries, whereas data annotation work is conducted mainly in low-income countries (Smart et al., 2024 ). This leaves untapped the potential that local – and, in particular, neurodivergent – individuals offer for this line of work. The present study investigated the annotation performance of individuals on the autism spectrum with normal intelligence in simple visual data annotation tasks (i.e., assignment of labels and placement of bounding boxes around objects). Participants were recruited through collaboration with a labor-integration project for autistic individuals. This research design incorporated a neurotypical, IT-trained comparison group. As part of a participatory approach, fair and respectful working conditions and compensation were ensured for all participants. In answer to Research Question 1, individuals on the AS with normal intelligence showed annotation quality equal to that of an IT-trained neurotypical comparison group. This finding is in line with other single case reports (Garrison et al., 2023 ; Schenkenfelder et al., 2024 ). Our study sought to gain a deeper understanding of intraindividual characteristics that may affect accuracy in visual data annotation, and thus non-verbal IQ, visual attention, reading comprehension, perceived stress and disability in daily life were assessed for the experimental group. Contrary to the seemingly plausible assumption that higher non-verbal IQ scores would be associated with higher visual annotation performance, our analysis indicated the opposite trend. A possible explanation consistent with feedback from participants is that autistic individuals who show higher non-verbal intellectual function may not be sufficiently motivated and challenged by annotations tasks with low complexity. In line with reports by trainees (Schenkenfelder et al., 2024 ), a positive but non-significant association was observed between reading comprehension and annotation performance. Self-perceived level of stress and disability demonstrated negative correlations with annotation performance, without attaining significance level. Personal characteristics over all participants such as resilience to monotony showed positive and conscientiousness a non-significant positive correlation with annotation performance. 5 Implications and Future Research Findings of this study support the promotion of annotation work in Europe in order to reduce biases rooted in culture or geography as identified, for instance, by Mark Díaz et al. ( 2022 ). The introduction of annotation work in Austria could mark a starting point, with future expansions to other high-income countries. Visual data annotation tasks are well suited to the target group described and do not seem to require specific strengths in nonverbal cognitive functioning or attention, although a minimum reading level might be beneficial, even though significance level was not attained in the small study sample. Another important skill for high-quality annotation results may be conscientiousness of the annotators. Further, it may be beneficial to have a fundamental understanding of and motivation for the need for accuracy in job-related tasks, an attitude that was more apparent in the EG in comparison to the CG in the limited sample. Annotation can be regarded as a process of continuous decision-making, a skill mentioned more by the CG sample. Individuals on the AS were granted specific working conditions, such as working in small groups, opportunities to relax, low-distraction environments, clear and short written and verbal instructions, and visualized guidelines. In a broader implementation, individuals with specific workplace preferences—such as working alone or in a noise-free environment—could be offered adapted working conditions, including working remotely, thereby reducing unemployment in this target group (Espelöer et al., 2023 ; Lorenz et al., 2016 ; Zimmermann & Falkner, 2018 ). The Conversation of Resources (COR) theory (Hobfoll et al., 2018 ) provides a helpful framework for better describing and understanding (stressful) job experiences in the workplace. A recent study (Tomczak et al., 2021 ) involved autistic individuals and applied the COR-theory to recruitment, selection, onboarding and job retention. The authors suggested inclusive, communication-based strategies, such as: “request to solve a given problem instead of a typical job advertisement”, “verbal instructions, short and to the point”, “practical skills tests, including gamification-based solutions”, “provide support of buddy, mentor, job coach”, “using onboarding checklists, manuals and guides”, “meetings in small groups”, or “non-direct, electronically mediated communication” (Tomczak et al., 2021 ). Such strategies, alongside some adaptations to work-related framework conditions (low-stimulus workspaces or other individual adjustments, Tomczak et al., 2021 ) could support communication between neurodiverse individuals with social communication difficulties and their colleagues and supervisors, thereby reducing barriers. Training sessions could be introduced for the entire team to ensure that the whole environment rather than just the immediate point of contact collaborates with and includes individual(s) with ASD. These adaptations should be seen not as exclusionary, but as promoting inclusion by respecting individual needs and integrating additional people into the labor market. Such a flexible work environment may also appeal to other neurodiverse individuals who prefer reduced sensory stimuli, for instance, individuals with attention-deficit/hyperactivity disorder (ADHD). Future studies should include larger samples and focus on optimized workplace designs for neurodivergent individuals, with the main aim of recognizing individuals‘ unique needs and addressing them effectively to improve their prospects in the labor market and their integration into society. Long-lasting, stable employment plays a crucial role in the life and well-being of individuals (Lord et al., 2020). This research project highlights that individuals with ASD can perform valuable work under well-adapted workplace conditions. 5.1 Limitations A clear limitation of this study is the overall small sample size (N = 39), which makes statistical interpretation difficult. However, the analyses provide relevant indications. Since test duration differed between EG and CG, the numbers of images annotated by each group are not directly comparable. Due to the varying levels, which the web-based image-annotation tool offered (level 1 to 6), and the increasing complexity of all images, which had to be annotated during the testing phase, the total amount of annotated images cannot simply be extrapolated. However, importantly the key-component of visual data annotation – annotation-quality – was quantifiable and statistically comparable in this trial. An expanded study in terms of sample size and temporal design would be necessary if the goal were to extend the use of statistical methods to enable the identification of group differences through significance tests for the amount of annotated data. Additionally, due to statistical constraints, the exclusion of outliers (high performers and very low performers) was neither feasible nor meaningful. Another critical aspect is the assumption of neurotypicality in the CG. The group was categorized as neurotypical because no diagnoses were reported. However, neurodivergence extends beyond the autism spectrum and includes learning disorders, such as dyslexia and dyscalculia, and ADHD. Statistically, it is unlikely that the control group was truly homogeneously neurotypical (which we assumed in this study due to a lack of information). This limitation was accepted, as a strict definition of neurodivergence was not the focus of this research. Declarations Acknowledgements The authors sincerely thank all participants for their participation and willingness to share their thoughts and experiences during the study. Many thanks to all autistic peers and co-workers for their support and help in developing an autism-friendly study procedure. The authors also thank all students, teachers and the principal of the Traun Technical College for their support and participation in this study. Contributions DL: Conceptualization, Funding acquisition, Investigation, Methodology, Statistical analysis, Data curation, Project administration, Writing, Editing, Review KK: Funding acquisition, Investigation, Methodology, Project administration, Writing, Editing, Review LF: Technical and Software Development, Methodology, Writing, Review BS: Technical and Software Development, Writing, Review SW: Autistic peer co-researcher, Methodology, Review MM: Investigation , Formal analysis, Writing, Review DH: Conceptualization, Methodology, Supervision, Writing, Editing, Review Funding : This research was funded by Open Innovation in Science Center of the Ludwig Boltzmann Gesellschaft. The research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The research reported in this paper has been partly funded by the Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation and Technology (BMK), the Federal Ministry for Labor and Economy (BMAW), and the State of Upper Austria in the frame of the SCCH competence center INTEGRATE [(FFG grant no. 892418)] in the COMET - Competence Centers for Excellent Technologies Programme managed by Austrian Research Promotion Agency FFG. Institutional Review Board Statement : The study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the JKU Ethics Committee. Informed Consent Statement : Informed consent was obtained from the participants. Data Availability Statement : The data presented in this study are available on request from the corresponding author. The data are not publicly available due to data protection issues. Conflicts of Interest : The authors declare no conflict of interest. The funder had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. References Adam, M. T. P., Gimpel, H., Maedche, A., & Riedl, R. (2017). Design Blueprint for Stress-Sensitive Adaptive Enterprise Systems. Business & Information Systems Engineering , 59 (4), 277–291. https://doi.org/10.1007/s12599-016-0451-3 Baranger, A [Aurélie] (2019). State of play of employment of people on the autism spectrum in Europe: barriers, good practices and trends. Autism Europe. Committee on Employment and Social Affairs of the European. https://web.archive.org/web/20231027172507/https://www.autismeurope.org/wp-content/uploads/2019/11/presentation_employment_autism_final2.pptx.pdf Baron-Cohen, S [Simon] (2009). Autism: The empathizing-systemizing (E-S) theory. Annals of the New York Academy of Sciences , 1156 , 68–80. https://doi.org/10.1111/j.1749-6632.2009.04467.x Baron-Cohen, S [Simon], Ashwin, E., Ashwin, C., Tavassoli, T., & Chakrabarti, B. (2009). Talent in autism: Hyper-systemizing, hyper-attention to detail and sensory hypersensitivity. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences , 364 (1522), 1377–1383. https://doi.org/10.1098/rstb.2008.0337 Boston Consulting Group. (2021). The Future of Jobs in the Era of AI . https://web-assets.bcg.com/ec/bc/da7341af41358367d26db742eb6c/bcg-the-future-of-jobs-in-the-era-of-ai-may-2021-r.pdf Bury, S. M., Hedley, D., & Uljarević, M. (2021). Restricted, Repetitive Behaviours and Interests in the Workplace: Barriers, Advantages, and an Individual Difference Approach to Autism Employment. In E. Gal & N. Yirmiya (Eds.), Repetitive and Restricted Behaviors and Interests in Autism Spectrum Disorders: From Neurobiology to Behavior (pp. 253–270). Springer International Publishing. https://doi.org/10.1007/978-3-030-66445-9_15 Davies, J., Heasman, B., Livesey, A., Walker, A., Pellicano, E., & Remington, A. (2023). Access to employment: A comparison of autistic, neurodivergent and neurotypical adults’ experiences of hiring processes in the United Kingdom. Autism , 27 (6), 1746–1763. Davies, J., Romualdez, A. M., Pellicano, E., & Remington, A. (2024). Career progression for autistic people: A scoping review. Autism , 13623613241236110. Espelöer, J., Proft, J., Falter-Wagner, C. M., & Vogeley, K. (2023). Alarmingly large unemployment gap despite of above-average education in adults with ASD without intellectual disability in Germany: A cross-sectional study. European Archives of Psychiatry and Clinical Neuroscience , 273 (3), 731–738. Federici, S., Bracalenti, M., Meloni, F., & Luciano, J. V. (2017). World Health Organization disability assessment schedule 2.0: An international systematic review. Disability and Rehabilitation , 39 (23), 2347–2380. Garrison, E., Singh, D., Hantula, D., Tincani, M., Nosek, J., Hong, S. R., Dragut, E., & Vucetic, S. (2023). Understanding the experience of neurodivergent workers in image and text data annotation. Computers in Human Behavior Reports , 11 , 100318. Gray, M., & Suri, S. (2019). Ghost Work: How Amazon, Google, and Uber Are Creating a New Global Underclass (1st ed.). Houghton Mifflin Harcourt Publishing Company. https://ebookcentral.proquest.com/lib/kxp/detail.action?docID=29437679 Hara, K., & Bigham, J. P. (2017). Introducing People with ASD to Crowd Work. In A. Hurst, L. Findlater, & M. R. Morris (Eds.), Proceedings of the 19th International ACM SIGACCESS Conference on Computers and Accessibility (pp. 42–51). ACM. https://doi.org/10.1145/3132525.3132544 Hobfoll, S. E., Halbesleben, J., Neveu, J.‑P., & Westman, M. (2018). Conservation of resources in the organizational context: The reality of resources and their consequences. Annual Review of Organizational Psychology and Organizational Behavior , 5 , 103–128. Jolliffe, T., & Baron-Cohen, S [S.] (2001). A test of central coherence theory: Can adults with high-functioning autism or Asperger syndrome integrate fragments of an object? Cognitive Neuropsychiatry , 6 (3), 193–216. https://doi.org/10.1080/13546800042000124 Lenhard, W., Lenhard, A., & Schneider, W. (2017). Ein Leseverständnistest für Erst- bis Siebtklässler (ELFE II). Lorenz, T., Frischling, C., Cuadros, R., & Heinitz, K. (2016). Autism and overcoming job barriers: Comparing job-related barriers and possible solutions in and outside of autism-specific employment. PloS One , 11 (1), e0147040. Mark Díaz, Ian Kivlichan, Rachel Rosen, Dylan Baker, Razvan Amironesei, Vinodkumar Prabhakaran, & Emily Denton. (2022). CrowdWorkSheets: Accounting for Individual and Collective Identities Underlying Crowdsourced Dataset Annotation. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (pp. 2342–2351). Association for Computing Machinery. https://doi.org/10.1145/3531146.3534647 Milanez, A. (2023). The impact of AIL on the workplace: Evidence from OECD case studies of AI implementation. OECD Social, Employment and Migration Working Papers (289), 1–115. https://doi.org/10.1787/2247ce58-en Nicolaidis, C., Raymaker, D., Kapp, S. K., Baggs, A., Ashkenazy, E., McDonald, K., Weiner, M., Maslak, J., Hunter, M., & Joyce, A. (2019). The AASPIRE practice-based guidelines for the inclusion of autistic adults in research as co-researchers and study participants. Autism , 23 (8), 2007–2019. https://doi.org/10.1177/1362361319830523 Perkowski, M., Oksztulski, M., Zoń, W., & Kaczyńska, I. (2024). Educated Persons on the Autism Spectrum in the Labour Market: Status Quo and Prospects . Temida 2 with the cooperation of the Law and Partnership Foundation … Perrigo, B. (2023). Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic . https://time.com/6247678/openai-chatgpt-kenya-workers/ Petermann, F. (2011). Frankfurter Aufmerksamkeits- Inventar 2 (FAIR-2). Zeitschrift Für Psychiatrie, Psychologie Und Psychotherapie , 59. https://doi.org/10.1024/1661-4747/a000088 Pisani, G., Averius, C., Baranger, A [A.], Garcia, J. M., Mazzoni, A., Thomander Neerland, H., Schneider, T., Tepper Singer, A., & Williams, Z. J. (2022). Guidebook for participant-friendly clinical trials in Autism: for investigators, researchers, clinical trials staff, and the autism community . https://www.ieepo.com/content/dam/websites/ieepo/2022/resources/learn-library/2022-ieepo-materials/Clinical%20Trials%20in%20Autism%20Guidebook.pdf Schenkenfelder, B., Brandstätter, U., Fischer, L., Ramler, R., Laister, D., Hartl, M., & Wurm, M. (2024). Responsible AI Engineering: The Case of an Inclusive Image Annotation Team in a Global Technology Company. In Proceedings of the 2nd International Workshop on Responsible AI Engineering (pp. 8–15). Association for Computing Machinery. https://doi.org/10.1145/3643691.3648583 Schneider, E. E., Schönfelder, S., Domke-Wolf, M., & Wessa, M. (2020). Measuring stress in clinical and nonclinical subjects using a German adaptation of the Perceived Stress Scale. International Journal of Clinical and Health Psychology : IJCHP , 20 (2), 173–181. https://doi.org/10.1016/j.ijchp.2020.03.004 Smart, A., Wang, D., Monk, E., Díaz, M., Kasirzadeh, A., van Liemt, E., & Schmer-Galunder, S. (2024, February 9). Discipline and Label: A WEIRD Genealogy and Social Theory of Data Annotation . http://arxiv.org/pdf/2402.06811 Solomon, C. (2020). Autism and Employment: Implications for Employers and Adults with ASD. Journal of Autism and Developmental Disorders , 50 (11), 4209–4217. https://doi.org/10.1007/s10803-020-04537-w Tan, R., & Cabato, R. (2023). Behind the AI boom, an army of overseas workers in ‘digital sweatshops’ . Taylor, A., Reynolds, C., & Kamphaus, R. (2006). The Reynolds Intellectual Assessment Scales (RIAS) and Assessment of Intellectual Giftedness. Gifted Education International , 21. https://doi.org/10.1177/026142940602100305 Tomczak, M. T., Szulc, J. M., & Szczerska, M. (2021). Inclusive communication model supporting the employment cycle of individuals with autism spectrum disorders. International Journal of Environmental Research and Public Health , 18 (9), 4696. Wang, D., Prabhat, S., & Sambasivan, N. (2022). Whose AI Dream? In search of the aspiration in data annotation. CHI Conference on Human Factors in Computing Systems , 1–16. https://doi.org/10.1145/3491102.3502121 World Economic Forum. (2023). AI: 3 ways articicial intelligence is changing the future of work . https://www.weforum.org/agenda/2023/08/ai-artificial-intelligence-changing-the-future-of-work-jobs/ World Health Organization. (2022). International Classification of Diseases Eleventh Revision (ICD-11) . https://icd.who.int/browse/2024-01/mms/en Zarifhonarvar, A. (2023). Economics of ChatGPT: a labor market view on the occupational impact of artificial intelligence. Journal of Electronic Business & Digital Economics , 3 (2), 100–116. https://doi.org/10.1108/JEBDE-10-2023-0021 Zimmermann, A., & Falkner, G. (2018). Inklusion von Menschen mit besonderen Bedürfnissen in den Arbeitsprozess. Personalmanagement: Internationale Perspektiven Und Implikationen Für Die Praxis , 133–156. Additional Declarations No competing interests reported. Supplementary Files UAISSM202505.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 02 Jul, 2025 Editor assigned by journal 02 Jul, 2025 Submission checks completed at journal 14 May, 2025 First submitted to journal 12 May, 2025 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-6646902","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":479832556,"identity":"5fac62c3-3cd2-4ec7-ad76-d2b4f26257f9","order_by":0,"name":"Dominik Laister","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYDCCAyBkkMDAx8DYAOLLMfYQq4UNpAXIMYZoScCvBayCDcpJbOAhoIXveHfi4YKCNAY29sPNnz9U3Etv7jljwPDzB24tkmfObjg8wyCHgY0nsU3iwJni3MbeHgPGHjy2GNzI3XCYx6AC6LDENoaDbQm5jf08Bgw8RGnhf9j84eC/hHRGoBbGP4S1AB0mkdggcbAhIYER6DBmfLaA/cJjkMbDJvGwTeLMsQTDxp5jBYdl0nBr4Tveu/kzz59kOX7+9McfKmoS5A17kjc+fGODWwsM8MBZhg2wyCIayJOmfBSMglEwCkYAAADPJVaSyjX0XwAAAABJRU5ErkJggg==","orcid":"","institution":"Johannes Kepler University Linz","correspondingAuthor":true,"prefix":"","firstName":"Dominik","middleName":"","lastName":"Laister","suffix":""},{"id":479832557,"identity":"ac09d80d-10be-4714-ba28-71013731a778","order_by":1,"name":"Kristina Köck","email":"","orcid":"","institution":"Johannes Kepler University Linz","correspondingAuthor":false,"prefix":"","firstName":"Kristina","middleName":"","lastName":"Köck","suffix":""},{"id":479832558,"identity":"bd49be80-9a87-4ec1-98fb-381db461b7ee","order_by":2,"name":"Lukas Fischer","email":"","orcid":"","institution":"Software Competence Center Hagenberg GmbH (SCCH)","correspondingAuthor":false,"prefix":"","firstName":"Lukas","middleName":"","lastName":"Fischer","suffix":""},{"id":479832559,"identity":"e1611328-8337-4007-a969-d8a652f962ab","order_by":3,"name":"Bernhard Schenkenfelder","email":"","orcid":"","institution":"Software Competence Center Hagenberg GmbH (SCCH)","correspondingAuthor":false,"prefix":"","firstName":"Bernhard","middleName":"","lastName":"Schenkenfelder","suffix":""},{"id":479832560,"identity":"55a09ecf-8149-4a75-b6ae-8ceb2616564e","order_by":4,"name":"Samuel Wiesinger","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Samuel","middleName":"","lastName":"Wiesinger","suffix":""},{"id":479832561,"identity":"76c13c82-7449-40aa-9602-23b9f662a33e","order_by":5,"name":"Miriam Massoumzadeh","email":"","orcid":"","institution":"Johannes Kepler University Linz","correspondingAuthor":false,"prefix":"","firstName":"Miriam","middleName":"","lastName":"Massoumzadeh","suffix":""},{"id":479832562,"identity":"0b9d5756-ff0b-4499-a445-8a90aea3b39b","order_by":6,"name":"Daniel Holzinger","email":"","orcid":"","institution":"Johannes Kepler University Linz","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Holzinger","suffix":""}],"badges":[],"createdAt":"2025-05-12 13:08:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6646902/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6646902/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86139522,"identity":"d0facf62-2cd5-49d2-8030-60004b5d2332","added_by":"auto","created_at":"2025-07-07 08:17:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":282957,"visible":true,"origin":"","legend":"\u003cp\u003eScreenshot of the “Computer Vision Annotation Tool” (CVAT)\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/55c3e25f0d9c180a96e805bf.png"},{"id":86139527,"identity":"564549c5-1f0b-43ef-ba6c-6c3ad5150852","added_by":"auto","created_at":"2025-07-07 08:17:07","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":18220,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of study measures. Neurocognitive Performance scores, Stress \u0026amp; Coping and Perceived Disability were assessed only for the EG\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/755b7669e8b68b4876c7968a.png"},{"id":86142069,"identity":"6a3c6d69-5561-46eb-9191-3f06d6b04a1c","added_by":"auto","created_at":"2025-07-07 08:33:07","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":165529,"visible":true,"origin":"","legend":"\u003cp\u003eAnnotation types with respective metrics used in this study\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/0f15d768514c1fc9e7a7fd87.png"},{"id":86139540,"identity":"93f2e0dc-094e-4510-8572-3b1b850aae30","added_by":"auto","created_at":"2025-07-07 08:17:07","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":116103,"visible":true,"origin":"","legend":"\u003cp\u003eVisualized annotation parameters for EG and CG. The left diagram illustrates F1- and right diagram IoU-annotation mean scores for both groups with no significant differences.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/7525a2dfbc102eea414e90ed.png"},{"id":86139541,"identity":"df1b799c-221b-437f-b4ed-7c4ce46ebb48","added_by":"auto","created_at":"2025-07-07 08:17:07","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":303971,"visible":true,"origin":"","legend":"\u003cp\u003eAssociations between participant characteristics (IQ, visual attention, reading comprehension, stress, disability in life) and F1 annotation score.\u003cbr\u003e\n \u003cu\u003e\u003cem\u003eNote\u003c/em\u003e\u003c/u\u003e\u003cem\u003e: RIAS_NIX … Nonverbal IQ; FAIR_Q_pr … Visual annotation performance; ELFE_text_corr… reading comprehension; PSS_Total_sc … PSS-10 stress score; WHO_MEAN_Total … WHODAS 2.0 Total score\u003c/em\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/87a666bb36a64aca00acff21.png"},{"id":86143385,"identity":"27ea1477-281e-4be0-be78-a72a6a3f4ce3","added_by":"auto","created_at":"2025-07-07 08:41:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1819724,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/ca910576-aa82-4e85-a426-ba7ab510c497.pdf"},{"id":86139521,"identity":"e095f46c-c110-4091-bf4d-c0e04f63a5ab","added_by":"auto","created_at":"2025-07-07 08:17:07","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":19393,"visible":true,"origin":"","legend":"","description":"","filename":"UAISSM202505.docx","url":"https://assets-eu.researchsquare.com/files/rs-6646902/v1/0500964eccac432db274b8f0.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Empowering Neurodiversity - Analyzing the Potential of Visual Data Annotation as Employment for Autistic Individuals","fulltext":[{"header":"1 Introduction","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003e1.1 Artificial Intelligence and Annotation\u003c/h2\u003e \u003cp\u003eThe digital revolution, driven by Artificial Intelligence (AI) and automation, is reshaping the employment landscape around the world, bringing both challenges and opportunities. While this transformation means that people are displaced from certain routine work and manual jobs, it is also creating opportunities in highly skilled roles such as data scientist, AI specialist, and other tech-focused positions (Milanez, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; World Economic Forum, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). ChatGPT has been reported to have a great impact on the labor market, with two-thirds of occupations being fully or partially affected, and thus reskilling and upskilling will be required to fight long-term unemployment (Zarifhonarvar, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For instance, a report by the Boston Consulting Group (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) highlighted that by 2030 the United States and Germany could face labor shortages of up to 12.5\u0026nbsp;million and up to 2.5\u0026nbsp;million people, respectively, particularly in tech-related fields.\u003c/p\u003e \u003cp\u003eData annotation is essential to the training, tuning, and evaluation of AI systems, and while research is underway to increase the level of automation, it is currently a manual task. Visual annotation is the process of determining the location and/or content of objects in pictures for an AI to learn and later recognize automatically. Drawing a bounding box on a traffic image defines, for example, the boundaries of a particular road user, and assigning a label to that box defines its class (i.e., bicycle, car, or truck). In the case of medical images, the label can also be applied to the entire image to indicate, for instance, whether a brain tumor is malignant. Regardless of the details, it is only by annotating images, videos, text, audio, or other data that the foundation for AI pipelines can be built. Despite the importance of data annotation to Western countries, where most AI research is conducted, the Global South, where most annotation work is done, experiences a vast recognition gap (Smart et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This imbalance also manifests as exploitation through unfair salaries and as dehumanization through ghost work (Gray \u0026amp; Suri, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Perrigo, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Tan \u0026amp; Cabato, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). While it would be desirable for annotated data to be value-free, neutral and therefore generalizable, various studies have shown that some data are biased due to annotators\u0026lsquo; cultural, social, or geographical backgrounds. Whether their backgrounds are relevant depends on the type of data (Gray \u0026amp; Suri, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Perrigo, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Smart et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Wang et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The importance of such a context depends on the type of data. For example, road traffic data is more useful when annotated by annotators in this particular geographical region (Mark D\u0026iacute;az et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In a recent workshop paper, Schenkenfelder et al (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) shared promising results from a data annotation training initiative with neurodivergent individuals which led to formation of a successful, well-integrated team within an IT company. Supervisors of this inclusive team stated that \u0026ldquo;resilience to monotony\u0026rdquo; and \u0026ldquo;average reading comprehension skills\u0026rdquo; might be essential skills for trainees.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.2 Autism Spectrum Condition and the Labor Market\u003c/h2\u003e \u003cp\u003eTo address the problems laid out above, it would be desirable to move annotation work to the countries in which AI systems are developed and researched. As demand for (local) data annotation workers is increasing in these countries, opportunities to participate in the job market might open up for marginalized groups that are underrepresented in the labor market. Individuals on the Autism Spectrum (AS) with IQ scores in the normal range have been identified as being part of this marginalized societal group (Davies et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Hara \u0026amp; Bigham, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Solomon, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). According to the eleventh revision of the International Statistical Classification of Diseases and Related Health Problems (World Health Organization, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), Autism Spectrum Disorder (ASD) is classified as a neurodevelopmental condition characterized by persistent differences in social communication and interaction, alongside restricted and repetitive and inflexible patterns of behavior, interests, or activities. Autistic individuals show characteristic features or have difficulties in social reciprocity, nonverbal communication, or in developing and maintaining relationships. Individuals may also exhibit a strong preference for routines and show intense focus on specific topics or interests. Characteristics include atypical sensory responses, such as heightened sensitivity to light, sounds, textures, or smells. Individuals with ASD have varying intellectual abilities, ranging from intellectual disability to average and above-average intelligence. Severity varies widely, and the condition may coexist with other medical or psychiatric conditions (World Health Organization, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite their ability and desire to work, the employment rate for individuals with ASD in the United Kingdom is 29% (Davies et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and the average employment rate in Europe is as low as 10% (Baranger, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). While the UN Convention on the Rights of Persons with Disabilities (Hendriks, 2007) acknowledges the importance of professional participation and earned income for full societal inclusion, there remains a significant shortage of suitable employment opportunities for individuals with ASD within the broader workforce (Perkowski et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Where access to appropriate jobs in the general labor market is difficult, people with ASD must overcome substantial barriers to find employment, regardless of their level of education (Espel\u0026ouml;er et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Lorenz et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Zimmermann \u0026amp; Falkner, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The reasons for these difficulties are manifold, ranging from barriers in finding appropriate jobs to lack of employer awareness of the strengths and potential of this group. Entering the labor market is challenging, often because insufficient training and on-the-job support for autistic individuals and their working environment is provided (Lorenz et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Furthermore, higher stress levels during application, training and employment have been demonstrated (Davies et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). For instance, a supportive approach to job training, developed by Adam et al. (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), proposes design considerations for an adaptive enterprise system which \"assesses users' individual stress levels continuously and unobtrusively\". This would allow enterprises to automatically adapt human\u0026ndash;computer interfaces (e.g., information filtering, changes in screen color, information presentation mode) based on physiological measures (e.g., pupil dilation, heart rate, mouse pressure, muscle tension, pulse transit time, and skin conductance).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e1.3 Autistic features as strengths in digital job tasks?\u003c/h2\u003e \u003cp\u003eDespite \u0026ndash; or because of \u0026ndash; their condition, individuals on the AS may possess skills and unique characteristics that make them particularly suited to annotation work, as suggested by a small but growing body of research (Garrison et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Firstly, individuals diagnosed with ASD often demonstrate an exceptional ability to focus on specific details, which may enable them to excel at tasks that require meticulous attention (Baron-Cohen, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Baron-Cohen et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Jolliffe \u0026amp; Baron-Cohen, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2001\u003c/span\u003e). The frequently described ability to identify even subtle differences or inconsistencies within data sets makes them well-suited to labeling and categorizing complex data with accuracy (Baron-Cohen, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Secondly, autistic individuals often demonstrate an ability to recognize patterns, which may allow them to identify and follow consistent patterns faster \u0026ndash; a key competence in tasks such as categorizing data, tagging images, and annotating text for machine learning (Baron-Cohen, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Many people on the AS prefer structured environments with clear rules and routines. Data annotation typically involves well-defined tasks and guidelines, which provides a structured framework. Furthermore, the focus on routine and precision in autistic individuals can result in consistent high-quality work when performing tasks such as reviewing and labeling datasets, where accuracy is crucial. Lastly and importantly, many autistic individuals have a high tolerance for repetitive tasks that others may find tedious or under-stimulating(Bury et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) .\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e1.4 Aims of this study\u003c/h2\u003e \u003cp\u003eIn summary, there is a growing need for highly accurate annotation in the emerging field of AI, and a concurrent disadvantaged employment situation for individuals with ASD with promising qualities for digital tasks. This study analyzed how individuals on the autism spectrum perform in simple visual data-annotation tasks and sought to gain a deeper understanding of putative intraindividual characteristics, which may be associated with visual data-annotation performance.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResearch Question 1\u003c/b\u003e:\u003c/p\u003e \u003cp\u003eHow do autistic participants perform in visual data annotation in terms of correctness (labeling tasks) and accuracy (bounding boxes) compared to a neurotypical comparison group?\u003c/p\u003e \u003cp\u003e \u003cb\u003eResearch Question 2\u003c/b\u003e:\u003c/p\u003e \u003cp\u003eIs annotation performance in autistic participants associated with a) nonverbal IQ, b) visual attention performance and c) reading comprehension skills?\u003c/p\u003e \u003c/div\u003e"},{"header":"2 Methods","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1 Participatory research design\u003c/h2\u003e\n \u003cp\u003eThis study was designed and implemented as a participatory research project that included two investigators, one study-assistant and two autistic co-researchers and followed guidelines for participant-friendly research (Nicolaidis et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e; Pisani et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). A focus group of six autistic individuals supported the final selection and adaptation of the measures implemented. Four consecutive meetings were co-led by an autistic co-researcher and implementation staff. All pre-selected measures suggested by the investigators were reflected critically in each focus group session to gather and apply specific feedback on aspects of feasibility (length, comprehension, quality of instruction, suitability).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2 Participants\u003c/h2\u003e\n \u003cp\u003eThe total sample consisted of 39 participants, including individuals with a formal autism spectrum diagnosis (experimental group (EG); n\u0026thinsp;=\u0026thinsp;22) and a neurotypical comparison group (CG; n\u0026thinsp;=\u0026thinsp;17) without any reported neurodevelopmental condition. The CG consisted of a group of high-potential students from an information technology college in Upper Austria. Individuals with an autism diagnosis were recruited via a labor-integration project for autistic people. No technical affinity or prior knowledge was required. The only requirements were an interest in participating in the study and a commitment to being present at the research institute on two separate assessment days. All participants had to be adults (\u0026gt;\u0026thinsp;18 years). After exclusion of three participants from the EG \u0026ndash; due to a technical error on one of the administration days, their respective annotation-parameter data sets had been lost \u0026ndash; the final EG sample comprised 19 participants. Participant demographic information (age, gender, education and occupation) was obtained using a demographic questionnaire. The EG included 5 cisgender females, 11 cisgender males and 3 nonbinary individuals. The mean age of the EG was 24.32 (SD\u0026thinsp;=\u0026thinsp;6.93). The CG consisted of 3 cisgender females and 14 cisgender males, with a mean age of 19.44 (SD\u0026thinsp;=\u0026thinsp;1.41). Based on the demographic questionnaire, the participants were assigned an occupational and an educational status. The descriptors for occupational status were \u0026ldquo;in work\u0026rdquo;, \u0026ldquo;in training/qualification program\u0026rdquo; and \u0026ldquo;unemployed\u0026rdquo;. The levels for educational status were \u0026ldquo;educational level below high-school diploma\u0026rdquo; (all individuals in the CG were in in the final grade (grade 13) of a technical college of higher vocational education (i.e., an upper level secondary technical and vocational school that awards the equivalent of a high-school diploma upon graduation, which qualifies for university entrance in Austria. Demographic information and group descriptions are presented in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tab1\" border=\"1\" class=\"fr-table-selection-hover\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003e\u003cem\u003eDescriptive statistics of experimental and comparison groups\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eEG\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCG\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003en\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003e% within group\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003en\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003e% within\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003cem\u003egroup\u003c/em\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003etotal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003e19\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003e17\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003egender\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e57.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e82.35\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003efemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17.65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003enonbinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" rowspan=\"3\"\u003e\n \u003cp\u003eoccupational status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ein work\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ein training/qualification\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eunemployed\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e47.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eeducational status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ebelow high-school diploma\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e47.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ehigh-school diploma or higher\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e42.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emissing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eage* (mean; SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e24.32 (6.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e19.44 (1.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\"\u003e\u003cspan type=\"ItalicUnderline\" class=\"ItalicUnderline\" name=\"Emphasis\"\u003eNotes\u003c/span\u003e:\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\"\u003e\u003cem\u003e*\u0026hellip;. p-value\u0026thinsp;\u0026le;\u0026thinsp;.05 (independent sample t-test)\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3 Procedures\u003c/h2\u003e\n \u003cp\u003eData acquisition differed between the two study groups because study settings were adapted to make them autism-friendly for the experimental group. For the EG, written information about the research (purpose, details about the process, general conditions, anticipated time and travel commitments, planned tests and measurements) was provided in advance via email in accordance with recommendations from the \u0026ldquo;Participant-Friendly Clinical Trials in Autism Guidebook\u0026rdquo; (Pisani et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Participants on the AS were welcomed individually and introduced to the testing environment. They were told that aids such as noise-canceling headphones and sunglasses could be used whenever needed, adjustment of light and temperature in the assessment room were offered, and the IT setup was adapted to individual preferences (e.g., screen adjustments, mouse speed). All instructions were presented in writing and had been tested for clarity by an autistic focus group. Short breaks outside the assessment environment (e.g., common room, garden) were possible on request. It was made explicit to the participants that they could end participation without giving a reason at any point. At the first appointment, an additional verbal introduction to the study with information on study participation and payment was given. The assessment was carried out in small groups of three or four participants on the premises of the research institute in the course of two different appointments, each of which lasted about 4\u0026frac12; hours. Individual neuropsychological assessments took place in a separate room. The duration of the individual assessments and questionnaires for the experimental group (EG) was on average about two hours per appointment. CG data was collected within one appointment with the whole group in their college classroom, which took about three hours. The CG was given a detailed explanation of the research purpose and the conditions for participation, and the data collected were limited to informed consent, demographic and feedback questionnaires and the annotation parameters.\u003c/p\u003e\n \u003cp\u003eThe visual data-annotation tasks were assessed using CVAT (Computer Vision Annotation Tool, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.cvat.ai/\u003c/span\u003e\u003c/span\u003e), a professional web-based image-annotation tool. Each annotation task (Level 1 to 6) included a certain amount of images. Scores were calculated using the \u0026ldquo;ground truth\u0026rdquo; provided by expert annotators. Figure \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the CVAT software used.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e2.4 Measures\u003c/h2\u003e\n \u003cp\u003eThe study\u0026rsquo;s measures can be subdivided into three categories: a) visual data-annotation parameters, b) neurocognitive performance and c) self-perception in daily life. Figure 2 summarizes all measures used.\u003c/p\u003e\n \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.1 Visual data-annotation parameters\u003c/h2\u003e\n \u003cp\u003eThe annotation parameters were obtained using an image data annotation software that was developed specifically for job-training purposes. Three types of annotation tasks had to be performed: 1) bounding-box annotation, 2) labeling and 3) a combination thereof.\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eBounding-box annotation was achieved by manually drawing rectangles around the object(s) of interest at the pixel level, and the \u003cem\u003eIntersection over Union\u003c/em\u003e (IoU) score was used to quantify annotation accuracy. The IoU score is useful for evaluating how well a predicted area of interest (e.g., a tumor in an MRI scan) aligns with the ground truth (the actual tumor boundary marked by experts).\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eLabeling was done by defining a textual label (e.g., car, truck) for a whole image, and accuracy was measured by calculating the \u003cem\u003eF1 score\u003c/em\u003e. This score combines Precision and Recall into a single value to give a balanced measure of a test\u0026rsquo;s accuracy; it is also used to evaluate the accuracy of diagnostic tests in which it is crucial to strike a balance between correctly identifying patients with a particular condition (recall) and not mislabeling healthy patients as sick (precision).\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe third annotation task combined labeling and bounding-box annotation. Here, multiple objects had to be annotated using bounding boxes, and to each object of interest a dedicated label had to be assigned. Annotation performance was computed per object and using F1 and IoU scores together for the whole image to ensure comprehensive evaluation. This combination of scores is particularly valuable for images that contain multiple objects, where both precise positioning and a balanced detection rate are crucial.\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eFigure \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e illustrates the scores for the three annotation tasks used in this study.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.2 Neurocognitive performance\u003c/h2\u003e\n \u003cp\u003eNon-verbal IQ and short-term memory were assessed using the \u003cem\u003eReynolds Intellectual Assessment Scales and Screening\u003c/em\u003e (RIAS; (Taylor et al., \u003cspan class=\"CitationRef\"\u003e2006\u003c/span\u003e). The test is normed for the age range 3;0 to 99;11, has an internal consistency of .81 to .95, and can be completed in approximately 30 minutes. This was the only test procedure that was carried out in a one-on-one test setting in a separate room. The other tests were carried out with groups of 4 to 6 participants. The following test instruments were used:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eVisual attention performance (accurate and quick discrimination of similar visual symbols) was tested with the \u003cem\u003eFrankfurter Aufmerksamkeits-Inventar 2\u003c/em\u003e (English: Frankfurt Attention Inventory; FAIR-2; (Petermann, \u003cspan class=\"CitationRef\"\u003e2011\u003c/span\u003e)), which is normed for 9 to 85 years, with a split-half reliability of over .90. It takes approximately 10 minutes to administer.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eReading comprehension was assessed using \u003cem\u003eEin Leseverst\u0026auml;ndnistest f\u0026uuml;r Erst- bis Siebtkl\u0026auml;ssler 2\u003c/em\u003e (English: A reading comprehension test for first- to seventh-graders; ELFE II;(Lenhard et al., \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e)) at the word, sentence and text levels. This test is normed from the first to the seventh grade, and the most recent standard value tables were used in this study. The odd-even-split reliability for this procedure is within the range from r\u0026thinsp;=\u0026thinsp;.87 to .98. The test takes approximately 20 minutes to complete.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eSelf-perceptions in everyday life (questionnaires; see next section)\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.3 Self-perceptions\u003c/h2\u003e\n \u003cp\u003eSubjectively perceived \u003cem\u003estress levels\u003c/em\u003e in everyday life were assessed by the \u003cem\u003ePerceived Stress Scale\u003c/em\u003e (PSS-10; (Schneider et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e)) with ten items related to the frequency of stress-associated conditions, using a 5-point scale from 1 (never) to 5 (very often). This procedure is intended for 18- to 65-year-olds and has \u0026ndash; according to the test authors \u0026ndash; good reliability (\u0026alpha;\u0026thinsp;=\u0026thinsp;.85).\u003c/p\u003e\n \u003cp\u003eSelf-perceived state of \u003cem\u003edisability\u003c/em\u003e in life was assessed with \u003cem\u003ethe WHO Disability Assessment Schedule 2.0\u003c/em\u003e (WHODAS 2.0; (Federici et al., \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e)) using a 5-point scale from \u0026quot;none\u0026quot; to \u0026quot;very severe/not possible\u0026quot;. All 36 items focus on everyday functionality and have good reliability scores of \u0026alpha;\u0026thinsp;=\u0026thinsp;0.70\u0026ndash;0.97.\u003c/p\u003e\n \u003cp\u003eWe created (but did not validate) a questionnaire for the self-assessment of \u003cem\u003ejob-related personality skills\u003c/em\u003e. This questionnaire comprises 20 items and focuses on the following skills: memory, resilience to monotony, on-task accuracy, detail orientation, experience in IT, endurance, decision-making, need for structure, searching for information, and conscientiousness. All items were rated on a 4-point scale from 0 (not true) to 3 (absolutely true) (see supplementary materials).\u003c/p\u003e\n \u003cp\u003eAdditionally, we constructed a questionnaire that comprises 13 items on a 4-point scale (see supplementary materials), which was handed out to gain \u003cem\u003eindividual feedback\u003c/em\u003e on the study procedures and annotation tasks from participants on the autism spectrum.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e2.5 Ethical considerations and data security\u003c/h2\u003e\n \u003cp\u003eThis study was approved by the Johannes Kepler University Linz Ethics Committee. Participants received written information prior to participation and gave written informed consent. To pseudonymize the data, identifiers were removed and replaced with participant identification numbers (e.g., AS01). Data obtained from the neuropsychological measures were stored securely at the Research Institute for Developmental Medicine and on a secure server at the Hospital of St. John of God in Linz in accordance with data protection law. The annotation data was saved anonymously on a server of the Software Competence Center Hagenberg GmbH. The researchers conducting the assessments were sensitive to the needs of individuals on the AS and, throughout the assessment process, tailored the environment to these participants\u0026rsquo; requirements (as far as possible). Following the assessments, participants received individual feedback on their annotation performance and the results of the neuropsychological assessments.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n \u003ch2\u003e2.6 Statistical analysis\u003c/h2\u003e\n \u003cp\u003eThis study used a between-subject design to measure differences between the experimental and comparison groups. All measures of interest (IoU and F1) were either assessed within one session (CG) or on two assessment days within one calendar week (EG). All data analyses were conducted with SPSS 29 and JAMOVI statistical software. Descriptive values were calculated and reported for all demographic, neurocognitive and questionnaire measures. Paired sample tests were performed to separately analyze differences in participants\u0026rsquo; dimensions. Bivariate Pearson correlations between individual characteristics and annotation performance measures were derived. Cohen\u0026rsquo;s d values were used to report on effect sizes. Scatter-plot and bar diagrams are provided.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3 Results","content":"\u003cdiv id=\"Sec17\"\u003e\n \u003ch2\u003e3.1 Visual Data Annotation\u003c/h2\u003e\n \u003cp\u003eIndependent sample t-tests revealed no significant differences in visual annotation tasks between EG and CG according to the F1 (t(34) = 1.59; p = .12) and IoU (t(34) = .95; p = .35) scores. However, a trend towards a moderately higher \u0026gt; F1 score annotation performance related to Cohen’s d (ES = .53) was found for the EG (see Table 2). The correlation between F1 and IoU scores was highly significant (r = .73; p \u0026lt; .001)\u003c/p\u003e\n \u003cdiv\u003e\u0026nbsp;\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 2\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eAnnotation performances for EG and CG.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"3\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eEG\u003c/p\u003e\n \u003cp\u003en = 19\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eCG\u003c/p\u003e\n \u003cp\u003en = 17\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003et-Test\u003c/p\u003e\n \u003cp\u003eCohen’s d\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eES\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.71\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003eIoU score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.77\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.32\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eFigure 4. Visualized annotation parameters for EG and CG. The left diagram illustrates F1- and right diagram IoU-annotation mean scores for both groups with no significant differences.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\"\u003e\n \u003ch2\u003e3.2 EG characteristics and correlations with annotation performances\u003c/h2\u003e\n \u003cp\u003eOur analysis investigated correlations between EG characteristics and F1 annotation scores. Pearson correlations revealed a non-significant negative correlation between non-verbal IQ and F1 annotation score (r=-.38; p = .105) and a non-significant positive correlation between reading skills and F1 annotation score (r = .37; p = .115). No other statistically relevant correlations between EG characteristics and F1 annotation score were found. EG mean characteristics and correlations with F1 score are listed in Table 3. Figure 5 illustrates individual associations between participant characteristics and F1 score.\u003c/p\u003e\n \u003cdiv\u003e\n \u003ctable id=\"Tab5\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 3\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eEG \u003cem\u003echaracteristics and correlations with annotation performance\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eEG\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"4\"\u003e\n \u003cp\u003eCorrelation F1 score\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePearson r\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"3\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eNon-verbal IQ (quotient-score) \u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e105.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e− .38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.105\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"1\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eVisual attention (standardized test score) \u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e60.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e27.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.773\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"1\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eReading comprehension (raw score) \u003csup\u003e3\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.115\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"1\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eStress (raw score) \u003csup\u003e4\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e33.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e− .21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.378\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"1\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eDisability in life (test score) \u003csup\u003e5\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003e− .30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.207\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colspan=\"1\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\"\u003eNotes:\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"8\"\u003e\u003csup\u003e\u003cem\u003e1\u003c/em\u003e\u003c/sup\u003e \u003cem\u003e… RIAS;\u003c/em\u003e \u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e \u003cem\u003e… FAIR-2 Quality-score;\u003c/em\u003e \u003csup\u003e\u003cem\u003e3\u003c/em\u003e\u003c/sup\u003e \u003cem\u003e… ELFE-II text-level;\u003c/em\u003e \u003csup\u003e\u003cem\u003e4\u003c/em\u003e\u003c/sup\u003e \u003cem\u003e… PSS-10;\u003c/em\u003e \u003csup\u003e\u003cem\u003e5\u003c/em\u003e\u003c/sup\u003e \u003cem\u003e… WHODAS 2.0\u003c/em\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec19\"\u003e\n \u003ch2\u003e3.3 Job-related attitudes of EG and CG\u003c/h2\u003e\n \u003cp\u003eExploratory analysis showed that \u003cem\u003eresilience to monotony\u003c/em\u003e was associated between all participants (EG and CG) and annotation performances F1 and IoU (r = .34, p = .044; r = .30, p = .077, respectively; \u003cem\u003e“Monotonous or repetitive tasks are not a problem for me.”\u003c/em\u003e). In addition, a trend towards a positive association between \u003cem\u003econscientiousness\u003c/em\u003e and annotation performances F1 and IoU was found (r = .30, p = .081; r = .28, p = .105, respectively; \u003cem\u003e“When I finish something, I consider what to do next or I ask relevant people.”\u003c/em\u003e).\u003c/p\u003e\n \u003cp\u003eWelch’s t-tests showed significant differences in self-perceived job-related attitudes between EG and CG: EG \u003cem\u003ehigher task accuracy\u003c/em\u003e (t(32.2) = 3.52, p = .001, ES = 1.17; \u003cem\u003e“I enjoy completing tasks in an accurate manner.”\u003c/em\u003e ), CG \u003cem\u003ehigher degree of decisiveness\u003c/em\u003e (t(34)=-3.39, p = .002, ES=-1.14; \u003cem\u003e”I find it easy to make decisions.”\u003c/em\u003e) and \u003cem\u003emore experience in IT\u003c/em\u003e (t(29.6)=-2.80, p = .009, ES=-.92; \u003cem\u003e“I have extensive experience with digital technologies (PCs, internet, software, etc.)”\u003c/em\u003e). Non-significant trends of differences between EG and CG were found: EG greater \u003cem\u003eneed for structure\u003c/em\u003e (t(33.4) = 1.80, p = .081, ES = .60; \u003cem\u003e“When completing a multi-step task, a structuring aid (e.g., visual, from other people) would help me.”\u003c/em\u003e), EG likes \u003cem\u003esearching for information\u003c/em\u003e more (t(31.8) = 1.91, p = .066, ES = .64; \u003cem\u003e“I enjoy searching for new information and explanations.”\u003c/em\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec20\"\u003e\n \u003ch2\u003e3.4 Qualitative feedback on annotation experiences\u003c/h2\u003e\n \u003cp\u003eAfter test administration, all participants from the EG were invited to give feedback on their experiences of the study and of annotation (N = 17). Overall, the participants felt very comfortable and well-informed about the study and reported that the general conditions during the tests were very accommodating to their needs. Wording and procedures were interpreted as clear, and the majority of participants (88%) would participate again. Twelve out of 17 participants wished to \u003cem\u003e“… learn more about annotation in the future.”\u003c/em\u003e and 10 participants \u003cem\u003e“…could imagine working in the field of annotation.”.\u003c/em\u003e\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eAs the digital revolution progresses (Boston Consulting Group, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), the field of AI is beginning to exert a significant influence on the global job market (Milanez, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This development also offers specific groups \u0026ndash; particularly those who face challenges in finding and maintaining suitable employment \u0026ndash; novel opportunities for employment (Davies et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This study aimed to investigate the extent to which this applies to autistic individuals. Over the past decade, data annotation \u0026ndash; a prerequisite for the training, tuning and evaluation of AI systems \u0026ndash; has emerged as a new line of work, which is expected to grow in the coming years (World Economic Forum, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Data scientists and IT enterprises pursue AI research in Western countries, whereas data annotation work is conducted mainly in low-income countries (Smart et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This leaves untapped the potential that local \u0026ndash; and, in particular, neurodivergent \u0026ndash; individuals offer for this line of work.\u003c/p\u003e \u003cp\u003eThe present study investigated the annotation performance of individuals on the autism spectrum with normal intelligence in simple visual data annotation tasks (i.e., assignment of labels and placement of bounding boxes around objects). Participants were recruited through collaboration with a labor-integration project for autistic individuals. This research design incorporated a neurotypical, IT-trained comparison group. As part of a participatory approach, fair and respectful working conditions and compensation were ensured for all participants.\u003c/p\u003e \u003cp\u003eIn answer to Research Question 1, individuals on the AS with normal intelligence showed annotation quality equal to that of an IT-trained neurotypical comparison group. This finding is in line with other single case reports (Garrison et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Schenkenfelder et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Our study sought to gain a deeper understanding of intraindividual characteristics that may affect accuracy in visual data annotation, and thus non-verbal IQ, visual attention, reading comprehension, perceived stress and disability in daily life were assessed for the experimental group. Contrary to the seemingly plausible assumption that higher non-verbal IQ scores would be associated with higher visual annotation performance, our analysis indicated the opposite trend. A possible explanation consistent with feedback from participants is that autistic individuals who show higher non-verbal intellectual function may not be sufficiently motivated and challenged by annotations tasks with low complexity. In line with reports by trainees (Schenkenfelder et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), a positive but non-significant association was observed between reading comprehension and annotation performance. Self-perceived level of stress and disability demonstrated negative correlations with annotation performance, without attaining significance level. Personal characteristics over all participants such as resilience to monotony showed positive and conscientiousness a non-significant positive correlation with annotation performance.\u003c/p\u003e"},{"header":"5 Implications and Future Research","content":"\u003cp\u003eFindings of this study support the promotion of annotation work in Europe in order to reduce biases rooted in culture or geography as identified, for instance, by Mark D\u0026iacute;az et al. (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). The introduction of annotation work in Austria could mark a starting point, with future expansions to other high-income countries.\u003c/p\u003e \u003cp\u003eVisual data annotation tasks are well suited to the target group described and do not seem to require specific strengths in nonverbal cognitive functioning or attention, although a minimum reading level might be beneficial, even though significance level was not attained in the small study sample. Another important skill for high-quality annotation results may be conscientiousness of the annotators. Further, it may be beneficial to have a fundamental understanding of and motivation for the need for accuracy in job-related tasks, an attitude that was more apparent in the EG in comparison to the CG in the limited sample. Annotation can be regarded as a process of continuous decision-making, a skill mentioned more by the CG sample. Individuals on the AS were granted specific working conditions, such as working in small groups, opportunities to relax, low-distraction environments, clear and short written and verbal instructions, and visualized guidelines. In a broader implementation, individuals with specific workplace preferences\u0026mdash;such as working alone or in a noise-free environment\u0026mdash;could be offered adapted working conditions, including working remotely, thereby reducing unemployment in this target group (Espel\u0026ouml;er et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Lorenz et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Zimmermann \u0026amp; Falkner, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). The Conversation of Resources (COR) theory (Hobfoll et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) provides a helpful framework for better describing and understanding (stressful) job experiences in the workplace. A recent study (Tomczak et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) involved autistic individuals and applied the COR-theory to recruitment, selection, onboarding and job retention. The authors suggested inclusive, communication-based strategies, such as: \u0026ldquo;request to solve a given problem instead of a typical job advertisement\u0026rdquo;, \u0026ldquo;verbal instructions, short and to the point\u0026rdquo;, \u0026ldquo;practical skills tests, including gamification-based solutions\u0026rdquo;, \u0026ldquo;provide support of buddy, mentor, job coach\u0026rdquo;, \u0026ldquo;using onboarding checklists, manuals and guides\u0026rdquo;, \u0026ldquo;meetings in small groups\u0026rdquo;, or \u0026ldquo;non-direct, electronically mediated communication\u0026rdquo; (Tomczak et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Such strategies, alongside some adaptations to work-related framework conditions (low-stimulus workspaces or other individual adjustments, Tomczak et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) could support communication between neurodiverse individuals with social communication difficulties and their colleagues and supervisors, thereby reducing barriers. Training sessions could be introduced for the entire team to ensure that the whole environment rather than just the immediate point of contact collaborates with and includes individual(s) with ASD. These adaptations should be seen not as exclusionary, but as promoting inclusion by respecting individual needs and integrating additional people into the labor market. Such a flexible work environment may also appeal to other neurodiverse individuals who prefer reduced sensory stimuli, for instance, individuals with attention-deficit/hyperactivity disorder (ADHD).\u003c/p\u003e \u003cp\u003eFuture studies should include larger samples and focus on optimized workplace designs for neurodivergent individuals, with the main aim of recognizing individuals\u0026lsquo; unique needs and addressing them effectively to improve their prospects in the labor market and their integration into society. Long-lasting, stable employment plays a crucial role in the life and well-being of individuals (Lord et al., 2020). This research project highlights that individuals with ASD can perform valuable work under well-adapted workplace conditions.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003e5.1 Limitations\u003c/h2\u003e \u003cp\u003eA clear limitation of this study is the overall small sample size (N\u0026thinsp;=\u0026thinsp;39), which makes statistical interpretation difficult. However, the analyses provide relevant indications. Since test duration differed between EG and CG, the numbers of images annotated by each group are not directly comparable. Due to the varying levels, which the web-based image-annotation tool offered (level 1 to 6), and the increasing complexity of all images, which had to be annotated during the testing phase, the total amount of annotated images cannot simply be extrapolated. However, importantly the key-component of visual data annotation \u0026ndash; annotation-quality \u0026ndash; was quantifiable and statistically comparable in this trial. An expanded study in terms of sample size and temporal design would be necessary if the goal were to extend the use of statistical methods to enable the identification of group differences through significance tests for the amount of annotated data. Additionally, due to statistical constraints, the exclusion of outliers (high performers and very low performers) was neither feasible nor meaningful. Another critical aspect is the assumption of neurotypicality in the CG. The group was categorized as neurotypical because no diagnoses were reported. However, neurodivergence extends beyond the autism spectrum and includes learning disorders, such as dyslexia and dyscalculia, and ADHD. Statistically, it is unlikely that the control group was truly homogeneously neurotypical (which we assumed in this study due to a lack of information). This limitation was accepted, as a strict definition of neurodivergence was not the focus of this research.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors sincerely thank all participants for their participation and willingness to share their thoughts and experiences during the study. Many thanks to all autistic peers and co-workers for their support and help in developing an autism-friendly study procedure. The authors also thank all students, teachers and the principal of the Traun Technical College for their support and participation in this study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003e\u003cu\u003eContributions\u003c/u\u003e\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eDL:\u0026nbsp;\u003c/strong\u003eConceptualization, Funding acquisition, Investigation, Methodology, Statistical analysis, Data curation, Project administration, Writing, Editing, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eKK:\u0026nbsp;\u003c/strong\u003eFunding acquisition, Investigation, Methodology, Project administration, Writing, Editing, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLF:\u0026nbsp;\u003c/strong\u003eTechnical and Software Development, Methodology, Writing, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBS:\u0026nbsp;\u003c/strong\u003eTechnical and Software Development, Writing, Review\u003cstrong\u003e\u003cbr\u003e\u0026nbsp;SW:\u0026nbsp;\u003c/strong\u003eAutistic peer co-researcher, Methodology, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMM:\u0026nbsp;\u003c/strong\u003eInvestigation\u003cstrong\u003e,\u0026nbsp;\u003c/strong\u003eFormal analysis,\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eWriting, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDH:\u0026nbsp;\u003c/strong\u003eConceptualization, Methodology, Supervision, Writing, Editing, Review\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e: This research was funded by Open Innovation in Science Center of the Ludwig Boltzmann Gesellschaft. The research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The research reported in this paper has been partly funded by the Federal Ministry for Climate Action, Environment, Energy, Mobility, Innovation and Technology (BMK), the Federal Ministry for Labor and Economy (BMAW), and the State of Upper Austria in the frame of the SCCH competence center INTEGRATE [(FFG grant no. 892418)] in the COMET - Competence Centers for Excellent Technologies Programme managed by Austrian Research Promotion Agency FFG.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInstitutional Review Board Statement\u003c/strong\u003e: The study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the JKU Ethics Committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed Consent Statement\u003c/strong\u003e: Informed consent was obtained from the participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e: The data presented in this study are available on request from the corresponding author. The data are not publicly available due to data protection issues.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest\u003c/strong\u003e: The authors declare no conflict of interest. The funder had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAdam, M. T. P., Gimpel, H., Maedche, A., \u0026amp; Riedl, R. (2017). Design Blueprint for Stress-Sensitive Adaptive Enterprise Systems. \u003cem\u003eBusiness \u0026amp; Information Systems Engineering\u003c/em\u003e, \u003cem\u003e59\u003c/em\u003e(4), 277\u0026ndash;291. https://doi.org/10.1007/s12599-016-0451-3\u003c/li\u003e\n\u003cli\u003eBaranger, A [Aur\u0026eacute;lie] (2019). \u003cem\u003eState of play of employment of people on the autism spectrum in Europe: barriers, good practices and trends. \u003c/em\u003eAutism Europe. Committee on Employment and Social Affairs of the European. https://web.archive.org/web/20231027172507/https://www.autismeurope.org/wp-content/uploads/2019/11/presentation_employment_autism_final2.pptx.pdf\u003c/li\u003e\n\u003cli\u003eBaron-Cohen, S [Simon] (2009). Autism: The empathizing-systemizing (E-S) theory. \u003cem\u003eAnnals of the New York Academy of Sciences\u003c/em\u003e, \u003cem\u003e1156\u003c/em\u003e, 68\u0026ndash;80. https://doi.org/10.1111/j.1749-6632.2009.04467.x\u003c/li\u003e\n\u003cli\u003eBaron-Cohen, S [Simon], Ashwin, E., Ashwin, C., Tavassoli, T., \u0026amp; Chakrabarti, B. (2009). Talent in autism: Hyper-systemizing, hyper-attention to detail and sensory hypersensitivity. \u003cem\u003ePhilosophical Transactions of the Royal Society of London. Series B, Biological Sciences\u003c/em\u003e, \u003cem\u003e364\u003c/em\u003e(1522), 1377\u0026ndash;1383. https://doi.org/10.1098/rstb.2008.0337\u003c/li\u003e\n\u003cli\u003eBoston Consulting Group. (2021). \u003cem\u003eThe Future of Jobs in the Era of AI\u003c/em\u003e. https://web-assets.bcg.com/ec/bc/da7341af41358367d26db742eb6c/bcg-the-future-of-jobs-in-the-era-of-ai-may-2021-r.pdf\u003c/li\u003e\n\u003cli\u003eBury, S. M., Hedley, D., \u0026amp; Uljarević, M. (2021). Restricted, Repetitive Behaviours and Interests in the Workplace: Barriers, Advantages, and an Individual Difference Approach to Autism Employment. In E. Gal \u0026amp; N. Yirmiya (Eds.), \u003cem\u003eRepetitive and Restricted Behaviors and Interests in Autism Spectrum Disorders: From Neurobiology to Behavior \u003c/em\u003e(pp. 253\u0026ndash;270). Springer International Publishing. https://doi.org/10.1007/978-3-030-66445-9_15\u003c/li\u003e\n\u003cli\u003eDavies, J., Heasman, B., Livesey, A., Walker, A., Pellicano, E., \u0026amp; Remington, A. (2023). Access to employment: A comparison of autistic, neurodivergent and neurotypical adults\u0026rsquo; experiences of hiring processes in the United Kingdom. \u003cem\u003eAutism\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(6), 1746\u0026ndash;1763.\u003c/li\u003e\n\u003cli\u003eDavies, J., Romualdez, A. M., Pellicano, E., \u0026amp; Remington, A. (2024). Career progression for autistic people: A scoping review. \u003cem\u003eAutism\u003c/em\u003e, 13623613241236110.\u003c/li\u003e\n\u003cli\u003eEspel\u0026ouml;er, J., Proft, J., Falter-Wagner, C. M., \u0026amp; Vogeley, K. (2023). Alarmingly large unemployment gap despite of above-average education in adults with ASD without intellectual disability in Germany: A cross-sectional study. \u003cem\u003eEuropean Archives of Psychiatry and Clinical Neuroscience\u003c/em\u003e, \u003cem\u003e273\u003c/em\u003e(3), 731\u0026ndash;738.\u003c/li\u003e\n\u003cli\u003eFederici, S., Bracalenti, M., Meloni, F., \u0026amp; Luciano, J. V. (2017). World Health Organization disability assessment schedule 2.0: An international systematic review. \u003cem\u003eDisability and Rehabilitation\u003c/em\u003e, \u003cem\u003e39\u003c/em\u003e(23), 2347\u0026ndash;2380.\u003c/li\u003e\n\u003cli\u003eGarrison, E., Singh, D., Hantula, D., Tincani, M., Nosek, J., Hong, S. R., Dragut, E., \u0026amp; Vucetic, S. (2023). Understanding the experience of neurodivergent workers in image and text data annotation. \u003cem\u003eComputers in Human Behavior Reports\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e, 100318.\u003c/li\u003e\n\u003cli\u003eGray, M., \u0026amp; Suri, S. (2019). \u003cem\u003eGhost Work: How Amazon, Google, and Uber Are Creating a New Global Underclass\u003c/em\u003e (1st ed.). Houghton Mifflin Harcourt Publishing Company. https://ebookcentral.proquest.com/lib/kxp/detail.action?docID=29437679\u003c/li\u003e\n\u003cli\u003eHara, K., \u0026amp; Bigham, J. P. (2017). Introducing People with ASD to Crowd Work. In A. Hurst, L. Findlater, \u0026amp; M. R. Morris (Eds.), \u003cem\u003eProceedings of the 19th International ACM SIGACCESS Conference on Computers and Accessibility \u003c/em\u003e(pp. 42\u0026ndash;51). ACM. https://doi.org/10.1145/3132525.3132544\u003c/li\u003e\n\u003cli\u003eHobfoll, S. E., Halbesleben, J., Neveu, J.‑P., \u0026amp; Westman, M. (2018). Conservation of resources in the organizational context: The reality of resources and their consequences. \u003cem\u003eAnnual Review of Organizational Psychology and Organizational Behavior\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e, 103\u0026ndash;128.\u003c/li\u003e\n\u003cli\u003eJolliffe, T., \u0026amp; Baron-Cohen, S [S.] (2001). A test of central coherence theory: Can adults with high-functioning autism or Asperger syndrome integrate fragments of an object? \u003cem\u003eCognitive Neuropsychiatry\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(3), 193\u0026ndash;216. https://doi.org/10.1080/13546800042000124\u003c/li\u003e\n\u003cli\u003eLenhard, W., Lenhard, A., \u0026amp; Schneider, W. (2017). \u003cem\u003eEin Leseverst\u0026auml;ndnistest f\u0026uuml;r Erst- bis Siebtkl\u0026auml;ssler (ELFE II). \u003c/em\u003e\u003c/li\u003e\n\u003cli\u003eLorenz, T., Frischling, C., Cuadros, R., \u0026amp; Heinitz, K. (2016). Autism and overcoming job barriers: Comparing job-related barriers and possible solutions in and outside of autism-specific employment. \u003cem\u003ePloS One\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(1), e0147040.\u003c/li\u003e\n\u003cli\u003eMark D\u0026iacute;az, Ian Kivlichan, Rachel Rosen, Dylan Baker, Razvan Amironesei, Vinodkumar Prabhakaran, \u0026amp; Emily Denton. (2022). CrowdWorkSheets: Accounting for Individual and Collective Identities Underlying Crowdsourced Dataset Annotation. In \u003cem\u003eProceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency \u003c/em\u003e(pp. 2342\u0026ndash;2351). Association for Computing Machinery. https://doi.org/10.1145/3531146.3534647\u003c/li\u003e\n\u003cli\u003eMilanez, A. (2023). The impact of AIL on the workplace: Evidence from OECD case studies of AI implementation. \u003cem\u003eOECD Social, Employment and Migration Working Papers\u003c/em\u003e(289), 1\u0026ndash;115. https://doi.org/10.1787/2247ce58-en\u003c/li\u003e\n\u003cli\u003eNicolaidis, C., Raymaker, D., Kapp, S. K., Baggs, A., Ashkenazy, E., McDonald, K., Weiner, M., Maslak, J., Hunter, M., \u0026amp; Joyce, A. (2019). The AASPIRE practice-based guidelines for the inclusion of autistic adults in research as co-researchers and study participants. \u003cem\u003eAutism\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(8), 2007\u0026ndash;2019. https://doi.org/10.1177/1362361319830523\u003c/li\u003e\n\u003cli\u003ePerkowski, M., Oksztulski, M., Zoń, W., \u0026amp; Kaczyńska, I. (2024). \u003cem\u003eEducated Persons on the Autism Spectrum in the Labour Market: Status Quo and Prospects\u003c/em\u003e. Temida 2 with the cooperation of the Law and Partnership Foundation \u0026hellip;\u003c/li\u003e\n\u003cli\u003ePerrigo, B. (2023). \u003cem\u003eExclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic\u003c/em\u003e. https://time.com/6247678/openai-chatgpt-kenya-workers/\u003c/li\u003e\n\u003cli\u003ePetermann, F. (2011). Frankfurter Aufmerksamkeits- Inventar 2 (FAIR-2). \u003cem\u003eZeitschrift F\u0026uuml;r Psychiatrie, Psychologie Und Psychotherapie\u003c/em\u003e, \u003cem\u003e59. \u003c/em\u003ehttps://doi.org/10.1024/1661-4747/a000088\u003c/li\u003e\n\u003cli\u003ePisani, G., Averius, C., Baranger, A [A.], Garcia, J. M., Mazzoni, A., Thomander Neerland, H., Schneider, T., Tepper Singer, A., \u0026amp; Williams, Z. J. (2022). \u003cem\u003eGuidebook for participant-friendly clinical trials in Autism: for investigators, researchers, clinical trials staff, and the autism community\u003c/em\u003e. https://www.ieepo.com/content/dam/websites/ieepo/2022/resources/learn-library/2022-ieepo-materials/Clinical%20Trials%20in%20Autism%20Guidebook.pdf\u003c/li\u003e\n\u003cli\u003eSchenkenfelder, B., Brandst\u0026auml;tter, U., Fischer, L., Ramler, R., Laister, D., Hartl, M., \u0026amp; Wurm, M. (2024). Responsible AI Engineering: The Case of an Inclusive Image Annotation Team in a Global Technology Company. In \u003cem\u003eProceedings of the 2nd International Workshop on Responsible AI Engineering \u003c/em\u003e(pp. 8\u0026ndash;15). Association for Computing Machinery. https://doi.org/10.1145/3643691.3648583\u003c/li\u003e\n\u003cli\u003eSchneider, E. E., Sch\u0026ouml;nfelder, S., Domke-Wolf, M., \u0026amp; Wessa, M. (2020). Measuring stress in clinical and nonclinical subjects using a German adaptation of the Perceived Stress Scale. \u003cem\u003eInternational Journal of Clinical and Health Psychology : IJCHP\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(2), 173\u0026ndash;181. https://doi.org/10.1016/j.ijchp.2020.03.004\u003c/li\u003e\n\u003cli\u003eSmart, A., Wang, D., Monk, E., D\u0026iacute;az, M., Kasirzadeh, A., van Liemt, E., \u0026amp; Schmer-Galunder, S. (2024, February 9). \u003cem\u003eDiscipline and Label: A WEIRD Genealogy and Social Theory of Data Annotation\u003c/em\u003e. http://arxiv.org/pdf/2402.06811\u003c/li\u003e\n\u003cli\u003eSolomon, C. (2020). Autism and Employment: Implications for Employers and Adults with ASD. \u003cem\u003eJournal of Autism and Developmental Disorders\u003c/em\u003e, \u003cem\u003e50\u003c/em\u003e(11), 4209\u0026ndash;4217. https://doi.org/10.1007/s10803-020-04537-w\u003c/li\u003e\n\u003cli\u003eTan, R., \u0026amp; Cabato, R. (2023). \u003cem\u003eBehind the AI boom, an army of overseas workers in \u0026lsquo;digital sweatshops\u0026rsquo;\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eTaylor, A., Reynolds, C., \u0026amp; Kamphaus, R. (2006). The Reynolds Intellectual Assessment Scales (RIAS) and Assessment of Intellectual Giftedness. \u003cem\u003eGifted Education International\u003c/em\u003e, \u003cem\u003e21. \u003c/em\u003ehttps://doi.org/10.1177/026142940602100305\u003c/li\u003e\n\u003cli\u003eTomczak, M. T., Szulc, J. M., \u0026amp; Szczerska, M. (2021). Inclusive communication model supporting the employment cycle of individuals with autism spectrum disorders. \u003cem\u003eInternational Journal of Environmental Research and Public Health\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(9), 4696.\u003c/li\u003e\n\u003cli\u003eWang, D., Prabhat, S., \u0026amp; Sambasivan, N. (2022). Whose AI Dream? In search of the aspiration in data annotation. \u003cem\u003eCHI Conference on Human Factors in Computing Systems\u003c/em\u003e, 1\u0026ndash;16. https://doi.org/10.1145/3491102.3502121\u003c/li\u003e\n\u003cli\u003eWorld Economic Forum. (2023). \u003cem\u003eAI: 3 ways articicial intelligence is changing the future of work\u003c/em\u003e. https://www.weforum.org/agenda/2023/08/ai-artificial-intelligence-changing-the-future-of-work-jobs/\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. (2022). \u003cem\u003eInternational Classification of Diseases Eleventh Revision (ICD-11)\u003c/em\u003e. https://icd.who.int/browse/2024-01/mms/en\u003c/li\u003e\n\u003cli\u003eZarifhonarvar, A. (2023). Economics of ChatGPT: a labor market view on the occupational impact of artificial intelligence. \u003cem\u003eJournal of Electronic Business \u0026amp; Digital Economics\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(2), 100\u0026ndash;116. https://doi.org/10.1108/JEBDE-10-2023-0021\u003c/li\u003e\n\u003cli\u003eZimmermann, A., \u0026amp; Falkner, G. (2018). Inklusion von Menschen mit besonderen Bed\u0026uuml;rfnissen in den Arbeitsprozess. \u003cem\u003ePersonalmanagement: Internationale Perspektiven Und Implikationen F\u0026uuml;r Die Praxis\u003c/em\u003e, 133\u0026ndash;156.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"universal-access-in-the-information-society","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"uais","sideBox":"Learn more about [Universal Access in the Information Society](http://link.springer.com/journal/10209)","snPcode":"10209","submissionUrl":"https://submission.nature.com/new-submission/10209/3","title":"Universal Access in the Information Society","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Visual Data Annotation, Crowd Work, Autism Spectrum Disorder, Labor Market, Inclusive Jobs","lastPublishedDoi":"10.21203/rs.3.rs-6646902/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6646902/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial Intelligence (AI) systems, which are currently reshaping the global employment landscape, require data to be annotated for training and evaluation \u0026ndash; a task performed mainly by annotation workers located in the Global South. That AI-related companies are located in Western countries, but annotation work is performed in regions that are culturally and socioeconomically different implies low-quality, potentially culturally biased annotation under unacceptable working conditions. It would thus be desirable to move annotation work to Western countries.\u003c/p\u003e \u003cp\u003eThis study investigated whether annotation could be a promising occupation for individuals on the autism spectrum (AS), an underprivileged group in the labor market in Western countries. Our participatory research project included adults on the AS (experimental group, EG; N\u0026thinsp;=\u0026thinsp;22) and a neurotypical comparison group (CG; N\u0026thinsp;=\u0026thinsp;17). Job-related self-perceptions were collected from all participants, neurocognitive parameters (nonverbal IQ, reading comprehension, visual attention) were assessed for EG participants, and correlations with annotation outcomes were analyzed.\u003c/p\u003e \u003cp\u003eFindings confirmed the hypothesized potential of autistic individuals to perform high quality data annotation. Good reading comprehension supported annotation outcomes, while higher IQ scores (\u0026gt;\u0026thinsp;110) were related to a tendency towards lower annotation quality, which might be explained by annotation tasks being insufficiently demanding, although neither correlation was significant in the limited study sample. Annotation performance might be supported by particular working conditions (i.e., working in small groups, minimal distraction, visual guidelines). Future studies should explore long-term trajectories (i.e., staying in work, income development, well-being) for annotation workers on the AS to confirm the reality of opportunities for this population.\u003c/p\u003e","manuscriptTitle":"Empowering Neurodiversity - Analyzing the Potential of Visual Data Annotation as Employment for Autistic Individuals","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-07 08:17:02","doi":"10.21203/rs.3.rs-6646902/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-07-02T17:17:25+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-02T17:13:23+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-05-14T06:16:53+00:00","index":"","fulltext":""},{"type":"submitted","content":"Universal Access in the Information Society","date":"2025-05-12T12:54:05+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"universal-access-in-the-information-society","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"uais","sideBox":"Learn more about [Universal Access in the Information Society](http://link.springer.com/journal/10209)","snPcode":"10209","submissionUrl":"https://submission.nature.com/new-submission/10209/3","title":"Universal Access in the Information Society","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"e8ac215f-57a4-4cf3-801d-722e1ae51c3b","owner":[],"postedDate":"July 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-07-07T08:17:02+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-07 08:17:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6646902","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6646902","identity":"rs-6646902","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}