Exploring the role of generative AI in higher education: Semi-structured interviews with students with disabilities

Classified as probably not a substantial claim.

Scientists with disabilities have expressed concerns that those working on AI guidelines do not consider how these tools can help people with disabilities or other under-represented groups.

“Yet, scientists with disabilities have expressed concerns that those working on AI guidelines do not consider how these tools can help people with disabilities or other under-represented groups (Heidt, 2024).”

• cites Heidt, 2024 sourceSupported

May combine two claims.

Students with disabilities must often self-advocate and disclose their disability, which can be emotionally demanding, especially for those with invisible disabilities.

“Moreover, students must often self-advocate and disclose their disability, which can be emotionally demanding, especially for those with invisible disabilities (Fichten et al., 2020;Moriña, 2017;Osborne, 2019).”

• cites Fichten et al., 2020 sourcePartly supported

  In “Statistics Regarding the Success of Students with Disabilities in Higher Education” (page not supplied), the paper says that many students who self-report disabilities do not register for disability-related services and must manage their access needs on their own. That supports the general point that students may have to handle access needs themselves, but the paper does not say they must often disclose their disability or that doing so is emotionally demanding, particularly for students with invisible disabilities. (LLM, conf=0.78)

  Experimental — accuracy not yet measured
• cites Moriña, 2017 sourceNot checked — no full text of the source
• cites Osborne, 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The literature on AI and accessibility suggests specific concerns regarding the use of AI and people with disabilities.

“More generally, the literature on AI and accessibility also suggests specific concerns regarding the use of AI and people with disabilities (Guo et al., 2020;Morris, 2020;Trewin, 2018).”

• cites Guo et al., 2020 sourceNot checked — no full text of the source
• cites Morris, 2020 sourceNot checked — no full text of the source
• cites Trewin, 2018 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Conversational agents such as ChatGPT can explain or simplify course content or give feedback to students who learn better through dialogue or through repetitive learning.

“Conversational agents such as ChatGPT can explain or simplify course content or give feedback (Morris, 2023) to students who learn better through dialogue (Singh, 2023) or through repetitive learning (McMurtrie, 2023).”

• cites Morris, 2023 sourceSupported
• cites Singh, 2023Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites McMurtrie, 2023Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Adopting UDL in higher education can especially reduce the need for students with disabilities to self-advocate for accommodations.

“Although this approach is meant to support everyone, adopting UDL in higher education can especially reduce the need for students with disabilities to self-advocate for accommodations (Hills et al., 2022).”

• cites Hills et al., 2022 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Lecturers can use technologies to account for the three UDL principles: Multiple Means of Engagement, Action and Expression, and Representation.

“Along with training and top-down strategies (Hills et al., 2022), lecturers can use technologies to account for the three UDL principles: Multiple Means of Engagement, Action and Expression, and Representation (Courtad, 2019).”

• cites Courtad, 2019 sourceNot checked — no full text of the source
• cites Hills et al., 2022 sourceSupported

May combine two claims.

Both mainstream and specialized technologies can be used as assistive technologies, and the difference between the two is blurry.

“Both can be used as assistive technologies, and the difference between the two is blurry (Fichten et al., 2020).”

• cites Fichten et al., 2020 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

ChatGPT can let students self-test their knowledge.

“ChatGPT can also let students self-test their knowledge (Morris, 2023), which is a manner to engage with learning.”

• cites Morris, 2023 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

A common barrier to implementing UDL is the lack of time and resources.

“A common barrier is the lack of time and resources to implement UDL (Hills et al., 2022).”

• cites Hills et al., 2022 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

University teachers have recognized the potential of ChatGPT to facilitate the writing process and reduce the inequalities for students whose native language is not English.

“University teachers have recognized the potential of ChatGPT to facilitate the writing process and reduce the inequalities for students whose native language is not English (Morris, 2023).”

• cites Morris, 2023 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Generative AI performed well with low-stake and easy-to-verify tasks, but did worse with specific tasks such as summarization of nuanced arguments or visualization.

“In an autoethnographic study on the use of GAI for researchers' accessibility needs, Glazko et al. (2023) found that while GAI performed well with low-stake and easyto-verify tasks, it did worse with specific tasks such as summarization of nuanced arguments or visualization.”

• cites Glazko et al., 2023 sourceSupported

May combine two claims.

Due to these characteristics, machine learning algorithms are likely to treat disability as noise in the data, and it is possible that there are not enough people with a specific disability to let the AI identify a category or pattern.

“Due to these characteristics, machine learning algorithms are likely to treat disability as noise in the data, and it is possible that there are not enough people with a specific disability to let the AI identify a category or pattern (Trewin, 2018).”

• cites Trewin, 2018 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The general public may not accept Google Glass due to privacy concerns, but its use to help people with visual impairments may receive more acceptance.

“For example, the general public may not accept Google Glass due to privacy concerns, but its use to help people with visual impairments may receive more acceptance (Profita et al., 2016).”

• cites Profita et al., 2016 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the current debate over GAI, many fear it will be misused for cheating, or that students will stop learning efficiently.

“In the current debate over GAI, many fear it will be misused for cheating, or that students will stop learning efficiently (Morris, 2023).”

• cites Morris, 2023 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Subtle ableist language was introduced when Glazko et al. summarized a research paper with ChatGPT.

“Glazko et al. (2023) also found that subtle ableist language was introduced when they summarized a research paper with ChatGPT.”

• cites Glazko et al., 2023 sourcePartly supported

  Vignette 1 (Mia; page not provided) describes a paper summary that turned a nuanced finding into an ableist trope. But Mia used ChatPDF.com for that summary, not ChatGPT, so the claim misidentifies the tool. (LLM, conf=0.91)

  Experimental — accuracy not yet measured

May combine two claims.

The latest version of OpenAI's model, GPT-4, is not free, which could hinder individuals with disabilities who often face financial constraints.

“The latest version of OpenAI's model, GPT-4, is not free, which could hinder individuals with disabilities who often face financial constraints (Vaughn, 2023;Wolanin, 2005).”

• cites Vaughn, 2023Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Wolanin, 2005 sourcePartly supported

  In “Financial Need and Students with Disabilities,” Wolanin describes particular financial barriers faced by students with disabilities from low-income families. “Having a Disability Is Expensive” also says available aid may not cover their disability-related expenses. This supports the financial-constraints portion, but the paper does not establish it for individuals with disabilities generally or address GPT-4’s price. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

Disability is a category that is very diverse, evolutive, and can be multiple.

“Disability is a category that is very diverse, evolutive, and can be multiple (Trewin, 2018).”

• cites Trewin, 2018 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Lyerly (2023) reported that ChatGPT could provide some form of support with the request for reasonable accommodations.

“While Lyerly (2023) reported that ChatGPT could provide some form of support with the request for reasonable accommodations, almost nobody reported using the tool in this manner.”

• cites Lyerly (2023) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

In Abbas et al. (2024), the authors suggested that students are looking for 'easy means' or 'shortcuts'.

“In Abbas et al. (2024), the authors' explanation regarding the relationship between academic workload and the use of ChatGPT suggested that students are looking for”

• cites Abbas et al. (2024) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In a quantitative analysis of a general population of students in Pakistan, usage of ChatGPT could increase the tendency to procrastinate, memory loss, and reduce academic performance.

“Abbas et al. (2024) found in a quantitative analysis of a general population of students in Pakistan that the usage of ChatGPT could increase the tendency to procrastinate, memory loss, and reduce academic performance.”

• cites Abbas et al. (2024) sourcePartly supported

  The Structural model (p. 15) reports that more frequent ChatGPT use was associated with greater procrastination and memory loss and lower CGPA, supporting the stated outcomes. However, Study 2 surveyed 494 university students (Sample and data collection procedures, pp. 9–10); it did not analyze a general population of students in Pakistan. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Several researchers have pointed out a possible overreliance on technology that could impede learning.

“Several researchers have pointed out a possible overreliance on technology that could impede learning (Abbas et al., 2024;Walter, 2024).”

• cites Abbas et al., 2024 sourceSupported
• cites Walter, 2024 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Kortemeyer (2023) found promising results in using GPT-4 to convert LaTeX documents containing equations into screen-reader-accessible HTML.

“For instance, Kortemeyer (2023) found promising results in using GPT-4 to convert LaTeX documents containing equations into screen-reader-accessible HTML.”

• cites Kortemeyer (2023) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Glazko et al. (2023) found that ChatGPT could successfully provide lists of accessibility rules to follow, but would fail in implementing them itself.

“For example, Glazko et al. (2023) found that ChatGPT could successfully provide lists of accessibility rules to follow, but would fail in implementing them itself.”

• cites Glazko et al. (2023) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Using a participatory approach like Lister et al. (2021) can help identify and guarantee user needs while balancing technical constraints and university requirements.

“Using a participatory approach like Lister et al. (2021) who co-developed a disability disclosure assistant can help identify and guarantee user needs while balancing technical constraints and university requirements.”

• cites Lister et al. (2021) sourcePartly supported

  Sections 2.3 and 4 describe co-designing the disability disclosure assistant with students, identifying their needs, and balancing preferences against university requirements and technical constraints. The paper does not say this approach guarantees that user needs will be met; Section 4 notes that some student ideas were not implemented. (LLM, conf=0.86)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

One student reported that ChatGPT could be talking about autism in a harmful and biased manner.

“One student reported that ChatGPT could be talking about autism in a harmful and biased manner, concurring with concerns about bias in research on AI ethics and disability (Guo et al., 2020;Morris, 2020;Trewin, 2018).”

• cites Guo et al., 2020 sourceNot checked — no full text of the source
• cites Morris, 2020 sourceNot checked — no full text of the source
• cites Trewin, 2018 sourceSupported

May combine two claims.

Glazko et al. (2023) reported subtle built-in ableism when summarizing texts, which could make it hard for non-experts to notice them.

“Notably, Glazko et al. (2023) reported subtle built-in ableism when summarizing texts, which could make it hard for non-experts to notice them.”

• cites Glazko et al. (2023) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

A majority of respondents were engaged in studies at the bachelor's level.

“Respondents were predominantly engaged in studies at the bachelor's level”

Probably not a substantial claim: may not pass S3 (load-bearing).

24 interviewees studied in the German-speaking part of Switzerland.

“A majority of interviewees ( 24) studied in the German-speaking part of Switzerland”

Probably not a substantial claim: may not pass S3 (load-bearing).

Interviews lasted between 24 and 65 minutes.

“Interviews lasted between 24 and 65 min.”

Probably not a substantial claim: may not pass S3 (load-bearing).

7 participants studied in France.

“7 participants in France”

Probably not a substantial claim: may not pass S3 (load-bearing).

2 participants studied in Germany.

“2 in Germany”

Probably not a substantial claim: may not pass S3 (load-bearing).

Another student was in a discovery phase.

“another was in a discovery phase”

Probably not a substantial claim: may not pass S3 (load-bearing).

Out of the 33 interviewed students, 24 different impairments were reported.

“Out of the 33 interviewed students, 24 different impairments were reported.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Seven students in the sample had a chronic disease (e.g., migraine, narcolepsy, chronic pain in the hand).

“Seven students had a chronic disease (e.g., migraine, narcolepsy, chronic pain in the hand).”

Probably not a substantial claim: may not pass S3 (load-bearing).

The median age of respondents was 25 years.

“with a median age of 25”

Probably not a substantial claim: may not pass S3 (load-bearing).

Seven participants first started using ChatGPT between March and May 2023.

“Seven first started between March and May 2023.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Respondents' ages ranged from 20 to 49 years.

“Respondents' ages ranged from 20 to 49 years”

May combine two claims.

Using ChatGPT to summarize reading assignments facilitated text understanding and reduced workload.

“This facilitated their text understanding and reduced workload.”

Probably not a substantial claim: may not pass S3 (load-bearing).

6 participants were enrolled in business sciences.

“followed by business sciences (6)”

Probably not a substantial claim: may not pass S3 (load-bearing).

The final respondent was undecided about the tool's usefulness.

“The final respondent was undecided about the tool's usefulness.”

May combine two claims.

Students with dyslexia/dysorthographia explained that writing can be arduous because they overlook their spelling, grammatical, or conjugation mistakes; it is also hard to acquire vocabulary and start writing.

“Students with dyslexia/dysorthographia explained that writing can be arduous because they overlook their spelling, grammatical, or conjugation mistakes; it is also hard to acquire vocabulary and start writing.”

May combine two claims.

Four neurodiverse students conveyed that they sometimes do not understand lecturers' instructions and could imagine that ChatGPT could help with this issue.

“Furthermore, four neurodiverse students conveyed that they sometimes do not understand lecturers' instructions and could imagine that ChatGPT could help with this issue.”

May combine two claims.

To cope with the barriers, students used assistive technologies, invested themselves personally, received paid-human assistance or support from their family and friends, got university reasonable accommodations, or followed therapy or medical treatment.

“To cope with the barriers, students used assistive technologies, invested themselves personally, received paid-human assistance or support from their family and friends, got university reasonable accommodations, or followed therapy or medical treatment.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Nine students in the sample had a visual impairment.

“Furthermore, there were nine students with a visual impairment.”

May combine two claims.

Among the four neurodiverse students, one had already experimented with the tool and believed they would not have solved the exercise without it.

“Among them, one student had already experimented with the tool and believed they would not have solved the exercise without it.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant studied engineering sciences.

“One participant studied engineering sciences.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant stressed that people with ADHD can benefit significantly from ChatGPT to overcome time management issues.

“One participant stressed that people with ADHD can benefit significantly from ChatGPT to overcome time management issues.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A neurodiverse student asked ChatGPT for learning strategies for someone with memorization difficulty or why he needed to move while he studied.

“A neurodiverse student also asked for learning strategies for someone with memorization difficulty or why he needed to move while he studied.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Information regarding one participant was unavailable.

“information regarding one participant was unavailable”

Probably not a substantial claim: may not pass S3 (load-bearing).

7 men were in the gender distribution.

“and 7 men”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The neurodiverse category includes students with ADHD, 'dys' conditions (e.g., dyslexia, dyscalculia, dyspraxia), autistic students, one student with bipolar type 1, and one student with a schizoaffective disorder.

“This category includes students with ADHD, with”

Probably not a substantial claim: may not pass S3 (load-bearing).

22 participants started using ChatGPT within the first three months after its release at the end of November 2022.

“Additionally, most participants (22) started using ChatGPT within the first three months after its release at the end of November 2022.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The fields of computer sciences, medical sciences, and natural sciences each had two interviewees.

“The fields of computer sciences, medical sciences, and natural sciences each had two interviewees.”

May combine two claims.

Reasons for not using ChatGPT to support self-organization were often that students did not think about this possibility, did not need assistance, or doubted it would be efficient.

“Reasons for not using ChatGPT to support their self-organization were often that they did not think about this possibility, did not need assistance, or doubted it would be efficient.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The largest group in the sample is neurodiverse students, with 20 individuals.

“The largest group in the sample is neurodiverse students, with 20 individuals.”

Probably not a substantial claim: may not pass S3 (load-bearing).

23 women were in the gender distribution.

“The gender distribution included 23 women”

Probably not a substantial claim: may not pass S3 (load-bearing).

Six respondents were pursuing studies at the master's level or equivalent.

“only six were pursuing their studies at the master's level or equivalent”

Probably not a substantial claim: may not pass S3 (load-bearing).

Nine participants exhibited a slightly higher level of familiarity with GAI, as they were keen on acquiring information on the subject.

“Nine exhibited a slightly higher level of familiarity as they were keen on acquiring information on the subject.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One student identified as diverse.

“one student identified as diverse”

May combine two claims.

The two respondents who were more skeptical than their peers explained either that they relied on human experience to study or that their studies were not adapted to ChatGPT.

“The two respondents who were more skeptical than their peers explained either that they relied on human experience to study or that their studies were not adapted to ChatGPT.”

May combine two claims.

Participants who did not see value in ChatGPT preferred interacting with a person when requesting reasonable accommodations (4) or felt they would be more efficient than the chatbot (3).

“On the one hand, participants who did not see value in ChatGPT preferred interacting with a person when requesting reasonable accommodations (4) or felt they would be more efficient than the chatbot (3).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Opinions diverged on the possible benefits of ChatGPT regarding reasonable accommodation.

“Opinions diverged on its possible benefits.”

May combine two claims.

Students with ADHD reported having difficulties structuring their abundant ideas or thoughts and overcoming writer's block.

“Parallelly, students with ADHD reported having difficulties structuring their abundant ideas or thoughts and overcoming writer's block.”

May combine two claims.

Five persons looked for references, one person for keywords, and one had the tool summarize text data from anonymized interviews.

“For instance, five persons looked for references, one person for keywords, and one had the tool summarize text data from anonymized interviews.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Two participants felt it would be unfair to be the only ones authorized to use ChatGPT.

“Whether ChatGPT could be an accommodation on its own was not supported, as only three could imagine this scenario, whereas two felt it would be unfair to be the only ones authorized to use it.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Four interviewees reckoned ChatGPT could help reduce administrative burdens by helping fill out forms or drafting emails.

“Four interviewees also reckoned ChatGPT could help reduce administrative burdens by helping fill out forms or drafting emails.”

May combine two claims.

Out of the 33 participants, 23 used ChatGPT at least weekly, among which ten reported utilizing it daily.

“Out of the 33 participants, 23 used ChatGPT at least weekly, among which ten reported utilizing it daily.”

May combine two claims.

Among the five students who found ChatGPT helpful for study plans, four were neurodiverse and one had a hearing impairment.

“Among the five students, four were neurodiverse and one had a hearing impairment.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Five participants possessed a more technical level of knowledge about GAI because of their current or previous studies or work.

“Five participants possessed a more technical level of knowledge because of their current or previous studies or work.”

Probably not a substantial claim: may not pass S3 (load-bearing).

17 participants were enrolled in social sciences.

“Most participants were enrolled in social sciences (17)”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The student with a hearing impairment also mentioned being a mother and having faced difficult times within the family, making it harder to balance her studies with her private life.

“The student with a hearing impairment also mentioned being a mother and having faced difficult times within the family, making it harder to balance her studies with her private life.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Some participants felt the topic was too sensitive or emotional to be written with ChatGPT (2), questioned its effectiveness (3), or believed that the procedure must be reformed (2).

“Some also felt the topic was too sensitive or emotional to be written with ChatGPT (2), questioned its effectiveness (3), or believed that the procedure must be reformed (2).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three respondents followed double-major degrees combining social sciences with fields such as business, computer science or natural sciences.

“Three respondents followed double-major degrees combining social sciences with fields such as business, computer science or natural sciences.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three participants had tried ChatGPT for the interview.

“Three had tried the tool for the interview.”

Classified as probably not a substantial claim. May combine two claims.

Following lectures, taking notes, and having the energy to study are common challenges for students with disabilities.

“Following lectures, taking notes, and having the energy to study are common challenges for students with disabilities.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The chart highlights the link between the participants' tasks that were reported as challenging and the assistance provided by ChatGPT.

“The chart highlights the link between the participants' tasks that were reported as challenging and the assistance provided by ChatGPT.”

May combine two claims.

16 participants highlighted the individual's internal empowerment to include others, emphasizing readiness for change and awareness as key components.

“16 participants highlighted the individual's internal empowerment to include others, emphasizing readiness for change and awareness as key components.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Five participants imagined lecturers could use ChatGPT to understand how to make their courses more accessible and become aware of issues students with disabilities face.

“Five participants also imagined that lecturers could use Chat-GPT to understand how they can make their courses more accessible and become aware of the issues that students with disabilities face.”

May combine two claims.

11 interviewees highlighted that ChatGPT is most useful when supplementing human teaching or expertise, underscoring its role as a tool rather than a standalone solution.

“Nevertheless, 11 interviewees highlighted that ChatGPT is most useful when supplementing human teaching or expertise, underscoring its role as a tool rather than a standalone solution.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Nine of the 33 interviewed students mentioned having more than one impairment.

“Nine mentioned having more than one impairment.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The writing style of ChatGPT was often found to be good.

“Additionally, although the writing style was often found to be good, about half of the participants noticed that ChatGPT tended to be repetitive, resulting in a robotic tone.”

May combine two claims.

Many students wished their university would allow general use of ChatGPT and provide guidelines to clarify when and how it can be used.

“While students could understand that the tool is not allowed for examinations, many wished their university would allow its general use and provide guidelines to clarify when and how it can be used.”

May combine two claims.

Accuracy issues prompted caution from participants or the use of other tools like Perplexity.ai.

“Issues regarding accuracy prompted caution from participants or the use of other tools like Perplexity.ai.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Seven participants explained universities or non-disabled people might get the false impression that students with disabilities no longer need assistance, which could result in less investment.

“Seven participants explained that universities or non-disabled people might get the false impression that students with disabilities no longer need assistance, which could result in less investment.”

May combine two claims.

Students with visual impairments noted lack of labels and a problematic navigation structure that forced users to scroll to find the latest question or wait while the answer was generated.

“They noted the lack of labels and a problematic navigation structure that forced users to scroll to find the latest question they asked or to wait while the answer was generated.”

May combine two claims.

Students with disabilities in this study already had expectations and would appreciate their universities allowing the utilization of ChatGPT.

“results indicate that students with disabilities already had expectations and would appreciate their universities allowing the utilization of ChatGPT.”

May combine two claims.

A student with ADHD found ChatGPT's answers were usually too long and wished for options to organize chats into topics and to notice when they lost track of time.

“For instance, a student with ADHD found that ChatGPT's answers were usually too long, wished for an option to organize chats into different topics and for a possibility to notice when they lost track of time on the platform.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant stated they would not want ChatGPT to be allowed at the cost of proctoring.

“Still, one person stated that they would not want ChatGPT to be allowed at the cost of proctoring.”

May combine two claims.

Participants identified limitations of ChatGPT relating to accuracy, capacity, writing style, necessary human skills, and platform inaccessibility.

“Limitations relate to accuracy, capacity, writing style, necessary human skills, and platform inaccessibility.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Two participants argued ChatGPT called for higher education reforms.

“Two participants argued that the tool called for higher education reforms.”

May combine two claims.

AI-based chatbots such as ChatGPT are used and perceived positively by students with disabilities.

“the results indicate that AI-based chatbots such as ChatGPT are used and perceived positively by students with disabilities.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Participants mentioned future AI-based applications to support their studies.

“Participants mentioned the following future AI-based applications to support their studies:”

Probably not a substantial claim: may not pass S3 (load-bearing).

Four participants highlighted that AI's reliability is unclear.

“Some highlighted the unknown consequences AI could have (6) and that its reliability is unclear (4).”

May combine two claims.

Participants mentioned lack of transparency in data management and that consequences are yet unknown.

“Specific consequences were not named, but they mentioned the lack of transparency in data management and the fact that the consequences are yet unknown.”

May combine two claims.

Participants mentioned concerns and challenges related to accessibility and people with disabilities in higher education: overreliance and impact on studies, cheating and plagiarism, data privacy, and loss of human contact.

“The following concerns and challenges related to accessibility and people with disabilities in higher education were mentioned: 1) overreliance and impact on studies, 2) cheating and plagiarism, 3) data privacy, and 4) loss of human contact.”

May combine two claims.

Participants in this study emphasized the intrinsic value of human assistance, highlighting its reliability and the emotional aspect of human interactions.

“In this study, participants have emphasized the intrinsic value of human assistance, highlighting its reliability and the emotional aspect of human interactions.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Two students would give GAI another chance as they did not see benefits at the moment.

“Only four students had more reserve concerning future use, two because of university policies that did not allow it, and two explained they would give it another chance as they did not see the benefits at the moment.”

May combine two claims.

A dyslexic student feared they would not notice they plagiarized a study due to their impairment and was more cautious with using ChatGPT for their work.

“In particular, a dyslexic student feared they would not notice they plagiarized a study due to their impairment and therefore was more cautious with using ChatGPT for their work.”

May combine two claims.

Twenty interviewees raised the issue that people or themselves could stop learning, lack critical thinking, and not check information.

“Twenty interviewees raised the issue that people or themselves could stop learning, lack critical thinking, and not check information.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A student with dyslexia explained there is unique value in having a conversation with someone who proofreads a text and doubted AI could replicate this.

“For instance, a student with dyslexia explained that there is a unique value in having a conversation with someone who proofreads a text and doubted that an AI could replicate this.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Some participants mentioned that performance and capacities of other technologies such as automatic captions, optical character recognition, and read-aloud functions could improve with GAI.

“Others mentioned that the performance and capacities of other technologies, such as automatic captions, optical character recognition, and read-aloud functions, could improve with the help of GAI.”

May combine two claims.

Four participants mentioned feeling confident and viewing their disability status as another aspect of their identity.

“Four mentioned feeling confident and viewing their disability status as another aspect of their identity.”

May combine two claims.

Thirteen participants mentioned data privacy as a risk, explaining data could be shared and misused.

“Thirteen participants mentioned data privacy as a risk, explaining that data could be shared and misused.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One student worried about what could happen to them if companies like OpenAI knew about their disabilities.

“one even worried about what could happen to them if companies like OpenAI knew about their disabilities.”

May combine two claims.

The limitations highlighted by students encompass accuracy problems, the absence of certain features like access to up-to-date information or document uploading, a somewhat robotic writing style, the requisite for users to proficiently master prompting skills and critical thinking, and accessibility issues.

“These encompass accuracy problems, the absence of certain features like access to up-to-date information or document uploading, a somewhat robotic writing style, the requisite for users to proficiently master prompting skills and critical thinking, and accessibility issues.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

One participant noted that ChatGPT tends to write positively and can lack criticism.

“as noted by one participant, ChatGPT tends to write positively and can lack criticism.”

May combine two claims.

Students acknowledge the utility of ChatGPT but also highlight specific limitations of the tool.

“Although students acknowledge the utility of ChatGPT, they also highlight specific limitations of the tool.”

May combine two claims.

Concerns and challenges included overreliance and impact on studies, cheating and plagiarism, data privacy, and loss of human contact.

“Concerns and challenges included overreliance and impact on studies, cheating and plagiarism, data privacy, and loss of human contact.”

May combine two claims.

Students wish for the following functions to assist them in their studies: transcription and summarization of courses, AI-generated explanations for visual content, support in communication, and the possibility to produce graphs.

“To assist them in their studies, students wish for the following functions: transcription and summarization of courses, AI-generated explanations for visual content, support in communication, and the possibility to produce graphs.”

May combine two claims.

Administrative procedures to receive reasonable accommodations can be burdensome and require a lot of additional energy.

“Nevertheless, administrative procedures to receive reasonable accommodations can be burdensome and thus require a lot of additional energy (García-González et al., 2021;Osborne, 2019).”

• cites García-González et al., 2021 sourceNot checked — no full text of the source
• cites Osborne, 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

When accessibility is considered an afterthought, it increases the financial costs and the time it takes for people with disabilities to get the support they need.

“When accessibility is considered an afterthought, it increases the financial costs and the time it takes for people with disabilities to get the support they need (Heiman et al., 2020).”

• cites Heiman et al., 2020 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Speech-to-text recognition may not function properly for people with unusual speech patterns.

“For instance, speech-to-text recognition may not function properly for people with unusual speech patterns (Guo et al., 2020).”

• cites Guo et al., 2020 sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

GAI-based technologies like ChatGPT can be referred to as assistive technologies.

“GAI-based technologies like ChatGPT can be referred to as assistive technologies.”

• cites Assistive Technology Industry Association, n.d.Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The UDL framework encourages lecturers to acknowledge that every learner has a unique way of understanding information, being motivated, and expressing their knowledge.

“The UDL framework encourages lecturers to acknowledge that every learner has a unique way of understanding information, being motivated, and expressing their knowledge (CAST, 2018).”

• cites CAST, 2018Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

Flexibility is a central element in UDL.

“Flexibility is a central element in UDL (Courtad, 2019), but it comes with challenges too.”

• cites Courtad, 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

None of the blog and news article authors have devoted much attention to the challenges of ChatGPT.

“While these blog and news article authors list many potential benefits of Chat-GPT, none have devoted much attention to the challenges of this technology.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Lecturers can ask ChatGPT to formulate suggestions to design a course that is more accessible to students with disabilities.

“More generally, lecturers can also ask ChatGPT to formulate suggestions to design a course that is more accessible to students with disabilities (Zhai, 2023).”

• cites Zhai, 2023 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Implementing video captions has become cheaper and more available, although automatic captioning still requires human editing.

“Although automatic captioning still requires human editing, implementing video captions has become cheaper and more available (Kent et al., 2018).”

• cites Kent et al., 2018 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Tools such as tracking, font changer, or text-to-speech can let lecturers present their learning materials in different formats.

“Tools such as tracking, font changer, or text-to-speech can let lecturers present their learning materials in different formats (Courtad, 2019).”

• cites Courtad, 2019 sourceNot checked — no full text of the source

May combine two claims.

This results in higher costs and delays in the inclusion of people with disabilities.

“This results in higher costs and delays in the inclusion of people with disabilities (Heiman et al., 2020).”

• cites Heiman et al., 2020 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

Excluding people from an education service would breach the right to equal opportunity to education as described in the UNCRPD, Article 24.

“Excluding people from an education service would breach the right to equal opportunity to education as described in the UNCRPD, Article 24 (2006).”

• cites UNCRPD, Article 24 (2006)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Financial obstacles and a lack of or inadequate training opportunities can impede the uptake of assistive technologies.

“Financial obstacles and a lack of or inadequate training opportunities can impede the uptake of assistive technologies (Heiman et al., 2020).”

• cites Heiman et al., 2020 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

No participant required an alternative form of communication during the interviews.

“Nobody required any alternative form of communication during the interviews.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Conversational agents might be ineffective for dyslexic individuals if the system has not been trained to interpret texts with atypical spelling or phrasing.

“Interestingly, Guo et al. (2020) highlighted that conversational agents might be ineffective for dyslexic individuals if the system has not been trained to interpret texts with atypical spelling or phrasing.”

• cites Guo et al., 2020 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

ChatGPT seems to provide correct answers to support students with disabilities in higher education.

“ChatGPT seems to provide correct answers to support students with disabilities in higher education (Zhai, 2023).”

• cites Zhai, 2023 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The difficulty of adopting AI-based technologies is due to a lack of resources and time.

“The study also highlighted the difficulty of adopting AI-based technologies due to a lack of resources and time (McGrath et al., 2023).”

• cites McGrath et al., 2023 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

After the first interview, the first author asked the student about the clarity of the questions and revised questions that did not come out well during the session.

“Additionally, after the first interview with a participant, the first author asked the student about the clarity of the questions and revised questions that did not come out well during the session.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

The question of inclusivity cannot be reduced to technical specificities.

“Additionally, the question of inclusivity cannot be reduced to technical specificities.”

Probably not a substantial claim: may not pass S3 (load-bearing).