AI and disability: A systematic scoping review

Abstract

Background: Artificial intelligence (AI) can enhance life experiences and present challenges for people with disabilities. Objectives: This study aims to investigate the relationship between AI and disability, exploring the potential benefits and challenges of using AI for people with disabilities. Methods: A systematic scoping review was conducted using eight online databases; 45 scholarly articles from the last 5 years were identified and selected for thematic analysis. Results: The review’s findings revealed AI’s potential to enhance healthcare; however, it showed a high prevalence of a narrow medical model of disability and an ableist perspective in AI research. This raises concerns about the perpetuation of biases and discrimination against individuals with disabilities in the development and deployment of AI technologies. Conclusion: We recommend shifting towards a social model of disability, promoting interdisciplinary collaboration, addressing AI bias and discrimination, prioritizing privacy and security in AI development, focusing on accessibility and usability, investing in education and training, and advocating for robust policy and regulatory frameworks. The review emphasizes the urgent need for further research to ensure that AI benefits all members of society equitably and that future AI systems are designed with inclusivity and accessibility as core principles.

Keywords

artificial intelligence critical disabilities equity informatics health policy machine learning people with disabilities

Introduction

Background

The Convention on the Rights of Persons with Disabilities (CRPD) declares that disability is a dynamic concept. It arises from the interplay between individuals with impairments and social and environmental barriers that obstruct their full and equal involvement in society.¹ This is known as the “social model” of disability. According to the Ontario Human Rights Commission, disability “must be interpreted to include its subjective component, as discrimination may be based as much on perceptions, myths, and stereotypes, as on the existence of actual functional limitations.”

Meanwhile, artificial intelligence (AI) has become embedded in everyday life.² In particular, AI for healthcare has multiple benefits. It can be used for social good, such as predicting disease onset, tracing pandemics,³ and advocating for people with disabilities.⁴ More specifically, advancements in the use of AI are being made in health informatics, which can help organize and utilize health data. Health informatics is an interdisciplinary field that uses computer science, decision-making, information processing, and management science to design and implement information systems in a user-centric manner. These information systems are designed to obtain, store, process, and provide information to users to assist them in achieving their objectives.⁵ AI is a field of study that allows computers to mimic human intelligence.⁶ Machine learning techniques are a subset of AI techniques that detect patterns in data to assist users in making decisions.⁷ Natural language processing (NLP) is another part of AI that allows computers to decipher the meaning of information and process phrases written or spoken in natural language.^8–12 However, AI can also perpetuate biases against certain groups, including people with disabilities. This is because AI software learns from pre-existing data and uses that data to make predictions. As a result, it is essential to understand the interplay between AI and people with disabilities, and there is no published review on that interplay. This scoping review aims to summarize the current knowledge on the subject, including the benefits and challenges of AI for people with disabilities.

We followed Arskey and O’Malley’s five steps: (1) identifying the research question; (2) identifying relevant studies (through searches of databases, search engines, etc.); (3) study selection (screening of abstracts for relevance); (4) charting the data (study features such as type of technology, disability, etc.); and (5) collating, summarizing, and reporting the results.¹³

Review question

Our review aims to answer the following question: What are the potential benefits and challenges of artificial intelligence (AI) and machine learning (ML) (Concept) for individuals with disabilities (Population) in various contexts (Context) as reflected in scholarly work, and what recommendations can we formulate for AI development?

Methods

Search strategy

The Population, Concept, and Context (PCC) method was used to guide the search strategy for conducting a scoping review. The following eight online databases were searched: ProQuest, IEEE Xplore, ACM Digital Library, Web of Science, Medline, PubMed, PsycINFO, and CINAHL.

• Population: The search focused on studies involving “people with disabilities,” including terms such as “disabilities,” “disability,” “disabled,” and “disable.”

• Concept: The search aimed to identify research related to “artificial intelligence” and “machine learning.” Search terms included “artificial intelligence,” “AI,” “A.I.,” and “machine learning.”

• Context: The review encompassed a broad range of contexts to capture the diverse applications and implications of AI for people with disabilities.

The search terms were established with the guidance of a supervisor to ensure comprehensive coverage of relevant scholarly literature. Of the 354 citations extracted, 84 articles were identified as non-duplicates. The complete search strategies are detailed in Appendix A.

Eligibility criteria

The inclusion criteria were predetermined and included scholarly studies published in the last 5 years, limited to journal articles and conference papers dealing with AI or machine learning and people with disabilities. The exclusion criteria were studies that were not written in English and were not scholarly publications. As this was a scoping review, the search terms were limited to the titles of the articles. The most recent search was done on February 17, 2023.

Selection process

The inclusion and exclusion criteria were applied to 84 abstracts. Two authors (XX and YY) employed Rayyan software to screen 84 articles, and 64 were identified as potentially eligible. Both authors reviewed the full-text version of the 64 articles retrieved for review and synthesis (Figure 1). The literature search and article collection were conducted by a single author (XX), and the literature review and data collection from articles were conducted by both authors (XX and YY) and repeated by another author (ZZ). A discussion was held to reach a consensus in case of any ambiguities and disagreements.

Figure 1.

PRISMA Diagram.

The data charting strategy involved extracting relevant information from the selected articles and organizing it into a standardized format. This included the study’s focus (e.g., diagnosis, rehabilitation), the specific disability addressed, the type of AI technology used, and the study’s outcomes. This structured approach facilitated the comparison and synthesis of findings across different studies.

Data synthesis was primarily qualitative, involving thematic analysis of the extracted data. The authors identified recurring themes and patterns related to the benefits and challenges of AI for people with disabilities. These themes were then categorized and summarized to provide an overview of the current state of knowledge in the field. This review was not registered in a prospective review registry.

Results

The search process identified 354 articles in total. After removing 216 duplicates, 138 articles underwent screening. Out of the 138, 74 were excluded for being unrelated to the scoping subject. Following this, the remaining 64 articles were assessed for eligibility, of which eight were excluded. After the full-text screening, 11 additional articles were excluded. This led to a final selection of 45 studies included in the analysis.

18 studies (40%) were dedicated to adults, eight to children (18%), and 4 (9%) to the elderly. One study (2%) combined child and adult participants, and 14 studies (31%) did not report addressing specific age groups (Table 1). The age distribution in Table 1 was determined by examining the age groups specified in each study. This information is important, as it highlights the focus of current research and potential gaps in addressing the needs of specific age groups.

Table 1.

Studies’ distribution by age group.

Population	Number of studies	Percentage(%)
Adults	18	40
Unreported	14	31
Children	8	18
Elderly	4	9
Children and young adults	1	2

Twenty-six (58%) studies had a clear medical perspective of disability, while 17 (38%) looked at disability from a social model perspective. The final two (4.5%) studies addressed assistive technologies (Table 2). The medical perspective of disability was determined by assessing whether the study primarily focused on disability as an impairment or medical condition rather than a social construct.

Table 2.

Study distribution model.

Model (Medical or Social)	Number of studies	Percentage(%)
Medical	26	58
Social	17	38
Assistive	2	4

In terms of data sources, 14 (31%) studies were conducted in a clinical setting, 8 (18%) used datasets from existing databases, 7 (16%) addressed educational settings, 5 (11%) were conducted in a laboratory setting, and 1 (2%) targeted home settings. The other 10 studies (22%) did not have (or need) a specific setting (Table 3).

Table 3.

Studies’ distribution by data source.

Data Source Type	Number of studies	Percentage(%)
Clinical setting	14	31
Database	8	18
Educational setting	7	16
Laboratory	5	11
Home setting	1	2
NA	10	22

All 34 studies (100%) that built AI models did not measure or address AI bias; the remaining 11 studies did not make AI models. Table 4 (see Appendix) provides a complete summary of the reviewed studies.

Advantages of AI for people with disabilities

Thirty-five articles uncovered three advantages toward which researchers are working: management of health (25 studies),^14–37 assistive technologies (6 studies),^38–43 and disability justice (4 studies).^4,12,44,45

Management of health

Twenty-four studies were related to the management of health conditions: seven studies on diagnosis,^14–20 10 studies on disability progression,^21–30 two studies on rehabilitation,^31,32 four studies on disability risk assessment,^33–36 and one study on access to care.³⁷

Diagnosis

Our review found that AI was used to diagnose three disabling conditions: multiple sclerosis (MS), developmental disorders, and autism.

Three studies used AI to diagnose MS. Flauzino et al. aimed to identify significant factors contributing to increased disability in MS patients to help healthcare professionals develop personalized treatment plans.¹⁴ Rehak Buckova et al. used ML to identify neuroimaging biomarkers related to motor disability in MS patients, which could improve understanding, diagnosis, and treatment for people with MS who have motor disabilities.¹⁵ Montolio et al. applied ML to optical coherence tomography (OCT) data to diagnose and predict disability in MS.¹⁶

Yang and Bai suggested using ML to identify causal genes and variants within patients to understand family diseases, especially X-linked conditions leading to developmental disorders.¹⁷ Nikam et al. used AI and ML to establish specific thresholds of analytes for different levels of intellectual disabilities (ID) to enhance our understanding of disabilities and improve diagnosis, treatment, and management.¹⁸

Two studies were concerned with diagnosing autism. Song et al. used ML with socio-demographic and behavioural observation data to diagnose children with autism spectrum disorder (ASD) and comorbid Intellectual Disabilities.¹⁹ Yperman et al. developed an ML model to improve the accuracy of the Autism Diagnostic Observation Schedule (ADOS/2) in accurately diagnosing ASD, motor evoked potentials,²⁸

Disability progression

Ten studies addressed the prediction of MS disability progression using the Expanded Disability Status Scale (EDSS),²¹ clinical notes,²² thalamic atrophy measurements,²³ optical coherence tomography,²⁵ genetic information,²⁶ motor evoked potentials,²⁸ neuroradiological features,²⁹ and clinical data.³⁰ One study used ML to predict secondary progressive MS (SPMS) disability progression, allowing those at the highest risk of worsening symptoms to access experimental treatments.²⁴ A final study used ML to predict the progression of primary progressive multiple sclerosis (PPMS).²⁷

Rehabilitation

Two studies considered AI and rehabilitation. Islam et al. implemented an expert system using ML to predict the self-care problems of children with physical and motor disabilities.³¹ This system helped therapists detect self-care issues early, make better decisions for children, and ultimately save time and costs in the healthcare system. Hori et al. used ML to identify the incidence of and risk factors linked to hospitalization-associated disability (HAD) in patients after cardiac surgery.³²

Disability risk assessment

Four studies predicted the risk of disabilities. Xiang et al. developed an ML model to predict disability risk in geriatric patients.³³ Koc et al. focused on predicting the permanent disability status of construction workers due to occupational accidents.³⁴ Erbeli et al. reported an ML model that predicts the risk of reading disabilities (RD) among students.³⁵ Finally, Modac et al. used NLP and a learning management system (Moodle) to detect dyslexic students.³⁶

Access to care

ML techniques were used in a study by Youssef and Youssef to develop a social network linking children with learning disabilities to their families and the appropriate specialists.³⁷

Assistive technologies

Six studies were related to using AI in assistive technologies: three in assistive communication^38–40 and three in mobility.^41–43

Communication

Tanabe et al. proposed an AI-based emotion recognition system that recognizes the emotions of children with profound intellectual and multiple disabilities (PIMD) using physiological and motion signals such as eye gaze, heartbeat, and tongue motions.³⁸ Herbuela et al. used ML to classify the movements of children with profound intellectual and multiple disabilities (PIMD) to support communication,³⁹ and Tamilselvan et al. reported the design of an AI-based writing robot to assist individuals with physical disabilities in writing without relying on others.⁴⁰

Mobility

Three studies focused on using AI for mobility. Ghazal et al. proposed an AI-powered service robot to assist elderly and disabled people during shopping experiences using a robotic arm with a lifting mechanism.⁴¹ Encinas Cantaro and Montano Isabel used ML to enable disabled individuals to choose the best routing decisions on the road in urban areas.⁴³ Blanc et al. reported a low-cost prototype of an autonomous electric vehicle (EV) for visually impaired people.⁴²

Disability justice and AI

Only four articles centred AI around disability justice: one related to disability-centred AI development⁴⁴ and three on social justice.^4,12,45

Disability-inspired ML models

One study suggested that AI entities like robots and intelligent agents could be implemented as disabled AI.⁴⁴ The researchers proposed that intentionally turning off certain aspects of an AI algorithm (i.e., disabling them) during the training process may also result in the algorithm developing compensational abilities, also known as coolabilities, thereby improving its performance after training.⁴⁴

Disability advocacy and social justice

Two studies related to social justice by El Morr et al. and Gorman et al. created a virtual platform for disability advocacy groups that utilizes AI techniques such as machine learning, the Semantic Web, and NLP.^4,12 The platform is based on an ontology representing the United Nations Convention on the Rights of Persons with Disabilities (CRPD). It provides a multilingual and accessible environment for uploading and searching disability-related documents.⁴ AI techniques enabled semi-automatic content tagging and intelligent semantic searches, giving access to trusted sources of information for people with disabilities and disability advocacy organizations.⁴ Advocacy groups face challenges in monitoring human rights and uncovering systemic discrimination against people with disabilities due to the lack of disability data availability, but ML approaches can help.

Sobnath et al. used an ML approach to identify features to enhance the possibility of engaging disabled school graduates in employment around 6 months post-graduation.⁴⁵ This research can provide valuable insights into creating more inclusive employment opportunities for disabled individuals.

Disadvantages and challenges of AI for people with disabilities

11 studies identified four weaknesses or challenges that research in the field of AI and disability faces; these include social and legal challenges (2 studies),^46,47 the exclusive design of AI solutions^46,48–53 (7 studies), and the lack of a disability justice approach^54–56 (3 studies).

Ethical and legal challenges

Despite the remarkable advancements in technology, every cutting-edge technology comes with challenges and potential drawbacks. Zdravkova et al. note that AI-based tools raise privacy concerns, while excluding people with disabilities from AI training can lead to biases, paternalistic designs, and neglect of privacy issues with “always-on” technologies.⁴⁶

Moreover, while AI offers significant benefits, it also poses privacy risks. AI relies on user data to learn and improve, raising concerns about how smart devices (e.g., assistive technology apps) collect, use, and protect this information. While some apps allow users to opt out of data collection, this may limit AI functionality. The risks associated with data collection include potential misuse by companies and the possibility of data breaches exposing sensitive information. These concerns raise questions about the extent and nature of data collected and the security measures in place to protect it. Such privacy and cyber-security concerns must be addressed through legal frameworks and incorporating solutions into the design of assistive technologies. Another challenge of using AI for people with disabilities raised by Huq et al. is the high cost of smart devices; this means that assistive tools may only be affordable for certain people, which further marginalizes those with disabilities who are already poor.⁴⁷

Exclusive design of AI solutions

AI classification of people uses big data, and technology developers tend to design for the majority, which makes it challenging to tackle the individual uniqueness of disabled people (e.g., unique visual or hearing impairments), making standard AI solutions inadequate for their needs as Domingo notices.⁴⁸

Disability is usually regarded as a homogenous concept, and data collection is generally not inclusive. Classification of disabled people as outliers can impact their equal access to primary services, such as health insurance, due to the marginalization caused by AI bias.⁴⁸ It is crucial to include individuals with disabilities in the development of AI systems. This helps better understand the implications of AI and design inclusive solutions for all. By creating technology that caters to the needs of people with disabilities, we can also improve satisfaction for those without disabilities. For instance, when sound is unavailable, everyone can benefit from the captions provided for deaf individuals.⁴⁸

Zdravkova et al. emphasize that a lack of engagement of people with disabilities can lead to paternalistic designs that fail to meet the actual needs of students with disabilities.⁴⁶ To address this issue, it is crucial to include people with disabilities in developing AI-based tools and potentially oversample them to ensure their needs and concerns are reflected in the data.⁴⁶

Newman-Griffis et al. notice that this is especially true for older adults with disabilities, who often cannot benefit from intelligent eldercare products due to the complexity of installation and use, resulting in imperfect support facilities.⁵²

AI-assisted decision-making can be biased due to specific design decisions, including problem formulation and data selection. Interpretations of disability can lead to biased AI design. Moreover, a lack of transparency and disabled participation during AI design can lead to inappropriate disability definitions and potential harm. Teng and Ren propose a framework to critically examine AI technologies in decision-making contexts and guide the development of a design praxis for AI analytics related to disabilities. The framework includes critical questions for disability-led design and participatory action to create more equitable AI technologies in disability contexts.⁴⁹

An AI-based medical ethics advisor tool has been developed by Wald.⁵¹ Since people with disabilities often experience health disparities and social challenges, it is crucial to ensure that such AI-based medical ethics tools prioritize equitable implementation and consider the unique needs and experiences of people with disabilities.⁵¹ To create more disability-responsive AI tools, developers should incorporate scoring systems that consider clinicians’ knowledge of patients’ needs. Additionally, approaches to evaluating a patient’s quality of life may need to adequately address the subjective nature of this assessment, particularly for people with disabilities. Therefore, AI tools should be trained to identify and address disability bias in quality-of-life reviews, ensuring fair and equitable decision-making processes for all patients.⁵¹

Protecting against algorithmic discrimination, whether intentional or unintentional, requires enacting laws that mandate the involvement of individuals in the design process of innovative digital systems and algorithms. This will aid in mitigating discriminatory harms while still enjoying the benefits of AI systems.⁵⁰

People with disabilities face disproportionately high unemployment rates compared to the general population, and AI further perpetuates discrimination against people with disabilities during recruitment. Packin proposes that AI developers must account for their implicit and explicit ableist biases during AI design and development to avoid prejudicial effects on job-seekers.⁵³

Lack of a disability justice approach to AI

AI-powered recruitment tools promise efficiency and reduce human bias but often fail to consider disability-related preferences and can reinforce problematic assumptions about disabilities.⁵⁴ To address these concerns, a disability justice approach should be taken, focusing on AI’s social and political nature and emphasizing the need to center the experiences and values of disabled individuals in discussions of AI ethics.⁵⁴

To ensure fairness in AI for people with disabilities, it is essential to consider the intersection of data protection and equality law.⁵⁶ This approach is unique because discrimination and data protection laws apply differently to disability than other protected characteristics. Various areas, such as disability human rights law, equality law, and privacy and data protection law, influence the laws that support AI fairness. However, since these laws can vary across jurisdictions, addressing AI fairness for people with disabilities is complex and context-dependent.⁵⁶

The current workflow for evaluating functional disability (FD) is based on the medical model of disability. The permanent physical impairment (PPI) calculation does not consider personal and environmental factors that can influence the applicant’s daily life. To address this issue, one study proposed a solution for a fairer process of disability evaluation that accurately portrays an individual’s functioning using AI.⁵⁵

Discussion

Addressing disability justice

Of 45 articles, 35 (76.09%) discussed AI for disability, and 11 addressed challenges. Of these 35, 25 (71.43%) were medically focused. In comparison, six articles addressed assistive technologies for people with disabilities (17.14%), and only four articles (11.43%) addressed disability justice issues and hence focused on the social context of people with disabilities. However, one of the four articles technically focused on AI development,⁴⁴ and two belonged to the same study addressing disability justice.^4,12 Hence, we conclude that only two studies out of 35 (5.71%) addressed disability justice issues. This reflects a social context where the medical model of disability is prevalent in AI.

There is a lack of use of the social model of disability in AI research. The social model focuses on the societal and environmental obstacles that disable individuals rather than solely attributing disability to an impairment.⁵⁷ There is a need for interdisciplinary research between AI researchers and disability researchers to expand the understanding of disability beyond impairment to the social context and to empower people with disabilities in the engineering and computer science fields. It is crucial to uphold the “nothing about us without us” principle in these fields to ensure that the disability community’s voice is heard and represented in AI research.⁵⁸

This can be achieved through concerted interdisciplinary research involving critical disabilities, social work, and health informatics researchers and by increasing targeted funding to advance the social agenda related to people with disabilities in AI.^59,60 Involving people with disabilities in collaborative efforts can contribute to developing more inclusive AI systems that prioritize the needs and rights of individuals with disabilities. Additionally, integrating the perspectives of people with disabilities into AI research can help address biases and ensure that AI technologies are leveraged to advance equity and justice.⁶¹

Prevalence of the medical perspective

The high prevalence of the medical perspective in the use of AI for people with disabilities is reflected in this review. This perspective sees disability primarily as a medical issue that requires treatment or management. It focuses on diagnosing and predicting the onset or progression of diseases or disabilities. For example, AI has been used to predict the progression of multiple sclerosis (MS) and to diagnose autism spectrum disorders, often utilizing clinical data and neuroimaging biomarkers.^19,62,63 Such applications underscore the dominance of the medical model, which prioritizes technological solutions for managing health conditions over addressing social and environmental factors that contribute to disability.⁶⁴ The emphasis on medicalization has been criticized for overlooking the broader social impacts and the rights of individuals with disabilities.⁶⁵ The failure to incorporate the social model of disability into AI research underscores the necessity of a comprehensive approach that takes into account the societal obstacles and prejudice experienced by people with disabilities.⁵⁷

Assistive perspective: Missing the broader implications

While studies on AI for people with disabilities have covered assistive technologies related to life experiences, communication, and support for older people,⁶⁶ these applications raise serious concerns about data security and individual privacy. This is especially pertinent when the individuals concerned are people with developmental or intellectual disabilities. AI-driven assistive technologies often collect and process sensitive personal data, which can be vulnerable to misuse and breaches.^67,68 Despite these significant concerns, the interplay between intellectual disability and privacy has received little attention in AI research.

The challenges faced by people with developmental or intellectual disabilities significantly impact their ability to consent and make decisions, affecting their privacy and autonomy.⁶⁹ Unlike elderly individuals or those with physical disabilities, this population often lacks decision-making capacity and is hindered by communication or literacy issues.⁷⁰ By the time privacy breaches are recognized, people with developmental disabilities may have already lost some privacy rights. Therefore, it is crucial, from a policy perspective, to increase awareness, education, and knowledge tools for caregivers and service providers to help protect privacy rights for people with developmental or intellectual disabilities effectively. Future research should focus on developing robust privacy frameworks to protect individuals with intellectual disabilities from potential data exploitation and ensure that these technologies are ethically designed and implemented.^71–73 It is essential to address these issues to build trust and guarantee the ethical use of AI in assistive technologies.

Disability risk: A narrow view

The use of AI to manage disability at work aligns with previous reviews that covered the use of AI to predict absenteeism and temporary disability.⁷⁴ However, out of 58 articles reviewed, none discussed the social implications of human rights approaches. There is a notable tendency to use AI from a technological perspective to achieve effectiveness and efficiency, often neglecting the broader impacts of technology use on social justice and human rights.^50,53 This narrow perspective overlooks the socio-economic factors that influence disability and can lead to biased outcomes that disadvantage already marginalized groups.⁷⁵ Broadening AI research and implementation is crucial to incorporate considerations of equity, inclusion, and the human rights of individuals with disabilities.

From AI ableism to knowledge creators

Previous studies have shown that individuals with disabilities are often portrayed as users of AI and machine learning (ML) technologies with a techno-optimistic tone in various sources, including academic literature, Canadian newspapers, and Twitter.⁷⁶ However, these individuals are rarely depicted as knowledge producers or influencers in AI/ML discussions. This exclusion highlights the need for a more diverse representation of disabled people in AI/ML contexts to ensure they become meaningful contributors and beneficiaries in AI/ML discussions.⁷⁷ The current ableist approach in AI research overlooks the valuable perspectives and expertise that people with disabilities can bring to the development and governance of AI technologies.⁴⁹ To bridge this gap, it’s crucial to promote inclusive approaches that acknowledge and empower people with disabilities as valuable contributors and ethical leaders in the AI/ML domain.⁷⁸

Bias is off the radar: the Absence of debiasing strategies

The articles that discussed prediction did not consider bias or measure it concerning different populations, such as gender or type of disability. This indicates a need for more awareness, research, and reporting on debiasing strategies in AI. Acknowledging and tackling bias in AI is essential to prevent it from perpetuating and worsening existing disparities. This is crucial for ensuring fair and just results for all users, particularly for those in marginalized communities, such as individuals with disabilities.^79,80 It is important to keep in mind that AI systems trained on biased data can perpetuate discriminatory practices and further entrench social biases. This is especially problematic in contexts such as healthcare, employment, and social services.^81,82

It is crucial to incorporate strategies to mitigate bias into the creation and implementation of AI technologies. This includes using varied datasets, utilizing machine learning algorithms that prioritize fairness, and regularly monitoring AI systems for biased results.^83,84 Furthermore, introducing ethical considerations and awareness of bias into the education of AI researchers can encourage a more responsible approach to AI development.⁶¹ Emphasizing the potential risks of AI, such as bias, and the actions taken to mitigate them should be a key focus in research to guarantee that AI technologies benefit all members of society equitably.

Study limitations

While the results of this study can inform future research, it has limitations. First, we have excluded grey literature and focused on scholarly evidence. This decision may have limited the scope of the review and excluded potentially valuable insights from non-academic sources, such as reports, policy documents, and the perspectives of people with disabilities. Second, this review primarily focused on summarizing the findings and identifying key themes. While this approach aligns with the exploratory nature of scoping reviews, it may limit the ability to assess the robustness and reliability of the evidence base. Also, this review’s inclusion criteria focused on studies published in the last 5 years, limited to journal articles and conference papers. This narrow timeframe and source selection excluded older but still relevant research, potentially restricting the comprehensiveness of the review.

Recommendations

Based on our findings, seven recommendations and implications can be drawn to enhance the development and implementation of AI for people with disabilities:

1. Shift Towards a Social Model of Disability: AI research and development needs to prioritize the social model of disability, understanding that disability is more than just a medical condition and is influenced by the interaction between individuals and their environment. This approach would foster the creation of AI solutions that address societal barriers and promote inclusivity.

2. Promote Interdisciplinary Collaboration: AI researchers, disability experts, and individuals with disabilities need to work together to guarantee that AI technologies are created and utilized in ways that truly support and empower people with disabilities. This cooperation should include individuals with disabilities in every phase of AI development, from initial idea generation to the testing and implementation stages.

3. Address Bias and Discrimination: AI researchers need to implement actions to identify and mitigate biases in AI algorithms and datasets. This includes ensuring diverse representation in training data, employing fairness-aware machine learning techniques, and continuously monitoring AI systems for discriminatory outcomes.

4. Prioritize Privacy and Security: Researchers and developers must establish and uphold strict privacy and security guidelines for AI systems that gather and analyze the personal data of people with disabilities. This entails acquiring informed consent, ensuring transparency in data utilization, and implementing strong security measures to safeguard against data breaches.

5. Focus on Accessibility and Usability: Designing AI-powered tools and applications that are accessible and usable for people with diverse disabilities is crucial. This involves integrating features like screen readers, voice recognition, and alternative input methods to guarantee that everyone can take advantage of AI technologies.

6. Invest in Education and Training: It is imperative to ensure that education and training programs for AI researchers, developers, and practitioners focus on disability rights, accessibility, and inclusive design. This will contribute to increasing awareness and building skills for creating AI solutions that are fair and advantageous for everyone in the community.

7. Advocate for Policy and Regulatory Frameworks: Advocacy groups should advocate for the creation and enforcement of policies and regulations that support the ethical and responsible application of AI for individuals with disabilities. This involves guaranteeing that AI tools are employed to uphold human rights, advance social equity, and prevent the worsening of current disparities.

Conclusion

This systematic scoping review highlights the transformative potential of AI in enhancing healthcare and improving the lives of people with disabilities. However, it also reveals a concerning prevalence of the medical model of disability and ableist perspectives within AI research. To harness the full potential of AI for social good, a paradigm shift is necessary to emphasize a disability justice approach, prioritize inclusivity, and address biases in AI algorithms and datasets.

Researchers must actively engage with and prioritize the social model of disability, recognizing its multifaceted nature and the importance of addressing societal barriers. Interdisciplinary collaboration between AI experts, disability scholars, and people with disabilities is crucial to ensure that AI technologies are developed and implemented inclusively and equitably.

Policymakers and practitioners also play a pivotal role in shaping the future of AI by enacting policies that promote ethical and responsible use, safeguard the rights and interests of people with disabilities, and ensure the accessibility and affordability of AI technologies.

By adopting a disability justice approach, prioritizing inclusivity, and addressing biases, we can ensure that AI technologies are used to create a more equitable and empowering future for people with disabilities.

Footnotes

Acknowledgements

The authors express their sincere gratitude to those colleagues within the School of Health Policy and Management at York University who exemplify an equitable, inclusive, innovative, and supportive environment, which has been instrumental in fostering this research. The authors are also deeply grateful for the vibrant intellectual community within the Center for Feminist Research at York University; its commitment to inclusivity, innovation, and collaboration has inspired this research and enriched our experience as scholars.

Author contributions

C.E., R.G., and Y.E. designed the study and received the funds. C.E. supervised B.K., F.M., and S.T., who performed and reported the summaries. C.E. verified the analysis and prepared the first draft. All authors provided critical feedback and revised it; they approved the final version and agreed to be accountable for all aspects of the submitted paper.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is funded by the Social Sciences and Humanities Research Council (SSHRC). (Grant No. 872-2022-1025).

ORCID iDs

Christo El Morr

Bushra Kundi

Appendix

References

United Nations . SDG-CRPD resource package: OHCHR and the rights of persons with disabilities. New York: United Nations. https://www.ohchr.org/en/disabilities/sdg-crpd-resource-package (accessed Augut 25, 2023).

Bialkova

. AI transforming business and everyday life. In: The Rise of AI User Applications: Chatbots Integration Foundations and Trends. Cham: Springer Nature Switzerland, 2024, pp. 143–165.

Sadhana

Pandiarajan

Sivaraman

, et al. AI-based power screening solution for SARS-CoV2 infection: a sociodemographic survey and COVID-19 cough detector. Procedia Comput Sci 2021; 194: 255–271. DOI: 10.1016/j.procs.2021.10.081.

El Morr

Maret

Muhlenbach

, et al. A virtual community for disability advocacy: development of a searchable artificial intelligence-supported platform. JMIR Form Res 2021; 5(11): e33335. DOI: 10.2196/33335.

El Morr

. Introduction to health informatics: a Canadian perspective. 2nd ed. Toronto, Canada: Canadian Scholars’ Press, 2023, p. 354.

El Morr

Jammal

Ali-Hassan

, et al. Machine learning for practical decision making: a multidisciplinary perspective with applications from healthcare. In: Engineering and business analytics (international series in operations research & management science). Cham, Switzerland: Springer International Publishing, 2022, p. 250.

El Morr

Ali-Hassan

. Analytics in healthcare (SpringerBriefs in health care management and economics). Cham, Switzerland: Springer International Publishing, 2019, p. 105.

Creighton

Dada

Gorman

, et al. Reimagining Disability Data Using Artificial Intelligence: A Transnational Disability Rights Approach Presented at the the federation for the Humanities and Social Sciences Congress 2023. Toronto, Ontario: The Canadian Sociological Association, 2023.

Kundi

Dharmalingam

Tadesse

, et al. The Disability Wiki Project: A Virtual Community for Disability Advocacy Using Artificial Intelligence Presented at the the conference for undergraduate health research (CUHR), may 27 - 28, 2022, Toronto. Toronto Canada: York University, 2022.

10.

Kundi

El Morr

Gorman

, et al. Artificial intelligence and bias: a scoping review. In: Yampolskiy

(ed). AI and Society. Chapman & Hall/CRC Artificial Intelligence and Robotics Series. New York, NY: CRC Press Taylor & Francis, 2022, pp. 199–215.

11.

El Morr

. Virtual communities, machine learning and IoT. In: Research Anthology on Mental Health Stigma, Education, and Treatment, A. Information Resources Management. Hershey, PA, USA: IGI Global, 2021, pp. 381–389.

12.

Gorman

Maret

Creighton

, et al. The potential of an artificial intelligence for disability advocacy: the WikiDisability project. Stud Health Technol Inform 2021; 281: 1025–1026. DOI: 10.3233/SHTI210338.

13.

Arksey

O'Malley

. Scoping studies: towards a methodological framework. Int J Soc Res Methodol 2005; 8(1): 19–32. DOI: 10.1080/1364557032000119616.

14.

Flauzino

Simão

ANC

de Carvalho Jennings Pereira

, et al. Disability in multiple sclerosis is associated with age and inflammatory, metabolic and oxidative/nitrosative stress biomarkers: results of multivariate and machine learning procedures. Metab Brain Dis 2019; 34(5): 1401–1413. DOI: 10.1007/s11011-019-00456-7.

15.

Rehak Buckova

Mareš

Škoch

, et al. Multimodal-neuroimaging machine-learning analysis of motor disability in multiple sclerosis. Brain Imaging Behav 2023; 17(1): 18–34. DOI: 10.1007/s11682-022-00737-3.

16.

Montolio

Martín-Gallego

Cegoñino

, et al. Machine learning in diagnosis and disability prediction of multiple sclerosis using optical coherence tomography. Comput Biol Med 2021; 133: 104416. DOI: 10.1016/j.compbiomed.2021.104416.

17.

Yang

Bai

. Prioritizing intellectual disability candidate genes and understanding family diseases using machine learning. In: 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Piscataway: IEEE, 2022, pp. 2886–2892. Available: DOI: 10.1109/BIBM55620.2022.9995119. https://ieeexplore.ieee.org/document/9995119/

18.

Nikam

Ranade

Shaik Mohammad

, et al. A pilot study on machine learning approach to delineate metabolic signatures in intellectual disability. Int J Dev Disabil 2019; 67(2): 94–100. DOI: 10.1080/20473869.2019.1599168.

19.

Song

Jiang

, et al. A machine learning-based diagnostic model for children with autism spectrum disorders complicated with intellectual disability. Front Psychiatry 2022; 13: 993077. DOI: 10.3389/fpsyt.2022.993077.

20.

Wolff

Eberlein

Stroth

, et al. Abilities and disabilities-applying machine learning to disentangle the role of intelligence in diagnosing autism spectrum disorders. Front Psychiatry 2022; 13: 826043. DOI: 10.3389/fpsyt.2022.826043.

21.

Roca

Attye

Colas

, et al. Artificial intelligence to predict clinical disability in patients with multiple sclerosis using FLAIR MRI. Diagn Interv Imaging 2020; 101(12): 795–802. DOI: 10.1016/j.diii.2020.05.009.

22.

Alves

Green

Leavy

, et al. Validation of a machine learning approach to estimate expanded disability status scale scores for multiple sclerosis. Mult Scler J Exp Transl Clin 2022; 8(2): 20552173221108635. DOI: 10.1177/20552173221108635.

23.

Zivadinov

Bergsland

Jakimovski

, et al. Thalamic atrophy measured by artificial intelligence in a multicentre clinical routine real-word study is associated with disability progression. J Neurol Neurosurg Psychiatry 2022; 93(10): 1128–1136. DOI: 10.1136/jnnp-2022-329333.

24.

Law

Traboulsee

, et al. Machine learning in secondary progressive multiple sclerosis: an improved predictive model for short-term disability progression. Mult Scler J Exp Transl Clin 2019; 5(4): 2055217319885983. DOI: 10.1177/2055217319885983.

25.

Montolio

Cegonino

Garcia-Martin

, et al. Comparison of machine learning methods using spectralis OCT for diagnosis and disability progression prognosis in multiple sclerosis. Ann Biomed Eng 2022; 50(5): 507–528. DOI: 10.1007/s10439-022-02930-3.

26.

Fuh-Ngwa

Zhou

Melton

, et al. Ensemble machine learning identifies genetic loci associated with future worsening of disability in people with multiple sclerosis. Sci Rep 2022; 12(1): 19291. DOI: 10.1038/s41598-022-23685-w.

27.

Mostafa

, et al. Predicting upper limb disability progression in primary progressive multiple sclerosis using machine learning and statistical methods. 2021 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Piscataway: IEEE, 2021, pp. 1453–1457. Available: DOI: 10.1109/BIBM52615.2021.9669665. https://ieeexplore.ieee.org/document/9669665/

28.

Yperman

Becker

Valkenborg

, et al. Machine learning analysis of motor evoked potential time series to predict disability progression in multiple sclerosis. BMC Neurol 2020; 20(1): 105. DOI: 10.1186/s12883-020-01672-w.

29.

Tommasin

Cocozza

Taloni

, et al. Machine learning classifier to identify clinical and radiological features relevant to disability progression in multiple sclerosis. J Neurol 2021; 268(12): 4834–4845. DOI: 10.1007/s00415-021-10605-7.

30.

De Brouwer

Becker

Moreau

, et al. Longitudinal machine learning modeling of MS patient trajectories improves predictions of disability progression. Comput Methods Programs Biomed 2021; 208: 106180. DOI: 10.1016/j.cmpb.2021.106180.

31.

Islam

Ashafuddula

NIM

Mahmud

. A machine learning approach to detect self-care problems of children with physical and motor disability. In: 21st International Conference of Computer and Information Technology (ICCIT). Piscataway: IEEE, 2018, pp. 1–4. Available: DOI: 10.1109/ICCITECHN.2018.8631960. https://ieeexplore.ieee.org/document/8631960/

32.

Hori

Usuba

Sakuyama

, et al. Hospitalization-associated disability after cardiac surgery in elderly patients - exploring the risk factors using machine learning algorithms. Circ Rep 2021; 3(8): 423–430. DOI: 10.1253/circrep.CR-21-0057.

33.

Xiang

Jia

, et al. Machine learning-based prediction of disability risk in geriatric patients with hypertension for different time intervals. Arch Gerontol Geriatr 2023; 105: 104835. DOI: 10.1016/j.archger.2022.104835.

34.

Koc

Ekmekcioğlu

Gurgun

. Integrating feature engineering, genetic algorithm and tree-based machine learning methods to predict the post-accident disability status of construction workers. Autom ConStruct 2021; 131: 103896. DOI: 10.1016/j.autcon.2021.103896.

35.

Erbeli

Cheek

, et al. Exploring the machine learning paradigm in determining risk for reading disability. Sci Stud Read 2023; 27(1): 5–20. DOI: 10.1080/10888438.2022.2115914.

36.

Modak

Warade

Saiprasad

, et al. Machine learning based learning disability detection using LMS. In: 2020 IEEE 5th International Conference on Computing Communication and Automation (ICCCA). Piscataway: IEEE, 2020, pp. 414–419. Available: DOI: 10.1109/ICCCA49541.2020.9250761. https://ieeexplore.ieee.org/document/9250761/

37.

Youssef

. Mathematical modeling combined with machine learning for social networks to match children with learning disabilities and specialists. In: 2019 IEEE 10th Annual Information Technology, Electronics and Mobile Communication Conference IEMCON). Piscataway: IEEE, 2019, pp. 0001–0004. Available: DOI: 10.1109/IEMCON.2019.8936233. https://ieeexplore.ieee.org/document/8936233/

38.

Tanabe

Shiraishi

Sato

, et al. A concept for emotion recognition systems for children with profound intellectual and multiple disabilities based on artificial intelligence using physiological and motion signals. Disabil Rehabil Assist Technol 2023; 19: 1319–1326. DOI: 10.1080/17483107.2023.2170478.

39.

Herbuela

Karita

Furukawa

, et al. Machine learning-based classification of the movements of children with profound or severe intellectual or multiple disabilities using environment data features. PLoS One 2022; 17(6): e0269472. DOI: 10.1371/journal.pone.0269472.

40.

Tamilselvan

Yogeshwaran

Pradheep

, et al. Development of Artificial Intelligence based assessment writing Robot for disable people. In: 2020 7th International Conference on Smart Structures and Systems (ICSSS). Piscataway: IEEE, 2020, pp. 1–6. Available: DOI: 10.1109/ICSSS49621.2020.9202067. https://ieeexplore.ieee.org/document/9202067/

41.

Ghazal

Yaghi

Gad

, et al. AI-powered service robotics for independent shopping experiences by elderly and disabled people. Appl Sci 2021; 11(19): 9007. DOI: 10.3390/app11199007.

42.

Encinas Cantaro

Montano Isabel

. Design of an autonomous electric vehicle for assistance in the movement of people with visual disabilities using vision algorithms and artificial intelligence. IOP Conf Ser Mater Sci Eng 2020; 852(1): 012081. DOI: 10.1088/1757-899x/852/1/012081.

43.

Blanc

Liu

Ertz

, et al. Building a crowdsourcing based disabled pedestrian level of service routing application using computer vision and machine learning. In: 2019 16th IEEE Annual Consumer Communications & Networking Conference (CCNC). Piscataway: IEEE, 2019, pp. 1–5. Available: DOI: 10.1109/CCNC.2019.8651850. https://ieeexplore.ieee.org/document/8651850/

44.

Terziyan

Kaikova

. Neural networks with disabilities: an introduction to complementary artificial intelligence. Neural Comput 2021; 34(1): 255–290. DOI: 10.1162/neco_a_01449.

45.

Sobnath

Kaduk

Rehman

, et al. Feature selection for UK disabled students’ engagement post higher education: a machine learning approach for a predictive employment model. IEEE Access 2020; 8: 159530–159541. DOI: 10.1109/access.2020.3018663.

46.

Fichten

Pickup

Asunsion

, et al. State of the research on artificial intelligence based apps for post-secondary students with disabilities. Eei 2022; 31(1): 62–76. DOI: 10.5206/eei.v31i1.14089.

47.

Zdravkova

Krasniqi

Dalipi

, et al. Cutting-edge communication and learning assistive technologies for disabled children: an artificial intelligence perspective. Front Artif Intell 2022; 5: 970430. DOI: 10.3389/frai.2022.970430.

48.

Wald

. AI data-driven personalisation and disability inclusion. Front Artif Intell 2020; 3: 571955. DOI: 10.3389/frai.2020.571955.

49.

Newman-Griffis

Rauchberg

Alharbi

, et al. Definition drives design: disability models and mechanisms of bias in AI technologies. Clin Hemorheol and Microcirc 2023; 28(1): 12903. DOI: 10.5210/fm.v28i1.12903.

50.

Packin

. Disability discrimination using artificial intelligence systems and social scoring: can we disable digital bias? J Int Comp Law 2021; 8(2): 487–511.

51.

Sabatello

. Wrongful birth: AI-tools for moral decisions in clinical care in the absence of disability ethics. Am J Bioeth 2022; 22(7): 43–46. DOI: 10.1080/15265161.2022.2075971.

52.

Teng

Ren

. Application and development prospect of artificial intelligence in the daily life of the elderly and disabled. In: 2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI). Piscataway: IEEE, 2021, pp. 767–770. Available:. DOI: 10.1109/MLBDBI54094.2021.00149. https://ieeexplore.ieee.org/document/9731038/

53.

Moss

. Screened out onscreen: disability discrimination, hiring bias, and artificial intelligence. Denver Law Review 2021; 98(4): 775–805.

54.

Tilmes

. Disability, fairness, and algorithmic bias in AI recruitment. Ethics Inf Technol 2022; 24(2): 21. DOI: 10.1007/s10676-022-09633-2.

55.

Vasudeva

Sheikh

Sahu

. International Classification of Functioning, Disability, and Health augmented by telemedicine and artificial intelligence for assessment of functional disability. J Family Med Prim Care 2021; 10(10): 3535–3539. DOI: 10.4103/jfmpc.jfmpc_692_21.

56.

Binns

Kirkham

. How could equality and data protection law shape AI fairness for people with disabilities? ACM Trans Access Comput 2021; 14(3): 1–32. DOI: 10.1145/3473673.

57.

58.

Charlton

. Nothing about us without us: disability oppression and empowerment. California: University of California Press, 1998.

59.

Barnes

. Understanding the social model of disability: past, present and future. In: Watson

Roulstone

Thomas

(eds). Routledge Handbook of Disability Studies. Milton Park: Routledge, 2020.

60.

Shakespeare

. Disability rights and wrongs. Milton Park: Taylor & Francis, 2006.

61.

Crawford

Roel

Genevieve

, et al. AI now 2018 report. New York: AI Now Institute, 2018.

62.

Flauzino

Simão

ANC

de Carvalho Jennings Pereira

63.

Montolío

Martín-Gallego

Cegoñino

, et al. Machine learning in diagnosis and disability prediction of multiple sclerosis using optical coherence tomography. Comput Biol Med 2021; 133: 104416. DOI: 10.1016/j.compbiomed.2021.104416.

64.

Gurevich

El Hassan

El Morr

. Equity within AI systems: what can health leaders expect? Healthc Manage Forum 2023; 36(2): 119–124. DOI: 10.1177/08404704221125368.

65.

Ontario Human Rights Commission (OHRC) . Policy on ableism and discrimination based on disability. Ontario Human Rights Commission (OHRC). https://www.ohrc.on.ca/en/policy-ableism-and-discrimination-based-disability (accessed June 7, 2024).

66.

Kharbat

Alshawabkeh

Woolsey

. Identifying gaps in using artificial intelligence to support students with intellectual disabilities from education and health perspectives. Aslib J Inf Manag 2020; 73(1): 101–128. DOI: 10.1108/ajim-02-2020-0054.

67.

Fichten

Pickup

Asunsion

, et al. State of the research on artificial intelligence based apps for post-secondary students with disabilities. Eei 2022; 31(1): 62–76. DOI: 10.5206/eei.v31i1.14089.

68.

Sabatello

. Wrongful birth: AI-tools for moral decisions in clinical care in the absence of disability ethics. Am J Bioeth 2022; 22(7): 43–46. DOI: 10.1080/15265161.2022.2075971.

69.

Joffe

. Enforcing the rights of people with disabilities in Ontario’s developmental services system. The Law as it affects persons with disabilities. Toronto, ON: ARCH Disability Law Centre, 2010.

70.

Khanlou

Mantini

Khan

, et al. Protection of privacy of information rights among young adults with developmental disabilities. Int J Ment Health Addict 2018; 16(3): 545–572. DOI: 10.1007/s11469-018-9904-x.

71.

Niemeijer

Frederiks

Riphagen

, et al. Ethical and practical concerns of surveillance technologies in residential care for people with dementia or intellectual disabilities: an overview of the literature. Int Psychogeriatr 2010; 22(7): 1129–1142, Available: https://www.cambridge.org/core/journals/international-psychogeriatrics/article/abs/ethical-and-practical-concerns-of-surveillance-technologies-in-residential-care-for-people-with-dementia-or-intellectual-disabilities-an-overview-of-the-literature/1B88C75F3CA757988C4263EF966AF59A

72.

Trewin

Basson

Muller

, et al. Considerations for AI fairness for people with disabilities. AI Matters 2019; 5(3): 40–63.

73.

Chalghoumi

Cobigo

Dignard

, et al.

Information privacy for technology users with intellectual and developmental disabilities: why does it matter?

Ethics Behav 2019; 29(3): 201–217.

74.

Montano

Marques

Alonso

, et al. Predicting absenteeism and temporary disability using machine learning: a systematic review and analysis. J Med Syst 2020; 44(9): 162. DOI: 10.1007/s10916-020-01626-2.

75.

Tilmes

. Disability, fairness, and algorithmic bias in AI recruitment. Ethics Inf Technol 2022; 24(2): 21. DOI: 10.1007/s10676-022-09633-2.

76.

Lillywhite

Wolbring

. Coverage of artificial intelligence and machine learning within academic literature, Canadian newspapers, and twitter tweets: the case of disabled people. Societies 2020; 10(1): 23. DOI: 10.3390/soc10010023.

77.

El Morr

Maret

Muhlenbach

, et al. A virtual community for disability advocacy: development of a searchable artificial intelligence–supported platform. JMIR Form Res 2021; 5(11): e33335. DOI: 10.2196/33335.

78.

Terziyan

Kaikova

. Neural networks with disabilities: an introduction to complementary artificial intelligence. Neural Comput 2022; 34(1): 255–290. DOI: 10.1162/neco_a_01449.

79.

Friedman

Nissenbaum

. Bias in computer systems. ACM Trans Inf Syst 1996; 14(3): 330–347. DOI: 10.1145/230538.230561.

80.

Mehrabi

Morstatter

Saxena

, et al. A survey on bias and fairness in machine learning. ACM Comput Surv 2021; 54(6): 1–35. DOI: 10.1145/3457607.

81.

Buolamwini

Gebru

. “Gender shades: intersectional accuracy disparities in commercial gender classification.” In: Proceedings of the 1st Conference on Fairness, Accountability and Transparency 2018. New York: Association for Computing Machinery, vol. 81, no. 77-91. Available: https://proceedings.mlr.press/v81/buolamwini18a.html. https://proceedings.mlr.press/v81/buolamwini18a.html

82.

Noble

. Algorithms of oppression: how search engines reinforce racism. New York: NYU Press, 2018.

83.

Binns

. Fairness in machine learning: lessons from political philosophy. Decision-Making in Computational Design & Technology eJournal 2017.

84.

Chouldechova

. Fair prediction with disparate impact: a study of bias in recidivism prediction instruments. Big Data 2017; 5(2): 153–163. DOI: 10.1089/big.2016.0047.