Artificial intelligence and Accessibility: Breaking Barriers for People with Disabilities

Susan Ferebee
Research Flow, Tucson, AZ, USA
Correspondence to: ferebees@gmail.com

DOI: https://doi.org/10.70389/PJAI.100012

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Susan Ferebee – Conceptualization, Writing – original draft, review and editing
• Guarantor: Susan Ferebee
• Provenance and peer-review:
  Commissioned and externally peer-reviewed
• Data availability statement: N/a

Keywords: AI-driven assistive technologies, Algorithmic bias in AI, Disability disclosure chatbots, Accessible stem solutions, Orcam myeye device.

Peer-review
Received: 31 December 2024
Revised: 19 January 2025
Accepted: 19 January 2025
Published: 30 January 2025

Introduction

Artificial intelligence (AI) has emerged as a transformative technology with profound implications for numerous domains, including healthcare, education, and employment. For individuals with disabilities, AI presents unprecedented opportunities to bridge gaps in accessibility, fostering greater inclusion and autonomy. The United Nations¹ estimates that over one billion people worldwide live with some form of disability, underscoring the urgency of addressing barriers to their full participation in society. In this context, AI-driven innovations such as speech recognition, natural language processing, and computer vision are becoming pivotal in enabling individuals with disabilities to navigate a world historically designed without their needs in mind. The intersection of AI and accessibility has garnered increasing scholarly attention. For instance, Iannone and Giansanti² highlight how AI can dismantle structural barriers by enabling assistive technologies, such as screen readers and voice assistants, to provide real-time, context-sensitive support in autism care. Similarly, Gligorea et al.³ examine the role of AI in adaptive learning systems, which tailor educational content to the needs of students with cognitive disabilities. However, while these advancements are promising, they also raise significant ethical and technical challenges. Issues such as algorithmic bias, data privacy, and the digital divide threaten to exacerbate, rather than alleviate, existing inequities.⁴

This literature review explores the state of research at the intersection of AI and accessibility, analyzing its potential to break barriers for individuals with disabilities. Drawing on a diverse body of scholarly work, this review addresses four primary objectives: (1) to understand the challenges of people with disabilities, (2) to evaluate current AI-driven solutions for accessibility, (3) to examine the challenges associated with these technologies critically, and (3) to identify future directions for inclusive AI research and development. This article contributes to a nuanced understanding of how AI can serve as both a tool for empowerment and a site of contestation within the broader struggle for disability rights and equity. This review offers an interdisciplinary perspective on AI and accessibility by situating this analysis within the broader theoretical frameworks of technological determinism and socio-technical systems. Through this approach, the article underscores the imperative to design AI systems that accommodate and actively empower individuals with disabilities, fostering a more inclusive digital and social landscape.

Methodology

This narrative review article synthesizes AI’s current research and application in reducing barriers for people with disabilities. The methodology was a narrative review of academic publications, government reports, industry white papers, and case studies. Sources were selected based on relevance, credibility, and contribution to the field, covering a period from 1968 to 2024 to ensure the inclusion of the most recent advancements. The literature search used academic databases such as Google Scholar, PubMed, and IEEE Xplore; AI-driven research tools like Semantic Scholar and Connected Papers were also used. Keywords such as “AI to empower people with disabilities,” “AI and accessibility,” “ethical applications of AI for people with disabilities,” and “AI tools for people with disabilities” were used to identify relevant sources. To ensure a balanced perspective, this review incorporated studies from various geographical regions, representing both developed and developing countries, and considered different AI technologies for resolving various challenges faced by people with disabilities. The findings were then critically analyzed to identify trends, challenges, and future research directions, offering a holistic view of the current literature and potential advancements in using AI to reduce barriers for people with disabilities.

Challenges for People with Disabilities: People with disabilities face more frequent barriers and barriers that have a more significant impact on their lives. The barriers are more than physical. World Health Organization (WHO) defines barriers as “factors in a person’s environment that, through their absence or presence, limit functioning and create disability.”⁵ These can include (1) services, systems, and policies that do not allow or hinder the involvement of people with disabilities, (2) a non-accessible physical space, (3) lack of or lack of access to assistive technology, and (4) negative attitudes about physical disability.⁵ The CDC⁶ lists the following seven barriers:

Attitudinal: Attitudinal barriers arise from misconceptions about disabilities, often leading to stereotyping, stigma, and discrimination. For instance, people may assume individuals with disabilities have a poor quality of life or view disability as a personal tragedy. Society’s growing understanding of disability as a mismatch between functional needs and the environment encourages inclusive thinking, emphasizing support and independence for all.⁶

Communication: Communication barriers affect individuals with disabilities related to hearing, speaking, or comprehension. Examples include inaccessible written or auditory health messages, lack of Braille or captions, and overly complex language. Addressing these barriers ensures equitable access to vital information for people with diverse communication needs.⁶

Physical: Physical barriers include structural obstacles that hinder mobility or access, such as steps blocking entryways, inaccessible medical equipment, or the absence of wheelchair-accommodating scales. Removing these barriers is essential to enable equal participation in daily activities.⁶

Policy: Policy barriers stem from inadequate enforcement of laws ensuring accessibility. Examples include denying reasonable accommodations or access to federally funded programs. Awareness and adherence to such policies are crucial to fostering an inclusive environment.⁶

Programmatic: Programmatic barriers hinder effective healthcare or public health services delivery. Examples include inaccessible equipment, poor scheduling, insufficient time for procedures, and providers’ lack of understanding about disabilities. Overcoming these barriers enhances program accessibility and effectiveness.⁶

Social: Social barriers involve societal factors like employment, education, income, and safety disparities faced by people with disabilities. For example, people with disabilities experience lower employment rates, higher poverty levels, and increased violence risks. Addressing these inequities requires systemic changes to social determinants of health.⁶

Transportation: Transportation barriers result from limited access to adequate transit, affecting independence and societal participation. Examples include inaccessible public transportation and the lack of convenient options for individuals with vision or cognitive impairments. Enhancing transportation infrastructure is vital for inclusivity.⁶

Models of Disability: The concept of disability has been interpreted and understood through various models that shape how society perceives and addresses the needs of individuals with disabilities. These models provide frameworks for understanding disability, from viewing it as a personal deficiency to recognizing it as a socially constructed phenomenon. By examining these models, we can better understand the diverse perspectives on disability and their implications for policy, accessibility, and inclusion.

Medical Model: In the medical model, disability is perceived as an impairment in a body system or function that is inherently pathological. From this perspective, the goal is to return the system or function as close to “normal” as possible. The medical model suggests that professionals with specialized training are the “experts” in disability. Persons with disability are expected to follow the advice of these “experts.” The language of the medical model is clinical and medical (e.g., left hemiplegia; partial lesion at the T4 level). This view can sometimes be seen within the fields of health, mental health, and education.⁷

Social Model: The social model considers disability an integral part of a person’s identity, similar to race, ethnicity, or gender. This perspective views disability as arising from a mismatch between the individual and their environment—both physical and social. It suggests that societal and environmental barriers, rather than the disability itself, create challenges. Addressing disability, therefore, requires transforming society and its environment rather than focusing on individuals with disabilities. Stereotypes, discrimination, and systemic oppression are seen as key obstacles to achieving full inclusion and removing these barriers.⁷

Identity Model: The identity model views disability as a marker of belonging to a minority group, similar to race or gender. It defines disability primarily through shared social and political experiences shaped by living in a world not designed to accommodate disabled individuals. While this model draws from the social model, it focuses less on how environments, policies, and institutions create barriers and more on building a positive sense of disability identity. This identity is based on shared experiences and circumstances that unite individuals into a distinct and recognizable minority group, “people with disabilities.”⁸

Barrier: Disclosure Requirements for Accommodations

Both students and employees with disabilities are required to disclose their disability in order to receive accommodation. However, many adults with disabilities prefer not to disclose. For example, only 17% of college students with disabilities disclose their disability to receive accommodation. Only 34% of students with disabilities graduate with a four-year degree compared to 51% of students without disabilities.⁹ Similarly, in a large workforce, about 20% of employees manage a disability, but they do not disclose it and, therefore, do not receive accommodations. For the students and employees not receiving accommodation, there is a gap between their work and the results. According to Return on Disability, there is no business case for disclosure. Businesses and colleges should proactively provide standard productivity adjustments to all employees and students. AI can assist in making this happen.¹⁰

Use of AI Chatbots for Disclosure

Even if disclosure continues to be a requirement, chatbots can take over the process. They allow employees and students to disclose information through a conversation with an anonymous chatbot whenever and wherever they want, using text or voice. This removes the stress and confusion of filling out long, complicated forms and reduces feelings of bias and stigma.

Case Study: Taylor, an AI-driven Chatbot for Disability Disclosure

Taylor was developed by the Open University (UK) ADMINS PROJECT, which Microsoft’s AI for Accessibility initiative initially funded. Taylor allows students to discuss their disabilities through a conversation (voice or text), then Taylor completes the form and submits it to disability services. One student, Rob Turner, at the Open University, provided the following comments:

“For me personally, the chatbot is easier to use than the disability support form. I like that I am able to ask questions from the chatbot that I cannot ask when I try to fill in the disability support form. It’s good that the chatbot is flexible. I prefer to speak to the chatbot, but I know other students prefer to type. The Chatbot shows that OU is concerned about disabled students. As a disabled student, I was very happy to take part in the project that would make it easier for disabled students to find out more about what is available and become involved with the OU” (Figure 1).¹¹

Users can respond at their own pace when dialoguing with the chatbot. Filling out the form can be an exhausting process for students with disabilities that overwhelms them. Having a conversation about their disabilities is more relaxed. Students can also select the pace at which the chatbot speaks to ensure thorough understanding.¹¹ A student at OU, Ruth Tudor, stated:

“It was a pleasure to be involved in this project. I liked the ethos of student involvement from start to finish. I believe the bot is a good idea and, once fully functional will be a valuable addition to the disability disclosure process for students.”¹¹

At the end of the disclosure conversation, the chatbot provides a text summary of the discussion, and the student can make any necessary corrections. When the student indicates that everything is correct, the information is sent securely to an OU Disability Support Advisor, who then contacts the student to discuss their needs and available accommodation.¹¹ This type of chatbot can be introduced into any environment where disabilities need to be disclosed before receiving accommodation (e.g., to receive government resources, to receive work accommodations, and to receive accommodation in educational institutions at all grade levels).

Barrier: Non-accessible Science, Technology, Engineering, and Mathematics (STEM) Materials

Individuals who must access STEM reading materials are limited if they have a vision disability and require a screen reader. Screen readers read the content of books and documents auditorily for vision-impaired people. They can comprise students at any grade level, teachers and professors, engineers, and technology, science, and engineering professionals. Previously, screen readers could not read formulas, equations, and symbols. AI is changing this.

The Benetech Process

Benetech is committed to serving people with learning and physical disabilities who cannot access and use the materials they need to be on a level playing field with their peers and colleagues.¹² Aligning with the social model of disability,⁷ Benetech sees equitable access as a human right. They partner with local communities to expand equitable work and education opportunities. Benetech has the most extensive global library of e-books for individuals who read in various ways (auditorily, with Braille, and with large font).¹² The Benetech process to create formulas, symbols, and equations that are readable by screen readers is capable of converting a math textbook with over 5000 equations and images to an accessible format in minutes. The process includes:

• Developing neural networks that identify math and chemistry equations and charts in textbooks.
• Equation images are fed into a math scanner that generates the equation into Math Markup Language (Math ML) (Figure 2).
• Screen readers can read the equations in Math ML.¹³

Benetech’s process is revolutionizing STEM education by using AI to break down accessibility barriers that exclude people with disabilities from fully participating in STEM education and careers.

Case Study: Benetech, a Technology Social Entrepreneurship Venture

Social entrepreneurship focuses on innovation that solves significant social problems. A technological, social venture innovates technology solutions to improve social issues. Benetech began as, and still is, a technological social venture. Founded by Jim Fruchterman in 1989, it began as a non-profit organization that combined a social entrepreneurial model with technology solutions to serve the visually impaired community. Benetech has expanded to provide technological solutions to improve global human rights and manage complicated environmental projects. Additionally, Benetech has created Route 66, a web-based teaching program to improve literacy for children and beginning adult readers.¹⁴

Bookshare.org     is Benetech’s online library of scanned publications and books for people with visual impairments or reading disabilities. The books and publications are accessible and freely available and include STEM materials made accessible with the Benetech process. Bookshare.org, dedicated to bringing full accessibility to people with disabilities, innovates technologically to bring affordable, accessible, and socially useful solutions to improve the world. Benetech encourages other organizations to follow the Technology Social Venture model to continue breaking barriers that exclude disadvantaged and disabled groups from full participation in education, employment, health, and social inclusivity.¹⁴

Barrier: Low Autonomy

AI has made some of the most significant differences in the autonomy of people with disabilities. To appreciate this, we must examine the autonomy of people with disabilities over the years. In 2010, Salvador-Carulla and Gasca¹⁵ discussed functioning and disability as key domains in person-centered medicine and integrated care within a holistic paradigm of disability. The Activities of Daily Living (ADL) model was developed in the United States after World War II. In the 1960s, Katz et al.¹⁶ and Lawton and Brody¹⁷ described ADLs as separated into self-care and instrumental. ADLs are the fundamental tasks that allow functioning with minimum independence and autonomy, like self-care, leisure, homemaking, and employment. Under self-care are included dressing, grooming, bathing, and walking. Instrumental includes doing dishes, preparing food, grocery shopping, taking medication, telephone use, and money management.

The ADL model evolved to include the dependency concept in the early 1990s. The European Council, in the late 1990s, defined dependency as severely disabled people who required support from another person. Dependency is related to losing autonomy. The WHO in 1980 created the International Classification of Impairments, Disabilities, and Handicaps (ICIDH), which evolved into the current International Classification of Functioning, Disability, and Health. The ICIDH altered how health and functioning were related away from the consequences of a particular disease to the result of complicated interactions between the disease/handicap, the individual, and the environment. This system has three factors: (a) activities and participation, (b) body function and structure, and (c) contextual elements. The definition of dependency shifted to mean a state caused by long-term health conditions constraining daily life to the extent that the person needs external support from special aids or another person to function minimally.¹⁵ It is in reducing dependency and improving autonomy that AI can make dramatic changes for a person with disabilities.

OrCam MyEye, Improved Social Engagement and Independence

Many people with vision disabilities cannot shop in a physical store as they cannot select products off the shelves, read labels, or read the currency value of money. Enter OrCam MyEye (Figure 3).

While prohibitively expensive at the moment ($4250), the OrCam MyEye demonstrates the capability of AI to change the independence of people with visual disabilities, and it can be assumed that these prices will reduce over time. OrCam MyEye is a wearable device that uses AI to help people with low or no vision, dyslexia, or reading difficulties understand text and identify objects:

How It Works: The OrCam MyEye attaches to the side of a person’s glasses and uses a smart camera to capture an image. The person can then activate the device using voice commands, pointing gestures, or gaze. The device converts the image into audio the user can hear through a speaker or Bluetooth.¹⁸

Features: OrCam MyEye uses AI to read text and barcodes from books, smartphones, and other surfaces. It can also recognize faces, identify products, scan barcodes, and detect colors.¹⁸ The user can use voice commands to ask the device to perform tasks like “Find canned green beans” or “Read the headlines.” The device can also verify the denomination of paper currency. This device also supports multiple languages.¹⁸

How to Use: To capture a full page of text, the user can tap the side of the device with their finger. They can swipe along the outside of the device to adjust the volume.¹⁸

Benefits: OrCam MyEye is hands-free and can be used without an internet connection. It is suitable for all levels of vision loss and eye conditions and requires no manual dexterity.¹⁸

Drawbacks: OrCam MyEye does not store information, so the user cannot save or annotate what they have scanned. It also does not recognize pictures or graphics.¹⁸

Case Studies: Impact of the OrCam MyEye on Life Quality and Rehabilitation Needs of Visual Impaired People

Study 1

Purpose: This study examined the effect of OrCam MyEye 2.0 (OrCam) on the quality of life and rehabilitation needs in patients with advanced retinitis pigmentosa (RP) or cone-rod dystrophies (CRD).¹⁹

Methods: Participant patients were selected who had a diagnosis of RP (n = 9, 45%) or CRD (n = 11, 55%) and an optimum corrected visual acuity of ≤20/400 Snellen. Three questionnaires (Dutch version of the National Eye Institute Visual Function Questionnaire, the Participation and Activity Inventory, and the OrCam Function Questionnaire) at baseline and after 5.2 weeks using OrCam. The study participants used OrCam to do the following tasks:

• facial recognition
• barcode recognition
• object recognition
• text recognition
• object recognition
• color recognition
• money recognition
• telling time19

Results: Results of the study showed that OrCam was an advantage for text recognition in optimal lighting but had difficulty in low lighting. Portability was rated positive, but the OrCam was unbalanced and heavy on glasses with thin light frames. Hands-free use was also a positive, but the OrCam has a short battery life, limiting its usefulness. Other features that were rated positively were color recognition and barcode recognition. Participants would want connectivity to their smartphones added to the device. Participants also wanted navigation assistance added and a connection to the internet. Participants with loss of peripheral vision rated the OrCam highest for moving about in an unfamiliar indoor environment. For patients with central vision loss, reading, and personal administration were rated highest. Object and facial recognition were rated lowest by participants as it required storing the object or person in OrCam’s memory, which was potentially time-consuming and exhausting for a person with severe visual loss. Eighty-five percent of the participants did not continue to use OrCam, primarily because of the high cost. A newer version of the OrCam, the MyEye Pro, was released after the study, which added orientation assistance and smart reading.¹⁹

Study 2

Purpose: This study aimed to evaluate OrCam My Eye’s effects on real-world activities for visually impaired people. The study also measured the participants’ satisfaction and usability.²⁰

Methods: This case study included participants from five vision rehabilitation centers. Patients had to meet the inclusion criteria of being older than 18 years, visually impaired or blind with an optimum corrected visual acuity less than 0.7 LogMAR, and/or a visual field less than 30% using the WHO classification. Patients were excluded if they had any impairment or health condition that could make it impossible to perform the tasks. All centers received the same spreadsheet for gathering data and specific survey administration instructions. The same functionality and activation modes existed in MyEye as in Study 1. However, this MyEye did include Wi-Fi connectivity. The device’s speech can only be heard by the user. Participants performed real-world activities in differing scenarios without using an LVD and with OrCam MyEye. The tasks included:

• reading a printed book page, a digital screen, and a newspaper article
• recognizing four money bills
• recognizing and reading wall-mounted signs at a distance of 4 m
• recognizing faces
• recognizing object colors20

Patient performance on each task was observed and recorded as a 1 for task completion and a 0 if not completed. Additionally, participants completed the System Usability Scale, Patient’s Global Impression of Change scale, and the Quebec User Evaluation of Satisfaction with Assistive Technology. One hundred subjects from the age range of 19 to 90 participated.²⁰

Results: Completing real-world tasks significantly differed between without and with OrCam use (p < 0.05). Daily living tasks improved for 85% of the participants and were unchanged for 9%. Four participants said task completion worsened when using OrCam. Participants with central vision loss had improved function for more tasks than participants with peripheral or general vision issues. Participants were generally satisfied with OrCam’s performance. Participants’ quality of life was positively affected, and improvements were seen immediately when the device was first used, showing no lengthy training was required.²⁰

Microsoft Seeing AI Smartphone App

The cost of the OrCam was prohibitive for most individuals. The good news is that Microsoft has created a smartphone app with much of the same capability, and it is free. Seeing AI lets you point your phone’s camera, select a channel, and hear a description of what the AI recognizes in your environment or point it at text, like a book or news article, and it immediately starts reading verbally. The app reads barcodes, recognizes and describes people you know, and describes scenery in your current space. Seeing AI recently added a feature to read money bills. It can read your email and describe the photos you store on your smartphone. Although the functionality is similar to the OrCam,²¹ the quality is lower, and it requires the use of hands and manual dexterity, which is not required to use OrCam.

Case Study: The Story Behind the Project

The Seeing AI project, showcased during Satya Nadella’s 2016 Build keynote, began as a passion project inspired by personal experiences and technological breakthroughs. Saqib Shaikh, a blind developer, demonstrated the app’s transformative potential for visually impaired users. The journey started in 2014 when Anirudh Koul, a data scientist, sought to help his grandfather, who was losing his vision, recognize him during Skype calls. Despite initial challenges, major advancements in vision-to-language technology and image classification accuracy in 2015 paved the way for progress.²¹ Koul assembled a diverse team for the 2015 Microsoft Hackathon, drawing from researchers and accessibility experts. The team focused on three key goals: indoor navigation without GPS, interpreting text and objects, and describing surroundings. Their efforts won multiple awards at the Hackathon, securing funding to turn their prototype into a full-time project.²¹

The app’s development was supported by interns from the Microsoft Garage program, who contributed to refining the prototype, including integrating Microsoft Cognitive Services. Released in 2017, Seeing AI quickly gained widespread acclaim, assisting users with over three million tasks and earning prestigious awards like the Helen Keller Achievement Award.²¹ Seeing AI exemplifies the power of innovation and collaboration, enhancing independence and enriching the lives of visually impaired individuals.

Case Study: Evaluating Usability and Identifying Improvements for Seeing AI

Purpose: The purpose of this study was to evaluate the usability of Microsoft’s Seeing AI application and identify areas for improvement to enhance the experience for visually impaired users. The study aimed to align findings with Microsoft’s “Guidelines for Human-AI Interaction” to propose actionable solutions.²²

Methods: The researcher conducted usability testing with six participants who were blind or had low vision. Over two weeks, participants completed four tasks: reading expiration dates, describing figures on holiday cards, identifying people in images, and reading documents. Data was collected through Zoom sessions, where participants shared their screens and audio. Interviews were transcribed, analyzed using qualitative coding, and synthesized with an affinity diagram. Results were cross-referenced with Microsoft’s guidelines to identify key insights and potential solutions.²²

Results: The study identified several usability challenges and proposed solutions:

1. Discoverability: Participants were unaware of the “Browse Photos” feature. Suggested improvement: Promote the feature to the main navigation for better visibility.
2. System Limitations: The app struggled with reading expiration dates on curved surfaces and recognizing digital screens. Suggested improvement: Communicate system capabilities and limitations clearly to users.
3. Accuracy Feedback: Inconsistent probability statements caused confusion. Suggested improvement: Refine AI confidence levels and provide more transparent feedback.
4. Pet Identification: Participants valued better descriptions of their pets. Suggested improvement: Enhance the app’s ability to describe pets’ attributes.
5. Document Reading: The app had difficulty reading multi-column text. Suggested improvement: Adjust reading logic based on document structure and learn from competitor solutions.
6. Color Identification: Users expressed a need for a more precise color identification feature. Suggested improvement: Narrow camera focus for color detection.

Despite the challenges, participants appreciated the app’s capabilities and potential, emphasizing the importance of continued development to meet the needs of a growing visually impaired population.²²

Discussion

The non-systematic narrative review and case study analysis highlight significant barriers faced by people with disabilities, identify systemic and technological shortcomings, and examine innovative solutions that foster inclusivity and independence. The findings emphasize the multidimensional nature of challenges, ranging from physical and attitudinal barriers to inadequate policies and inaccessible technologies, while showcasing the transformative potential of AI-driven innovations. Misconceptions about disabilities perpetuate stigma and discrimination, creating hurdles to full societal participation. Addressing these biases requires societal transformation supported by inclusive policies and educational initiatives. The identity model of disability8 further stresses the importance of cultivating a shared sense of belonging and advocacy among people with disabilities to combat systemic inequities. Inaccessible communication channels, including the lack of Braille or captioning, hinder equitable access to information. Similarly, physical barriers, such as inadequate infrastructure or inaccessible medical equipment, exacerbate exclusion. These findings align with the social model of disability,8 which calls for systemic changes to reduce environmental mismatches. Additionally, enforcement gaps in disability accommodation policies and programmatic healthcare and public services barriers demonstrate systemic failures. These issues reinforce the need for rigorous policy adherence and program redesign to prioritize accessibility.

Inaccessible STEM materials and reliance on outdated assistive technologies restrict opportunities for people with disabilities in education and employment. Additionally, the high costs of advanced devices like OrCam MyEye limit accessibility for many individuals, underscoring the need for affordable solutions. On the other hand, AI chatbots, exemplified by the Open University’s “Taylor,” revolutionize the disability disclosure process by offering anonymous, flexible, and stress-free interactions. These tools reduce stigma and empower individuals to access accommodations, showcasing the potential for scalable applications in workplaces, education, and government programs. Likewise, Benetech’s groundbreaking process converts complex STEM materials into accessible formats within minutes, enabling equitable participation in science and technology fields. This aligns with the social model’s focus on reducing environmental barriers and advancing systemic change.

Devices like OrCam MyEye18 and apps like Microsoft Seeing AI²¹ exemplify AI’s ability to improve autonomy for individuals with disabilities. These technologies address critical needs, including reading, navigation, and object recognition, while fostering greater independence. However, limitations such as high costs (OrCam) and functionality gaps (Seeing AI) highlight areas for further innovation and refinement. The findings underscore the necessity of a holistic approach to disability inclusion that combines systemic reforms, technological innovation, and societal attitudinal shifts. Key recommendations include:

• Policy Enforcement and Reform: Strengthen compliance with accessibility laws and introduce proactive accommodations without requiring disclosure.
• Scalable AI Solutions: To ensure widespread access, develop and subsidize affordable assistive technologies.
• Community Engagement: Involve individuals with disabilities in designing and testing solutions to ensure relevance and usability.
• Awareness Campaigns: Promote inclusive narratives that challenge stereotypes and emphasize the value of diversity.

Addressing the challenges faced by people with disabilities demands a collaborative effort across sectors. AI-driven solutions like Taylor, Benetech’s STEM accessibility process, and OrCam MyEye demonstrate the potential of technology to remove barriers and empower individuals. By combining innovation with systemic change, society can progress toward a more inclusive future that upholds all individuals’ rights, dignity, and autonomy.

Challenges and Limitations

Integrating AI into assistive technologies has offered unprecedented opportunities to improve accessibility, independence, and quality of life for individuals with disabilities. AI-powered tools, like augmentative and alternative communication (AAC) systems, object recognition for individuals with visual impairments, and adaptive mobility aids, hold potential. However, advancing these technologies requires addressing substantial challenges at the intersection of technological limitations, societal barriers, and systemic inequities. Recent literature highlights key concerns, including bias in AI systems, usability shortcomings, and limited intersectional inclusivity. Research on AI-driven assistive technologies has consistently emphasized the detrimental effects of bias in AI design stemming from non-representative datasets and ableist assumptions.

Studies such as Guo et al.²³ identify fairness and representation challenges that exclude diverse disability experiences and emphasize insufficient intersectional data to guide inclusive AI development. Similar findings critique the reinforcement of ableist norms in AI systems while proposing disability-first frameworks and participatory approaches as potential solutions.²⁴ Additionally, intersectionality-focused research^25,26 underscores the compounded barriers faced by individuals with overlapping marginalized identities (e.g., disability and racial minority status), but empirical studies offering actionable strategies remain limited.

Technological limitations also hinder the effective implementation of emerging AI tools in assistive contexts. Generative AI and adaptive learning algorithms have shown potential for improving accessibility in AAC systems²⁷ and personalizing non-standard speech recognition.²⁸ However, these technologies introduce risks, including cognitive overload, alignment mismatches with user needs, and privacy concerns. Concurrently, the high cost of commercial assistive technologies and the lack of scalable, affordable design solutions intensify the digital divide for individuals in low-resource settings.²⁹

In response to these systemic challenges, co-design and participatory frameworks have emerged as central methodologies to ensure the inclusivity and efficacy of assistive technologies. Curtis et al.³⁰ advocate for the involvement of disabled individuals in every stage of technological design to produce culturally sensitive, intersectional, and user-centered tools. However, other researchers suggest that scaling these participatory practices globally remains a significant obstacle.²³ Broader governance and policy frameworks have also been identified as critical in addressing regulatory gaps around accountability and equity in AI development for disabilities.³⁰ This section discussed technological and societal limitations and challenges. However, we must also examine the ethical issues surrounding AI use for people with disabilities.

Policy and Ethical Considerations

The rapid integration of AI technologies into daily life has created significant opportunities to enhance the quality of life and independence of people with disabilities. AI-driven solutions, from assistive robotics and voice recognition tools to data-powered decision-making systems, increase accessibility and inclusion in education, healthcare, and workplace settings. However, these advancements also bring complicated ethical challenges, including algorithmic bias, data privacy concerns, autonomy, and accountability issues. Effective implementation of such technologies relies on an interplay between ethical considerations and policy frameworks that ensure inclusivity while minimizing harm.

A range of research addresses this intersection by exploring both the ethical implications and the role of policy in guiding AI development for disability-focused applications. For example, several studies highlight the persistent risks of algorithmic bias and discrimination in AI systems, which disproportionately affect individuals with disabilities due to non-representative datasets and ableist assumptions embedded in system design.³¹ Researchers have recognized these issues and emphasized anti-ableist approaches, disability-led design frameworks, and the importance of grassroots advocacy to ensure fairness and equitable AI deployment.²⁶ Complementary to this, works like Newman-Griffis et al. (2024)²⁴ propose using social and relational models of disability, which emphasize systemic barriers and inclusion, to inform algorithmic design and ethical considerations.

The critical role of privacy protections and data governance in AI for disabilities has also been highlighted in the literature. Studies investigating frameworks such as the European Union’s General Data Protection Regulation and the Convention on the Rights of Persons with Disabilities identify both their strengths in protecting individuals’ rights and their limitations in addressing AI-specific contexts, such as data exploitation or emergent medical data mining.³² The potential for violating privacy laws by AI systems that infer disability status without consent,³³ and calls for trust-building mechanisms to facilitate ethical data collection from disabled communities³⁴ are ethical issues that must be addressed.

Future Implications and Future Research

Future Implications

Integrating AI into technologies that support individuals with disabilities is a growing research domain, with a broad focus on improving personal autonomy, accessibility, and inclusivity across diverse societal domains. AI-powered tools, such as adaptive learning platforms, assistive workplace accommodations, and real-time communication aids, exhibit the potential to redefine education, employment, and societal perceptions of disability.³⁵ However, the broader societal implications of these advancements—specifically in terms of shifts in social attitudes, equitable access to education, and workplace inclusivity—are complex and require further exploration. A recurring theme in the literature is the combined risk and opportunity of AI amplifying either inclusivity or harmful biases. There is concern that AI-driven solutions will be designed based on ableism-biased datasets and may inadvertently reinforce the medical model of people with disabilities.³⁶ The promise of personalized learning systems and assistive tools in breaking down barriers for disabled students is critical, but barriers to implementation, such as the digital divide, algorithmic biases, and ethical concerns around autonomy and equity, exist.³⁷

In the workplace, AI has shown a potential to enhance inclusivity through tools that support accommodations, flexible working environments, and equitable hiring; however, the continuance of bias in automated hiring systems and the chance of isolating disabled employees by over-relying on remote roles suggest the need for critical evaluations of AI systems.³⁸ Despite these advances, the literature lacks longitudinal studies investigating AI’s systemic impacts on social values, educational systems, and organizational culture. The convergence of challenges such as data bias, ethical risks, and digital inequality underscores the complexity of ensuring that AI supports meaningful and equitable societal transformations.³⁶ While AI holds substantial potential to improve the lives of individuals with disabilities, significant gaps remain in understanding its broader societal influence. Further research is needed to examine how these technological innovations can foster systemic inclusivity and societal change across social, educational, and workplace domains.

Future Research Directions

• Personalization and User-Centric Design: Enhancing usability and user satisfaction through individualized support is critical to meaningful improvements for people with disabilities. Effective AI technologies must adapt to the user’s unique needs, contexts, and preferences. People with disabilities often have combined disabilities, which must be considered in the technology’s adaptive design.

• Intersectional Inclusivity: Solutions must be accessible and inclusive for highly diverse populations. Overlapping identities such as socioeconomic status, age, ethnicity, and disability must be addressed in the technology’s design.
• Disability Bias Removal and Ethical AI Development: Non-ableist models and disability characteristics and needs datasets must be developed to provide a dataset that effectively represents a wide range of people with disabilities. Methods must be improved to detect and eliminate disability biases in AI models.
• Enhanced Interaction Options: Continue to improve intuitive communication and control techniques, ensuring multiple options in each technology to account for audio, visual, and physical dexterity issues.
• Strategies to Reduce Costs and Improve Availability and Accessibility: The high cost of OrCam is a good example of how costs must be reduced for these technologies to improve the lives of people with disabilities.
• Collaborate with Stakeholders to Guide Technology Design: Engage end-users, caregivers, family members, healthcare professionals, and disability advocates in the design of AI technologies for people with disabilities. This will ensure that solutions align with real-world needs.
• Policy and Regulatory Frameworks: Inform policies, regulations, and frameworks that protect people with disabilities as they take advantage of greater autonomy through AI technology.

Conclusion

AI can potentially revolutionize accessibility for people with disabilities, offering improved autonomy, inclusivity, and the ability to perform daily living activities. From enhancing communication and mobility to increasing educational and employment opportunities, AI-driven technologies have demonstrated their ability to empower and include. However, realizing this potential requires addressing barriers such as algorithmic bias, affordability, usability, and ethical concerns. Collaborative efforts among researchers, policymakers, developers, and disability advocates are essential to create inclusive, equitable, and effective AI systems. By prioritizing user-centric design, reducing costs, and fostering intersectional inclusivity, AI can break down barriers and reshape societal perceptions of disability, paving the way for a future where all individuals can thrive.

References

1 United Nations. Factsheet on Persons with Disabilities. United Nations. c2024 [cited 2024 December 14]. Available from: https://www.un.org/development/desa/disabilities/resources/factsheet-on-persons-with- disabilities.html
 
2 Iannone A, Giansanti D. Breaking barriers: The intersection of AI and assistive technology in Autism care: A narrative review. J Pers Med. 2023;14(1):41. https://doi.org/10.3390/jpm14010041
https://doi.org/10.3390/jpm14010041
 
3 Gligorea I, Cioca M, Oancea R, Gorski A, Gorski H, Tudorache P. Adaptive learning using artificial intelligence in e-learning: A literature review. Educ Sci. 2023;13(12):1216. https://doi.org/10.3390/educsci13121216
https://doi.org/10.3390/educsci13121216
 
4 Joamets K, Chochia A. Access to artificial intelligence for persons with disabilities: Legal and ethical questions concerning the application of trustworthy AI. Acta Balt Hist Philos Sci. 2021;9(1):51. https://doi.org/10.11590/abhps.2021.1.04
https://doi.org/10.11590/abhps.2021.1.04
 
5 World Health Organization. International Classification of Functioning, Disability, and Health. WHO. c2001. p. 214.
 
6 CDC. Common Barriers to Participation Experienced by People with Disabilities. CDC. c2024 [cited 2024 December 14]. Available from: https://www.cdc.gov/ncbddd/disabilityandhealth/disability-barriers.html#:~:text=These %20include%20aspects%20such%20as:%20*%20a,*%20negative%20attitudes%20of%20people%20towards%20disability%2C
 
7 Olkin R. Could you hold the door for me? Including disability in diversity. Cult Divers Ethn Minor Psychol. 2002;8:130-7.
https://doi.org/10.1037/1099-9809.8.2.130
 
8 Brewer E, Brueggemann B, Hetrick N, Yergeau M. Introduction background and history. In B Brueggemann, editor. Arts and Humanities 2012;(pp. 1-62). Sage.
https://doi.org/10.4135/9781452218335.n1
 
9 National Center for Educational Statistics. Use of Support Among Students with Disabilities and Special Needs in College. NCES. c2022 [cited 2024 December 14]. Available from: https://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2022071
 
10 Return on Disability. Accommodations and Productivity: The Case Against Disclosure. Return on Disability. c2024 [cited 2024 December 14]. Available from: https://www.rod-group.com/research-insights/accommodations-and-productivity-the-case- against-disclosure/
 
11 Institute of Educational Technology. How the OU’s New Chatbot is Helping Disabled Students. IET. c2024 [cited 2024 December 21]. Available from: https://iet.open.ac.uk/research/new-chatbot-for-ou-disabled-students
 
12 Benetech. Bookshare. Benetech: c2024 [cited 2024 December 21]. Available from: https://benetech.org/work-area/bookshare/
 
13 Benetech. Nonprofit Uses AI to Make STEM Materials More Accessible. K-12 DIVE. c2022 [cited 2024 December 21]. Available from: https://www.k12dive.com/news/nonprofit-uses-ai-to- make-stem-materials-more-accessible/627442/
 
14 Ismail K, Sohel MH, Ayuniza UN. Technology social venture: A new genre of social entrepreneurship? Procedia Soc Behav Sci. 2012;40:429-34. https://doi.org/10.1016/j.sbspro.2012.03.211
https://doi.org/10.1016/j.sbspro.2012.03.211
 
15 Salvador-Carulla L, Gasca V. Defining disability, functioning, autonomy, and dependency in person-centered medicine and integrated care. Int J Integr Care. 2010;10 Suppl:3025. https://doi.org/10.5334/ijic.495
https://doi.org/10.5334/ijic.495
 
16 Katz S, Ford A, Moskowitz RW, Jackson BA, Jaffe MW. Studies of illness in the aged. The index of ADL: A standardized measure of biological and psychosocial function. J Am Med Assoc. 1963;185(12):914-9. https://doi.org/10.1001/jama.1963.03060120024016
https://doi.org/10.1001/jama.1963.03060120024016
 
17 Lawton MP, Brody EM. The functional assessment in rehabilitation of elderly people: Self maintaining and instrumental activities of daily living. Gerontologist. 1969;9:179-86.
https://doi.org/10.1093/geront/9.3_Part_1.179
 
18 Webster S. OrCam’s MyEye Pro Device on Glasses Can Help Visually Impaired Users to Read. TechTimes. c2022 [cited 2024 December 21]. Available from: https://www.techtimes.com/articles/270220/20220107/orcams-myeye-pro-device-help-visually-impaired-users-read.htm
 
19 Nguyen X, Koopman J, van Genderen M, Stam H, Boon C. Artificial vision: The effectiveness of OrCam in patients with advanced inherited retinal dystrophies. Acta Ophthalmol. 2021;100(4):e986-3. https://doi.org/10.1111/aos.15001
https://doi.org/10.1111/aos.15001
 
20 Amore FM, Silvestri V, Guidobaldi M, Sulfaro M. Efficacy and patients’ satisfaction with the ORCAM MyEye device among visually impaired people: A multicenter study. J Med Syst. 2023;47(1). https://doi.org/10.1007/s10916-023-01908-5
https://doi.org/10.1007/s10916-023-01908-5
 
21 Microsoft. SeeingAI: An App for Visually Impaired People that Narrates the World Around You. Microsoft Garage. c2024 [cited 2024 December 28]. Available from: https://www.microsoft.com/en-us/garage/wall-of-fame/seeing-ai/#:~:text=Download%20from:%20*%20Microsoft%20Store%20support.%20*%20Certified%20Refurbished.%20*%20Flexible%20Payments
 
22 Dost B. How Can We Improve Microsoft’s “Seeing AI” Application? 2021. Medium. c2021 [cited 2024 December 28]. Available from: https://bengisudost.medium.com/how-can-we-improve-microsofts-seeing-ai-application-4cb4b074e0d6
 
23 Guo A, Kamar E, Vaughan JW, Wallach H, Morris M. Toward fairness in AI for people with disabilities SBG@a research roadmap. ACMSIGACCESS Access Comput. 2020;125:1. https://doi.org/10.1145/3386296.3386298
https://doi.org/10.1145/3386296.3386298
 
24 Newman-Griffis D, Rauchberg JS, Alharbi R, Hickman L, Hochheiser H. Definition drives design: Disability models and mechanisms of bias in AI technologies. ArXiv:2206.08287. https://doi.org/10.48550/arXiv.2206.08287
 
25 Newman-Griffis D, Swenor B, Valdez R, Mason G. Disability data futures: Achievable imaginaries for AI and disability data justice. 2024;ArXiv:2411.03885. https://doi.org/10.48550/arXiv.2411.03885
https://doi.org/10.31219/osf.io/j52n6
 
26 Gorman R. Disability data justice from the ground up: A practice-led, participatory co-design approach to building an AI search engine and data repository for local, national, and transnational disability organizations. Crit Stud Int Interdiscip J. 2024;18(1):28-43. https://doi.org/10.51357/cs.v18i1.229
https://doi.org/10.51357/cs.v18i1.229
 
27 Valencia S, Cave R, Kallarackal K, Seaver K, Terry M, Kane SK. “The less I type, the better”: How AI language models can enhance or impede communication for AAC users. CHI’23 Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems 2023;(Article 830, pp. 1-14).
https://doi.org/10.1145/3544548.3581560
 
28 Murero M, Vita S, D’Ancona G. Artificial intelligence for severe speech impairment: Innovative approaches to AAC and communication. Symposium on Psychology-based Technologies 2020;(pp. 1-5).
 
29 Griffiths T, Slaughter R, Waller A. Use of artificial intelligence (AI) in augmentative and alternate communication (AAC): Community consultation on risks, benefits and the need for a code of practice. J Enabling Technol. 2024;18(4):232-47. https://doi.org/10.1108/JET-01-2024-0007
https://doi.org/10.1108/JET-01-2024-0007
 
30 Almufareh MF, Kausar S, Humayun M, Tehsin S. A conceptual model for inclusive technology: Advancing disability inclusion through artificial intelligence. J Disabil Res. 2024;3(1). https://doi.org/10.57197/JDR-2023-0060
https://doi.org/10.57197/JDR-2023-0060
 
31 El Moor C, Kundi B, Gorman R. AI and disability: A systematic scoping review. Health Inform J. 2024;30(3). https://doi.org/10.1177/14604582241285743
https://doi.org/10.1177/14604582241285743
 
32 Krupiy T, Scheinin M. Disability discrimination in the digital realm: How the ICRPD applies to artificial intelligence decision-making processes and helps in determining the state of international human rights law. Human Rights Law Rev. 2023;23(3). https://doi.org/10.1093/hrlr/ngad019
https://doi.org/10.1093/hrlr/ngad019
 
33 Marks M. Algorithmic disability discrimination. In I Glenn Cohen, Shachar C, Silvers A, Stein MA, editors. Disability, Health, Law, and Bioethics 2020. Cambridge University Press.
https://doi.org/10.1017/9781108622851.026
 
34 Tsang M. Building trust for data sourcing with the disabled community to build robust AI systems. 2021. IEEE Xplore. https://doi.org/10.1109/ISTAS52410.2021.9629180
https://doi.org/10.1109/ISTAS52410.2021.9629180
 
35 Kumar V, Barik S, Aggarwal S, Kumar D, Raj V. The use of artificial intelligence for persons with disability: A bright and promising future ahead. Disabil Rehabil Assist Technol. 2024;19(6). https://doi.org/10.1080/17483107.2023.2288241
https://doi.org/10.1080/17483107.2023.2288241
 
36 Khan MR. Role of AI in enhancing accessibility for people with disabilities. Adv Artif Intell Interdiscip Front. 2024;3(1):142. https://doi.org/10.60087/jaigs.vol03.issue01.p142
https://doi.org/10.60087/jaigs.vol03.issue01.p142
 
37 Wulandari C, Firdaus FA, Saifulloh F. Promoting inclusivity through technology: A literature review in educational settings. J Learn Technol. 2024;3(1):19-28. https://doi.org/10.33830/jlt.v3i1.9731
https://doi.org/10.33830/jlt.v3i1.9731
 
38 Buyl M, Cociancig C, Frattone C, Roekens N. Tackling algorithmic disability discrimination in the hiring process: An ethical, legal, and technical analysis. FAccT’22 Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency 2022;(pp. 1071-82). https://doi.org/10.1145/3531146.3533169
https://doi.org/10.1145/3531146.3533169
 

Cite this article as:
Ferebee S. AI and Accessibility: Breaking Barriers for People with Disabilities. Premier Journal of Artificial Intelligence 2025;2:100012

Export to EndNote Export to Mendeley