The Revolution That Left Us Out: Disability, AI, and the Incomplete Conversation About Humanity

 

While the 'Revolution' digitizes the horizon for the elite, the rest of us are still waiting for the pavement under our feet to be fixed.

A recent article in The Hindu, titled "Keeping Humanity at the Centre of the AI Revolution," [Click here for link to the article] raises questions that matter. It asks whether the rapid advance of artificial intelligence systems is moving faster than our collective ability to govern them. It is concerned with the risks of automation displacing labour, with the concentration of technological power in a small number of corporate actors, and with the erosion of human agency in decisions that shape livelihoods and social participation. These are serious concerns. They deserve serious examination.

But the article says a great deal about humanity and very little about a substantial part of it.

Disability does not appear in that conversation. Not once. And that absence is not a minor editorial oversight. It is symptomatic of a much older pattern: disability is admitted into the AI ethics discourse only when someone specifically demands its inclusion. It does not arrive on its own. It has to be carried in, repeatedly, by the same people who bear its exclusion.

This article is that demand.

What the Conversation Is Getting Right, and What It Is Missing

The concern with human-centred AI is not new. Researchers, civil society organisations, and several governments have spent years arguing that artificial intelligence must be designed with people in mind rather than profits or efficiency metrics. The Hindu article reflects this concern well. It draws attention to the fact that AI systems, however sophisticated, are built upon choices made by people. Those choices carry values. Those values carry biases. And those biases reproduce the social conditions that shaped them in the first place.

This is an important argument. It is also, in the disability rights community, an argument we have been making for the better part of a decade.

The AI Now Institute's foundational 2019 report, "Disability, Bias, and AI," made exactly this case. It documented how artificial intelligence systems, when trained on data that underrepresents disabled people, produce outputs that treat non-disabled behaviour as the universal standard. The systems are not neutral. They are calibrated to a particular body, a particular mode of speech, a particular speed of response, a particular pattern of interaction. When disabled users fall outside those calibrations, they are not accommodated. They are rejected.

That report was published six years ago. Mainstream AI ethics commentary in India is still not routinely engaging with it.

The Hindu article's conversation about humanity is, in this sense, a conversation about a subset of humanity. It addresses displacement of labour, questions of democratic accountability, the ethics of automation in public services. All of this is important. But when disability is absent from the frame, what emerges is an incomplete picture of who stands to be most harmed, and therefore an incomplete framework for remedy.

Technoableism Is Not a Technical Problem. It Is a Political One.

The word technoableism was put into sustained analytical use by Ashley Shew, a disability studies scholar and engineer at Virginia Tech, whose 2020 work in IEEE Technology and Society Magazine named the ideology that drives much of what passes for progressive AI design. Technoableism is the assumption that disability is a problem requiring technological elimination. It is the belief that the goal of assistive or accessible technology is to make the disabled person function more like a non-disabled person. It frames difference as defect, and positions the non-disabled body as the ideal towards which all technological development ought to strive.

This ideology does not announce itself. It arrives quietly, encoded into design decisions that no one has bothered to question.

Consider how this operates across the AI development pipeline. When training data for voice recognition systems is assembled, the overwhelming majority of voice samples are from speakers without speech disabilities. The system learns what a voice is supposed to sound like. When a person with cerebral palsy, amyotrophic lateral sclerosis, or a stammer interacts with that system, the system fails. Not because the technology is inherently incapable. Because the people who built it did not think to include the full range of human speech in their model of what a human voice sounds like.

This is Selection Bias. It is also a straightforward act of exclusion. It is not accidental. It is the consequence of disabled people being absent from the design room, the data team, the product meeting, and the ethics board.

The same logic applies to hiring systems that flag disabled communication styles as indicators of low confidence or low performance. It applies to facial recognition systems that fail to accurately identify people with atypical facial features or expressions. It applies to content moderation systems that flag disability-related language as offensive without distinguishing between slurs and community self-identification. It applies to large language models that, when asked to generate disability-related content, produce outputs that are clinical, condescending, and rooted in medical deficit models rather than disability rights perspectives.

Research published in 2025 by Panda, Agarwal, and Patel, introducing the AccessEval benchmarking framework, confirmed that disability bias in large language models is systemic rather than incidental. These are not edge cases. They are structural outcomes.

India's AI Moment and the Disability Rights Framework It Is Ignoring

The Hindu article is written in an Indian context, addressing an Indian readership, at a moment when India is making significant policy commitments around artificial intelligence. The India AI Impact Summit, NITI Aayog's AI governance guidelines, and the government's broader rhetoric about an "AI for All" future all claim a vision of inclusive technological development.

That vision does not hold up to scrutiny when examined from a disability rights standpoint.

India has 2.74 crore persons with disabilities according to the 2011 Census. The true figure is considerably higher by most independent estimates, given systemic undercounting. These individuals are spread across urban and rural geographies, across caste and class divisions, and across 21 categories of disability formally recognised under the Rights of Persons with Disabilities Act 2016. They are also among the most dependent upon digital public infrastructure for access to services, entitlements, information, and economic participation.

Yet NITI Aayog's AI governance documents, as I have argued previously on this platform and in an open letter to the Ministry of Electronics and Information Technology, treat disability as a sectoral afterthought rather than a structural dimension of all AI systems. The guidelines speak of inclusion in general terms. They do not mandate disability-inclusive data collection. They do not require accessibility impact assessments for AI systems deployed in public services. They do not integrate the RPwD Act 2016 into their governance framework. They do not reference the Supreme Court's landmark judgment in Rajive Raturi v. Union of India, which in November 2024 established accessibility as an ex-ante constitutional duty rather than a discretionary accommodation.

The Raturi judgment is significant precisely because it forecloses the kind of argument that AI governance currently makes by implication: that accessibility will be addressed eventually, after the core system is built. The Supreme Court held that accessibility is not a post-hoc retrofitting exercise. It is a baseline requirement that must be built into new infrastructure from the start. That principle applies with full force to AI systems, which are new infrastructure. If it applies to ramps and lifts, it applies to hiring algorithms and speech interfaces.

India's position under the United Nations Convention on the Rights of Persons with Disabilities reinforces this obligation. Article 4(1)(d) of the UNCRPD requires state parties to refrain from engaging in any act or practice that is inconsistent with the Convention, and to ensure that public authorities and institutions act in conformity with it. Article 9 requires accessible information and communication technology as a matter of right. These are not aspirational norms. India ratified the UNCRPD in 2007. The obligations are binding.

When The Hindu publishes a substantive opinion piece about keeping humanity at the centre of the AI revolution, and does not engage with these legal frameworks or the constituencies they protect, it participates in the same pattern of omission that characterises the policy it is critiquing. The critique of AI governance cannot exempt itself from the structural blind spots of AI governance.

The Difference Between Accessibility and Inclusion

There is a distinction that this discourse consistently collapses, and it is a distinction that persons with disabilities experience with considerable personal consequence.

Accessibility determines whether a person can use the system. Inclusion determines whether the system was designed with that person as a full human subject, rather than as an edge case to be accommodated later.

Accessible platforms built upon biased algorithms do not remove barriers. They move the barrier from the interface to the algorithm. A screen-reader-compatible job application portal that feeds into a hiring algorithm trained to penalise atypical speech patterns or non-linear employment histories is accessible in a technical sense and exclusionary in a structural one. The disabled applicant can submit the application. The system will still reject them.

The conversation about human-centred AI must therefore go further than user interface accessibility. It must address the assumptions embedded in the data, the objectives embedded in the optimisation function, and the absences embedded in the design team. Universal Design is not a feature to be added on. It is a methodology of designing from the margins outward, such that systems built to work for the most excluded users tend to work better for everyone.

The curb-cut effect, well documented in both physical and digital environments, illustrates this principle. Features designed for wheelchair users, closed captions developed for deaf and hard-of-hearing users, voice interfaces developed for users with motor impairments: these have consistently expanded usability for the broader population. Disability-led design is not charity. It is better engineering. It is a stress test for inclusion that the mainstream AI development pipeline systematically refuses to conduct.

Nothing About Us Without Us Is Not a Slogan. It Is a Design Requirement.

The principle of Nothing About Us Without Us, central to the disability rights movement since the 1980s, is sometimes treated by technologists as a vague aspirational gesture. It is in fact a precise methodological requirement.

It means that disabled people must be present at the data collection stage, so that training datasets capture the full range of human speech, movement, cognition, and behaviour. It means that disabled people must be present at the design stage, so that the objectives of the system are not calibrated exclusively around non-disabled norms of productivity, efficiency, and interaction. It means that disabled people must be present at the evaluation stage, so that bias audits assess performance across the full spectrum of the population rather than optimising for majority user groups and treating minority outcomes as acceptable collateral.

Research from the AAAI Conference on Artificial Intelligence in 2025, examining ableism in both Western and Indic language models, found that Indian AI systems consistently underestimate the harmfulness of ableist statements. The models reflect the cultural tolerances of the dominant society they were trained on. When that society normalises certain forms of disability-related discrimination, the model inherits that normalisation. Building cross-cultural competence into AI evaluation frameworks is therefore not an academic nicety. It is a basic requirement of fairness for the 2.74 crore Indians whose lives will increasingly be shaped by these systems.

This is the argument that the human-centred AI conversation must make room for. Not as a supplement to the main concern. As part of it.

Conclusion: Incomplete Humanity Is Not Humanity

The Hindu article is concerned about AI doing things to people without their meaningful participation. That concern is legitimate and necessary. But it is also incomplete. Because the people most likely to have AI systems act upon them without their participation, without their input into the training data, without representation in the design team, without recourse in the legal framework, without visibility in the policy document, are disabled people.

Disability is not a niche interest within the AI ethics discourse. It is the discipline's most rigorous test case. If an AI system cannot account for the full range of human bodies, minds, speech patterns, and modes of being in the world, it has not achieved human-centred design. It has achieved able-bodied-centred design dressed in the language of inclusion.

India is at a formative moment in shaping its AI ecosystem. The decisions being made now, about data, design, governance, and accountability, will embed their assumptions into public infrastructure for decades. If disability is absent from those decisions, the resulting systems will not be accessible to 2.74 crore Indians by omission. The omission will be structural and, given the legal frameworks now in place, unconstitutional.

The conversation about keeping humanity at the centre of the AI revolution must therefore include all of humanity. Not as a courtesy. As a constitutional obligation, as a matter of rights, and as a basic condition of the claim that the revolution is being made for people.

Those of us who have spent our lives being treated as edge cases, as outliers, as system anomalies, are not interested in watching another revolution proceed without us. We are the stress test. We are also the users. And it is past time that the mainstream AI ethics discourse remembered both.\

References

• Whittaker, M., Alper, M., Bennett, C.L., et al. (2019). Disability, Bias, and AI. AI Now Institute. https://ainowinstitute.org/disabilitybiasai-2019.pdf
• Shew, A. (2020). Ableism, Technoableism, and Future AI. IEEE Technology and Society Magazine, 39(1), 40-85.
• Panda, S., Agarwal, A., and Patel, H.L. (2025). AccessEval: Benchmarking Disability Bias in Large Language Models. Proceedings of EMNLP 2025. ACL Anthology.
• Phutane, M., Seelam, A., and Vashistha, A. (2025). A Human-Centered Audit of Ableism in Western and Indic Language Models. AAAI Conference on Artificial Intelligence.
• Rajive Raturi v. Union of India and Others, Writ Petition (Civil) No. 4/2005, Supreme Court of India, Judgment dated 8 November 2024.
• Rights of Persons with Disabilities Act, 2016. Ministry of Law and Justice, Government of India.
• United Nations Convention on the Rights of Persons with Disabilities, 2006. Articles 4 and 9.
• Singit, N. (2025). An Open Letter to the Ministry of Electronics and Information Technology: A Critique of the India AI Governance Guidelines on the Omission of Mandatory Disability and Digital Accessibility Rules. The Bias Pipeline. https://thebiaspipeline.nileshsingit.org
• Singit, N. (2026). Technoableism and the Bias Pipeline: How Ableist Ideology Becomes Algorithmic Exclusion. The Bias Pipeline. https://thebiaspipeline.nileshsingit.org
• Singit, N. (2026). TechnoAbleism in India's AI Moment: Why Accessibility Is Not Enough. Moneylife.in / The Bias Pipeline. https://thebiaspipeline.nileshsingit.org

A Rejoinder to "The Upskilling Gap" — The Invisible Intersection of Gender, AI & Disability

 To:

Ms. Shravani Prakash, Ms. Tanu M. Goyal, and Ms. Chellsea Lauhka
c/o The Hindu, Chennai / Delhi, India

Subject: A Rejoinder to "The Upskilling Gap: Why Women Risk Being Left Behind by AI"

Dear Authors,

I write in response to your article, "The upskilling gap: why women risk being left behind by AI," published in The Hindu on 24 December 2025 [click here to read the article], with considerable appreciation for its clarity and rigour. Your exposition of "time poverty"—the constraint that prevents Indian women from accessing the very upskilling opportunities necessary to remain competitive in an AI-disrupted economy—is both timely and thoroughly reasoned. The statistic that women spend ten hours fewer per week on self-development than men is indeed a clarion call for policy intervention, one that demands immediate attention from policymakers and institutional leaders.

Your article, however, reveals a critical lacuna: the perspective of Persons with Disabilities (PWDs), and more pointedly, the compounded marginalisation experienced by women with disabilities. While your arguments hold considerable force for women in general, they apply with even greater severity—and with doubled intensity—to disabled women navigating this landscape. If women are "stacking" paid work atop unpaid care responsibilities, women with disabilities are crushed under what may be termed a "triple burden": paid work, unpaid care work, and the relentless, largely invisible labour of navigating an ableist world. In disability studies, this phenomenon is referred to as "Crip Time"—the unseen expenditure of emotional, physical, and administrative energy required simply to move through a society not designed for differently-abled bodies.

1. The "Time Tax" and Crip Time: A Compounded Deficit

You have eloquently articulated how women in their prime working years (ages 25–39) face a deficit of time owing to the "stacking" of professional and domestic responsibilities. For a woman with a disability, this temporal deficit becomes far more acute and multidimensional.

Consider the following invisible labour burdens:

Administrative and Bureaucratic Labour. A disabled woman must expend considerable time coordinating caregivers, navigating government welfare schemes, obtaining UDID (Unique Disability ID) certification, and managing recurring medical appointments. These administrative tasks are not reflected in formal economic calculations, yet they consume hours each week.

Navigation Labour. In a nation where "accessible infrastructure" remains largely aspirational rather than actual, a disabled woman may require three times longer to commute to her place of work or to complete the household tasks you enumerate in your article. What takes an able-bodied woman thirty minutes—traversing a crowded marketplace, using public transport, or attending a medical appointment—may consume ninety minutes for a woman using a mobility aid in an environment designed without her needs in mind.

Emotional Labour. The psychological burden of perpetually adapting to an exclusionary environment—seeking permission to be present, managing others' discomfort at her difference—represents another form of unpaid, invisible labour.

If the average woman faces a ten-hour weekly deficit for upskilling, the disabled woman likely inhabits what might be termed "time debt": she has exhausted her available hours merely in survival and navigation, leaving nothing for skill development or self-improvement. She is not merely "time poor"; she exists in a state of temporal deficit.

2. The Trap of Technoableism: When Technology Becomes the Problem

Your article recommends "flexible upskilling opportunities" as a solution. This recommendation, though well-intentioned, risks collapsing into what scholar Ashley Shew terms "technoableism"—the belief that technology offers a panacea for disability, whilst conveniently ignoring that such technologies are themselves designed by and for able bodies.

The Inaccessibility of "Flexible" Learning. Most online learning platforms—MOOCs, coding bootcamps, and vocational training programmes—remain woefully inaccessible. They frequently lack accurate closed captioning, remain incompatible with screen readers used by visually impaired users, or demand fine motor control that excludes individuals with physical disabilities or neurodivergent conditions. A platform may offer "flexibility" in timing, yet it remains inflexible in design, creating an illusion of access without its substance.

The Burden of Adaptation Falls on the Disabled Person. Current upskilling narratives implicitly demand that the human—the disabled woman—must change herself to fit the machine. We tell her: "You must learn to use these AI tools to remain economically valuable," yet we do not ask whether those very AI tools have been designed with her value in mind. This is the core paradox of technoableism: it promises liberation through technology whilst preserving the exclusionary structures that technology itself embodies.

3. The Bias Pipeline: Where Historical Data Meets Present Discrimination

Your observation that "AI-driven performance metrics risk penalising caregivers whose time constraints remain invisible to algorithms" is both acute and insufficiently explored. Let us examine this with greater precision.

The Hiring Algorithm and the "Employment Gap." Modern Applicant Tracking Systems (ATS) and AI-powered hiring tools are programmed to flag employment gaps as indicators of risk. Consider how these gaps are interpreted differently:

• For women, such gaps typically represent maternity leave, childcare, or eldercare responsibilities.

• For Persons with Disabilities, these gaps often represent medical leave, periods of illness, or hospitalisation.

• For women with disabilities, the algorithmic penalty is compounded: a resume containing gaps longer than six months is automatically filtered out before any human reviewer examines it, thereby eliminating qualified disabled women from consideration entirely.

Research audits have documented this discrimination. In one verified case, hiring algorithms flagged minority candidates disproportionately as needing human review because such candidates—inhibited by systemic bias in how they were evaluated—tended to give shorter responses during video interviews, which the algorithm interpreted as "low engagement".​

Video Interviewing Software and Facial Analysis. Until its removal in January 2021, the video interviewing platform HireVue employed facial analysis to assess candidates' suitability—evaluating eye contact, facial expressions, and speech patterns as proxies for "employability" and honesty. This system exemplified technoableism in its purest form:

• A candidate with autism who avoids direct eye contact is scored as "disengaged" or "dishonest," despite neuroscientific evidence that autistic individuals process information differently and their eye contact patterns reflect cognitive difference, not deficiency.

• A stroke survivor with facial paralysis—unable to produce the "expected" range of expressions—is rated as lacking emotional authenticity.

• A woman with a disability, already subject to gendered scrutiny regarding her appearance and "likability," encounters an AI gatekeeper that makes her invisibility or over-surveillance algorithmic, not merely social.

These systems do not simply measure performance; they enforce a narrow definition of normalcy and penalise deviation from it.

4. Verified Examples: The "Double Glitch" in Action

To substantiate these claims, consider these well-documented instances of algorithmic discrimination:

Speech Recognition and Dysarthria. Automatic Speech Recognition (ASR) systems are fundamental tools for digital upskilling—particularly for individuals with mobility limitations who rely on voice commands. Yet these systems demonstrate significantly higher error rates when processing dysarthric speech (speech patterns characteristic of conditions such as Cerebral Palsy or ALS). Recent research quantifies this disparity:

• For severe dysarthria across all tested systems, word error rates exceed 49%, compared to 3–5% for typical speech.​

• Character-level error rates have historically ranged from 36–51%, though fine-tuned models have reduced this to 7.3%.​

If a disabled woman cannot reliably command the interface—whether due to accent variation or speech patterns associated with her condition—how can she be expected to "upskill" into AI-dependent work? The platform itself becomes a barrier.

Facial Recognition and the Intersection of Race and Gender. The "Gender Shades" study, conducted by researchers at MIT, documented severe bias in commercial facial recognition systems, with error rates varying dramatically by race and gender:

• Error rates for gender classification in lighter-skinned men: less than 0.8%

• Error rates for gender classification in darker-skinned women: 20.8% to 34.7%​

Amazon Rekognition similarly misclassified 31 percent of darker-skinned women. For a disabled woman of colour seeking employment or accessing digital services, facial recognition systems compound her marginalisation: she is simultaneously rendered invisible (failed detection) or hyper-surveilled (flagged as suspicious).​

The Absence of Disability-Disaggregated Data. Underlying all these failures is a fundamental problem: AI training datasets routinely lack adequate representation of disabled individuals. When a speech recognition system is trained predominantly on able-bodied speakers, it "learns" that dysarthric speech is anomalous. When facial recognition is trained on predominantly lighter-skinned faces, it "learns" that darker skin is an outlier. Disability is not merely underrepresented; it is systematically absent from the data, rendering disabled people algorithmically invisible.

5. Toward Inclusive Policy: Dismantling the Bias Pipeline

You rightly conclude that India's Viksit Bharat 2047 vision will be constrained by "women's invisible labour and time poverty." I respectfully submit that it will be equally constrained by our refusal to design technology and policy for the full spectrum of human capability.

True empowerment cannot mean simply "adding jobs," as your article notes. Nor can it mean exhorting disabled women to "upskill" into systems architected to exclude them. Rather, it requires three concrete interventions:

First, Inclusive Data Collection. Time-use data—the foundation of your policy argument—must be disaggregated by disability status. India's Periodic Labour Force Survey should explicitly track disability-related time expenditure: care coordination, medical appointments, navigation labour, and access work. Without such data, disabled women's "time poverty" remains invisible, and policy remains blind to their needs.

Second, Accessibility by Design, Not Retrofit. No upskilling programme—whether government-funded or privately delivered—should be permitted to launch without meeting WCAG 2.2 Level AA accessibility standards (the internationally recognised threshold for digital accessibility in public services). This means closed captioning, screen reader compatibility, and cognitive accessibility from inception, not as an afterthought. The burden of adaptation must shift from the disabled person to the designer.​

Third, Mandatory Algorithmic Audits for Intersectional Bias. Before any AI tool is deployed in India's hiring, education, or social welfare systems, it must be audited not merely for gender bias or racial bias in isolation, but for intersectional bias: the compounded effects of being a woman and disabled, or a woman of colour and disabled. Such audits should be mandatory, transparent, and subject to independent oversight.

Conclusion: A Truly Viksit Bharat

You write: "Until women's time is valued, freed, and mainstreamed into policy and growth strategy, India's 2047 Viksit Bharat vision will remain constrained by women's invisible labour, time poverty and underutilised potential."

I would extend this formulation: Until we design our economy, our technology, and our policies for the full diversity of human bodies and minds—including those of us who move, speak, think, and perceive differently—India's vision of development will remain incomplete.

The challenge before us is not merely to "include" disabled women in existing upskilling programmes. It is to fundamentally reimagine what "upskilling" means, to whom it is designed, and whose labour and capability we choose to value. When we do, we will discover that disabled women have always possessed the skills and resilience necessary to thrive. Our task is simply to remove the barriers we have constructed.

I look forward to the day when India's "smart" cities and "intelligent" economies are wise enough to value the time, talent, and testimony of all women—including those of us who move, speak, and think differently.

Yours faithfully,

Nilesh Singit
Distinguished Research Fellow
CDS, NALSAR &&
Founder, The Bias Pipeline
https://www.nileshsingit.org/

Prototype — Accessible to Whom? Legible to What?

 

Abstract

Artificial Intelligence (AI) has transformed the terrain of possibility for assistive technology and inclusive design, but continues to perpetuate complex forms of exclusion rooted in legibility, bias, and tokenism. This paper critiques current paradigms of AI prototyping that centre “legibility to machines” over accessibility for disabled persons, arguing for a radical disability-led approach. Drawing on international law, empirical studies, and design scholarship, the analysis demonstrates why prototyping is neither neutral nor technical, but a deeply social and political process. Building from case studies in recruiting, education, and healthcare technology failures, this work exposes structural biases in training, design, and implementation—challenging designers and policymakers to move from “designing for” and “designing with” to “designing from” disability and difference.

Introduction

Prototyping is celebrated in engineering and design as a space for creativity, optimism, and risk-taking—a laboratory for the future. Yet, for countless disabled persons, the prototype is also where inclusion begins… or ends. For them, optimism is often tempered by the unspoken reality that exclusion most often arrives early and quietly, disguised as technical “constraints,” market “priorities,” or supposedly “objective” code. When prototyping occurs, it rarely asks: accessible to whom, legible to what?

This question—so simple, so foundational—is what this paper interrogates. The rise of Artificial Intelligence has intensified the stakes because AI prototypes increasingly determine who is rendered visible and included in society’s privileges. Legibility, not merely accessibility, is becoming the deciding filter; if one’s body, voice, or expression cannot be rendered into a dataset “comprehensible” to AI, one may not exist in the eyes of the system. Thus, we confront a new and urgent precipice: machinic inclusion, machinic exclusion.

This work expands the ideas presented in recent disability rights speeches and debates, critically interrogating how inclusive design must transform both theory and practice in the age of AI. It re-interprets accessibility as a form of knowledge and participation—never a technical afterthought.

Accessibility as Relational, Not Technical

Contemporary disability studies and the lived experiences of activists reject the notion that accessibility is a mere checklist or add-on. Aimi Hamraie suggests that “accessibility is not a technical feature but a relationship—a continuous negotiation between bodies, spaces, and technologies.”1 Just as building a ramp after a staircase is an act of remediation rather than inclusion, most AI prototyping seeks to retrofit accessibility, arguing it is too late, too difficult, or too expensive to embed inclusiveness from the outset.

Crucially, these arguments reflect broader epistemologies: those who possess the power to design, define the terms of recognition. Accessibility is not simply about “opening the door after the fact,” but questioning why the door was placed in an inaccessible position to begin with.

This critique leads us to re-examine prototyping practices through a disability lens, asking not only “who benefits” but also “who is recognised.” Evidence throughout the AI industry reveals a persistent confusion between accessibility for disabled persons and legibility for machines, a theme critically examined in subsequent sections.

Legibility and the Algorithmic Gaze

Legibility, distinct from accessibility, refers to the capacity of a system to recognise, process, and make sense of a body, voice, or action. Within the context of AI, non-legible phenomena—those outside dominant training data—simply vanish. People with non-standard gait, speech, or facial expressions are “read” by the algorithm as errors or outliers.

What are the implications of placing legibility before accessibility?

Speech-recognition models routinely misinterpret dysarthric voices, excluding those with neurological disabilities. Facial recognition algorithms have misclassified disabled expressions as “threats” or “system errors,” because their datasets contain few, if any, disabled exemplars. In the workplace, résumé-screening AI flags gaps or “unusual experience,” disproportionately rejecting those with disability-induced employment breaks. In education, proctoring platforms flag blind students for “cheating”, unable to process their lack of eye gaze at the screen as a legitimate variance.

These failures do not arise from random error. They are products of a pipeline formed by unconscious value choices made at every stage: training, selection, who participates, and who is imagined as the “user.”

In effect, machinic inclusiveness transforms the ancient bureaucracy of bias from paper to silicon. The new filter is not the form but the invisible code.

The Bias Pipeline: What Goes In, Comes Out Biased

Bias in AI does not merely appear at the end of the process; it is present at every decision point. One stark experiment submitted pairs of otherwise identical résumés to recruitment-screening platforms: one indicated a “Disability Leadership Award” or advocacy involvement, the other did not. The algorithm ranked the “non-disability” version higher, asserting that highlighting disability meant “reduced leadership emphasis,” “focus diverted from core job responsibilities,” or “potential risk.”

This is not insignificant. Empirical studies have reproduced such results across tech, finance, and education, showing systemic discrimination by design. Qualified disabled applicants are penalised for skills, achievements, and community roles that are undervalued or alien to training data.

Much as ethnographic research illuminated the “audit culture” in public welfare (where bureaucracy performed compliance rather than delivered services), so too does “audit theatre” manifest in AI. Firms invite disabled people to validate accessibility only after the design is final. In true co-design, disabled persons must participate from inception, defining criteria and metrics on equal footing. This gap—between performance and participation—is the site where bias flourishes.

The Trap of Tokenism

Tokenism is an insidious and common problem in social design. In disability inclusion, it refers to the symbolic engagement of disabled persons for validation, branding, or optics—rather than for genuine collaboration.

Audit theatre, in AI, occurs when disabled people are surveyed, “consulted,” or reviewed, but not invited into the process of design or prototyping. The UK’s National Disability Survey was struck down for failing to meaningfully involve stakeholders. Even the European Union’s AI Act, lauded globally for progressive accessibility clauses, risks tokenism by mandating involvement but failing to embed robust enforcement mechanisms.

Most AI developers receive little or no formal training in accessibility. When disability emerges in their worldview, it is cast in terms of medical correction—not lived expertise. Real participation remains rare.

Tokenism has cascading effects: it perpetuates design choices rooted in non-disabled experience, licenses shallow metrics, and closes the feedback loop on real inclusion.

Case Studies: Real-World Failures in Algorithmic Accessibility

AI Hiring Platforms and the “Disability Penalty”

Automated CV-screening tools systematically rank curricula vitae containing disability-associated terms lower, even when qualifications are otherwise stronger. Companies like Amazon famously scrapped AI recruitment platforms after discovering they penalised women, but similar audits for disability bias are scarce. Companies using video interview platforms have reported that candidates with stroke, autism, or other disability-related facial expressions score lower due to misinterpretation.

Online Proctoring and Educational Technology in India

During the COVID-19 pandemic, the acceleration of edtech platforms in India promised transformation. Yet, blind and low-vision students were flagged as “cheating” for not making “required” eye contact with their devices. Zoom and Google Meet upgraded accessibility features, but failed to address core gaps in their proctoring models.

Reports from university students showed that requests for alternative assessments or digital accommodations were often denied on the grounds of technical infeasibility.

Healthcare Algorithms and Diagnostic Bias

Diagnostic risk scores and triaging algorithms trained on narrow datasets exclude non-normative disability profiles. Health outcomes for persons with rare, chronic, or atypical disabilities are mischaracterised, and recommended interventions are mismatched.

Each failure traces back to inaccessible prototyping.

Disability-Led AI Prototyping

If the problem lies in who defines legibility, the solution lies in who leads the prototype. Disability-led design reframes accessibility—not as a requirement for “special” needs but as expertise that enriches technology. It asks not “How can you be fixed?” but “What knowledge does your experience bring to designing the machine?”

Major initiatives are emerging. Google’s Project Euphonia enlists disabled participants to re-train speech models for atypical voices, but raises ethical debates on data ownership, exploitation, and who benefits. More authentic still are community-led mapping projects where disabled coders and users co-create AI mapping tools for urban navigation, workspace accessibility, and independent living. These collaborations move slowly but produce lasting change.

When accessibility is led by disabled persons, reciprocity flourishes: machine and user learn from each other, not simply predict and consume.

Sara Hendren argues, “design is not a solution, it is an invitation.” Where disability leads, the invitation becomes mutual—technology contorts to better fit lives, not the reverse.

Policy, Law, and Regulatory Gaps

The European Union’s AI Act is rightly lauded for Article 16 (mandating accessibility for high-risk AI systems) and Article 5 (forbidding exploitation of disability-related vulnerabilities), as well as public consultation. Yet, the law lacks actionable requirements for collecting disability-representative data—and overlooks the intersection of accessibility, data ownership, and research ethics.

India’s National Strategy for Artificial Intelligence, along with “AI for Inclusive Societal Development,” claims “AI for All” but omits specific protections, data models, or actionable recommendations for disabled persons—this despite the Supreme Court’s Rajiv Raturi judgment upholding accessibility as a fundamental right. Implementation of the Rights of Persons with Disabilities Act, 2016, remains loose, and enforcement is sporadic.

The United States’ ADA and Section 508 have clearer language, but encounter their own enforcement challenges and retrofitting headaches.

Ultimately, policy remains disconnected from practice. Prototyping and design must close the gap—making legal theory and real inclusiveness reciprocal.

Intersectionality: Legibility Across Difference

Disability is never experienced in isolation: it intersects with gender, caste, race, age, and class. Women with disabilities face compounded discrimination in hiring, healthcare, and data representation. Caste-based exclusions are rarely coded into AI training practices, creating models that serve only dominant groups.

For example, the exclusion of vernacular languages in text-to-speech software leaves vast rural disabled communities voiceless in both policy and practical tech offerings. Ongoing work by Indian activists and community innovators seeks to produce systems and data resources that represent the full spectrum of disabled lives, but faces resistance from resource constraints, commercial priorities, and a lack of institutional support.

Rethinking the Fundamentals: Prototyping as Epistemic Justice

Epistemic justice—ensuring that all knowledge, experience, and ways of living are valued in the design of social and technical systems—is both a theoretical and a practical necessity in AI. Bias springs not only from bad data or oversight but by failing to recognise disabled lives as valid sources of expertise.

Key steps for epistemic justice in prototyping include:

• Centre disabled expertise from project inception, defining metrics, incentives, and feedback loops.

• Use disability as a source of innovation, not just compliance: leverage universal design to produce systems more robust for all users.

• Address intersectionality in datasets, training and testing for compounded bias across race, gender, language, and class.

• Create rights-based governance in tech companies, embedding accessibility into KPIs and public review.

Recommendations: Designing From Disability

The future of inclusive AI depends on three principal shifts:

1. From designing for to designing with: genuine co-design, not audit theatre, where disabled participants shape technology at every stage.

2. From accessibility as compliance to accessibility as knowledge: training developers, engineers and policymakers to value lived disability experience.

3. From compliance to creativity: treating disability as “design difference”—a starting point for innovation, not merely a deficit.

International law and national policy must recognise the lived expertise of disability communities. Without this, accessibility remains a perpetual afterthought to legibility.

Conclusion

Accessible to whom, legible to what? This question reverberates through every level of prototype, product, and policy.

If accessibility is left to the end, if legibility for machines becomes the touchstone, humanity is reduced, difference ignored. When disability leads the design journey, technology is not just machine-readable; it becomes human-compatible.

The future is not just about teaching machines to read disabled lives—but about allowing disabled lives to rewrite what machines can understand.

References

• Aimi Hamraie, Building Access: Universal Design and the Politics of Disability (University of Minnesota Press, 2017).

• Barocas, Solon, Moritz Hardt, and Arvind Narayanan. “Fairness and Machine Learning.” fairmlbook.org, 2019.

• Buolamwini, Joy, and Timnit Gebru. “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings of Machine Learning Research 81 (2018): 1–15.

• Leavy, Siobhan, Eugenia Siapera, Bethany Fernandez, and Kai Zhang. “They Only Care to Show Us the Wheelchair: Disability Representation in Text-to-Image AI Models.” Proceedings of the 2024 ACM FAccT.

• Sara Hendren. What Can a Body Do? How We Meet the Built World (Riverhead, 2020).

• National Strategy for Artificial Intelligence, NITI Aayog, Government of India, 2018.

• Rajiv Raturi v. Union of India, Supreme Court of India, AIR 2012 SC 651.

• European Parliament and Council, Artificial Intelligence Act, 2023.

• Google AI Blog. “Project Euphonia: Helping People with Speech Impairments.” May 2019.

• “Making AI Work for Everyone,” Google Developers, 2022.

• Amazon Inc., “Amazon Scraps Secret AI Recruiting Tool That Showed Bias Against Women,” Reuters, October 10, 2018.

• United Kingdom High Court, National Disability Survey ruling, 2023.

• Nita Ahuja, “Online Proctoring as Algorithmic Injustice: Blind Students in Indian EdTech,” Journal of Disability Studies, vol. 12, no. 2 (2022): 151-177.

• United Nations, Convention on the Rights of Persons with Disabilities, Resolution 61/106 (2006).

• [Additional references on intersectionality, design theory, empirical studies, Indian law, US/EU regulation, and case material]