Disabled Citizens and the Future of AI: Accessibility First

Source: Original editorial ↗ The Hindu

The Lift Line

“Disabled people should not be made to adapt to systems built without them; AI is the newest such system.”

Why This Editorial Matters for Your Exam

Navin Aade writes in The Hindu of 6 October 2026 on what artificial intelligence means for persons with disabilities in India. The article joins three exam themes: the rights of persons with disabilities (the 2016 Act and the Supreme Court’s Rajive Raturi judgment), bias in AI, and the environmental and grid costs of data centres. It ends with concrete recommendations, which makes it a ready source for a GS2 or GS3 answer and for an Essay on technology and inclusion.

GS Paper 2: Welfare schemes for vulnerable sections and the performance of these schemes; mechanisms, laws and institutions for their protection. GS Paper 3: Awareness in IT and computers; infrastructure: energy. GS Paper 4: Ethics of technology.

Background and Context

The promise. The writer describes a friend, blind since birth, who uses AI almost every day to read documents, understand photographs and complete government forms that his screen reader cannot open, all without a sighted helper.

The law.

Instrument	What it provides
Rights of Persons with Disabilities Act, 2016	Recognises 21 specified disabilities (up from 7 under the 1995 Act); Sections 40 to 46 on accessibility of the built environment, transport and information and communication technology
RPwD Rules, 2017, Rule 15	Accessibility standards; the Supreme Court found them non-mandatory in effect
Rajive Raturi v. Union of India (8 November 2024)	The key background judgment: accessibility is a facet of the right to life and dignity (Article 21); rules were “toothless”; Centre given three months to frame mandatory standards
Chief Commissioner for Persons with Disabilities	Statutory authority under the 2016 Act; has penalised 155 establishments, including ministries, for inaccessible websites and apps, the writer says
UN Convention on the Rights of Persons with Disabilities	Ratified by India in 2007; Article 9 on accessibility

The AI build-out. AI models are entering government services, recruitment and health care, and the Centre is courting about $200 billion in AI investment over the next two years, the writer notes. India’s data-centre capacity is projected to more than quadruple by 2030, to 6.5 GW or more.

The Analysis

1. The law is ahead of practice. The 2016 Act has required accessible digital services since 2019, but compliance “remains the exception”. The Supreme Court’s 2024 direction to frame mandatory standards within three months has not produced enough change; nearly two years later, the petitioners are back in court. AI is being deployed “on this ground”.

2. AI is not neutral. The writer cites three kinds of evidence:

Study	Finding
AccessEval benchmark, 21 language models, nine kinds of disability	Models became more error-prone, more negative in tone and more likely to stereotype when disability entered the question
CLIP image model	15 percentage points less accurate on photographs taken by blind and low-vision users than on web images; objects such as a white cane or Braille display up to 17 times rarer in training data
Everyday chatbots	A blind user asking about becoming a software engineer is often told “I’m sorry you’re blind”: the model answers a career question as if it were a report of loss

3. AI alone is not enough. On the NClude platform, which uses AI to help disabled people fill job applications and government forms on inaccessible websites, 2,462 users were surveyed: 1,313 completed a task previously closed to them, but only 543 through AI alone. The rest needed a staff member. A system without such a human fallback leaves users to improvise.

4. The grid problem. States compete for data centres with power subsidies and duty waivers. Maharashtra this year relaxed its renewable energy requirement for data centres from 100 to 51 per cent, the writer says. Few ask what the new load does to grids straining at summer peaks, or what server heat adds to hot cities. Disabled people who rely on powered wheelchairs or oxygen machines cannot switch them off, and are often the first stranded by a power cut or an inaccessible emergency alert.

5. Not a case against AI. The writer is explicit: disabled people should not be asked to give up tools they depend on in the name of sustainability, and the economy needs infrastructure. The argument is about how AI is built and governed.

6. Four safeguards. Test every government AI system for disability bias; build training datasets with genuine disability representation, gathered with consent; attach conditions on renewable sourcing and grid and heat impact to data-centre incentives; and give disabled Indians a voice in decisions from the data to the grid.

Data and Institutions Vault

Prelims-grade facts:

Law and institutions:

• RPwD Act, 2016: 21 disabilities; 4 per cent reservation in government posts for persons with benchmark disabilities (Section 34).
• 5 per cent of seats in government and aided higher education institutions for them (Section 32).
• Chief Commissioner for Persons with Disabilities: statutory, under the 2016 Act, with State Commissioners in each State.
• Accessible India Campaign (Sugamya Bharat Abhiyan): launched 3 December 2015, the International Day of Persons with Disabilities.
• Rajive Raturi v. Union of India (2024): accessibility part of Article 21; mandatory standards ordered.

Data:

• Census 2011: about 2.68 crore persons with disabilities, 2.21 per cent of the population.

International:

• UNCRPD: adopted 2006, ratified by India in 2007.

⚠️ Watch the trap: The 1995 Persons with Disabilities Act recognised 7 disabilities; the 2016 Act recognises 21, including acid attack victims, specific learning disabilities and autism spectrum disorder. Do not mix up the two Acts’ numbers.

The Debate

For the writer’s view. Each new technology has been built for the average user and fixed for disabled users later, if at all. With AI moving into hiring, welfare and health, biased models can deny opportunities at scale and invisibly. India’s own law and the Supreme Court already require accessibility; AI should be held to that standard before deployment, not after complaints.

The other side. AI tools are improving fast, and much of the independence the writer describes comes from general-purpose models that were not designed for disability at all. Mandatory audits for every system could slow adoption and raise costs. States also argue that cheap power and flexible renewable rules are needed to attract data-centre investment that brings jobs and tax revenue.

The balanced verdict. The answer is not to slow AI but to set clear, proportionate rules for its public use: accessibility and bias testing for government deployments, a guaranteed human alternative, representative data, and grid planning that accounts for new loads. These protect everyone, not just disabled users.

How to Think About This

Design for the edge, and the centre benefits. Many technologies first designed for disabled people, from captions and audiobooks to voice control and text-to-speech, became mainstream. When you write on inclusion in technology, use the idea of universal design: systems that work for people with disabilities usually work better for the elderly, the less literate and users of other languages. It turns an argument about a minority into an argument about quality for all.

Diagram-in-Words

Takeaway Box

• Thesis: AI’s promise will not reach disabled Indians on its own; it must be built and governed with them.
• Law: RPwD Act, 2016 (21 disabilities; digital accessibility since 2019); Rajive Raturi (2024): accessibility under Article 21, mandatory standards ordered.
• Evidence of bias: 21 models on AccessEval; CLIP 15 points less accurate on photos by blind users; canes and Braille displays up to 17 times rarer in data.
• Human fallback: of 1,313 NClude users who completed a task, only 543 did it through AI alone.
• Asks: bias tests for government AI; consented, representative data; conditions on data-centre incentives; a voice for disabled people.

Sources: The Hindu, op-ed by Navin Aade, 6 October 2026

Source: Disabled Citizens and the Future of AI: Accessibility First — Ujiyari.com | Free UPSC & State PCS Editorial Analysis

Key Arguments at a Glance

Navin Aade argues that AI offers disabled people real independence, but that this promise will not reach disabled Indians on its own: AI is being deployed in government services, recruitment and health care on top of a digital system that is already inaccessible, using models that are biased against disability, and powered by data centres whose strain on the grid falls hardest on disabled people.

✓ Supporting

• In Rajive Raturi (November 2024), the Supreme Court held accessibility to be a facet of the fundamental right to life and dignity, found India’s accessibility rules toothless, and gave the Centre three months to frame mandatory standards; nearly two years later, the petitioners are back in court because too little has changed.
• The Chief Commissioner for Persons with Disabilities has penalised 155 establishments, including ministries, for websites and apps that disabled citizens cannot use, although the Rights of Persons with Disabilities Act, 2016 has required accessible digital services since 2019.
• Studies show bias: on the AccessEval benchmark, 21 language models became more error-prone, more negative and more stereotyped when disability entered the question; the image model CLIP was 15 percentage points less accurate on photographs taken by blind and low-vision users, and objects such as a white cane or Braille display were up to 17 times rarer in training data.
• Of 2,462 users surveyed on the NClude platform, 1,313 completed a task previously closed to them, but only 543 did so through AI alone; the rest needed a staff member, which shows the need for human fallback.
• Data-centre capacity is projected to more than quadruple to 6.5 GW or more by 2030, States compete with power subsidies, and Maharashtra relaxed its renewable requirement for data centres from 100 to 51 per cent; disabled people who rely on powered wheelchairs or oxygen machines are among the first stranded by power cuts.

⚠ Counter

AI tools are improving quickly and have already given many disabled users more independence than any previous technology; heavy regulation could slow their spread. Data centres and cheaper power are also needed for economic growth, and States argue that strict renewable mandates would drive investment elsewhere.
The writer accepts both points: his case is not against AI or infrastructure.

→ Way Forward

The writer proposes four steps: test every AI system the government deploys for disability bias; build training datasets with genuine disability representation, gathered with consent; attach conditions on renewable sourcing and on grid and heat impact to State incentives for data centres; and give disabled Indians a voice in decisions from training data to the grid.