AI and Disability: Build It With Disabled Users From Day One

For many persons with disabilities, AI is already part of daily life. A blind student uses a phone app that describes a photograph or reads out a printed page. A deaf professional follows a meeting through live captions. A person with a motor disability writes emails by voice. These tools were expensive, rare or impossible a decade ago. Now many are free.

At the same time, AI is moving into public services: chatbots that answer questions about schemes, voice systems that take complaints, automated checks that decide who is eligible. If those systems are designed without persons with disabilities in mind, they can lock people out of services they have a legal right to.

What the law already requires

The Rights of Persons with Disabilities Act, 2016 requires that information and communication technology be accessible. Section 40 asks the government to set accessibility standards, including for ICT, and Section 42 asks it to ensure that content in audio, print and electronic media is available in accessible formats.

In November 2024, the Supreme Court made this sharper. In Rajive Raturi v. Union of India, it held that accessibility standards under the Act must be mandatory, not just guidance, and directed the Union government to frame binding rules. Any public digital service, including one powered by AI, falls under this duty.

Where AI and accessibility fail persons with disabilities

Voice systems that do not understand some voices

Speech recognition is trained mostly on typical speech. People with cerebral palsy, hearing impairments or speech disabilities are often misunderstood. A helpline that only works by voice, with no other option, can be unusable for them.

Interfaces that do not work with assistive technology

A chatbot embedded in a government website may look simple. But if it cannot be used with a screen reader, cannot be navigated by keyboard, or hides its answers in images, blind and low-vision users cannot use it at all.

Biometrics that assume every body is the same

Systems that rely on fingerprints, iris scans or face matching can fail for people whose bodies do not match the assumptions built into them. Without a simple alternative, a failed match can mean a missed ration, pension or payment.

Decisions trained on incomplete data

When AI is used to sort applications, flag fraud or predict need, it learns from past data. Persons with disabilities are often under-counted or misrecorded in that data. A system trained on it can quietly repeat old exclusions at greater speed.

Language that leaves people out

India's national language platform, Bhashini, supports translation and speech across the 22 scheduled languages. That is a major step. But plain language matters as much as the choice of language. Many persons with intellectual disabilities, and many people with limited schooling, need short sentences and simple words. AI tools can produce these, if they are designed to.

Five principles for public AI that works for everyone

1. Involve persons with disabilities from the start. Test early prototypes with blind, deaf, low-vision, physically disabled and neurodivergent users, and pay them for their time. Problems found at the design stage are cheap to fix. Problems found after launch are not.
2. Always offer more than one way in. Every service should work by text and by voice, by touch and by keyboard. No one should be forced to use a single channel that does not work for them.
3. Keep a human route. When an automated system fails, a person should be able to reach a trained human quickly, without having to start again.
4. Meet accessibility standards before launch, not after. Public digital services in India should follow the Guidelines for Indian Government Websites and accessibility standards for ICT. Accessibility testing should be a condition of going live.
5. Audit outcomes, not just accuracy. Check who is being approved, rejected or misunderstood by an AI system, and whether persons with disabilities are being left out at higher rates.

An opportunity worth getting right

India is investing heavily in AI. The IndiaAI Mission, approved in 2024 with an outlay of over Rs 10,000 crore, aims to build computing capacity, datasets and applications for public good. Persons with disabilities stand to gain a great deal from that investment, through better access to information, education, work and services.

The gains will only reach them if accessibility is designed in. That means disability expertise at the table when public AI systems are planned, not called in after something has gone wrong.

Sutra works on disability rights and on how public systems deliver. Making sure that new digital and AI-based services are accessible is a natural part of that work. If you are building or buying AI tools for public services, and want them to work for persons with disabilities, we would be glad to talk. Write to us at info@sutraindia.org.