Role of AI in the future of disabled

What does accessibility mean for persons with disabilities?

• Accessibility means ensuring that persons with disabilities can independently access and use physical and digital systems, services and information.
• For AI, this includes:
  
• Screen-reader compatible interfaces for persons with visual disabilities.
• Accessible government forms and websites.
• AI tools that understand diverse disability-related needs.
• Assistive technologies that improve independence.
• AI can already help a blind user read documents, interpret photographs and navigate government forms. 

What is the constitutional and legal basis for digital accessibility?

• Article 14 – Equality before law.
• Article 21 – Right to life and dignity.
• Rights of Persons with Disabilities Act, 2016 – Provides a statutory framework for the rights and inclusion of persons with disabilities.
• Supreme Court's Rajive Raturi judgment (2024) – Held accessibility to be an aspect of the fundamental right to life and dignity and criticised accessibility rules for being treated as non-binding recommendations.
• The Court directed the Centre to frame mandatory accessibility standards.
• The Chief Commissioner for Persons with Disabilities had also penalised 155 establishments, including government ministries, for accessibility-related violations. 

How can AI improve the lives of persons with disabilities?

• Greater independence
  
• Enables visually impaired users to read documents and interpret images without depending entirely on another person.
• Can assist users in navigating otherwise inaccessible digital services.
• Access to employment and public services
  
• AI-based tools can help persons with disabilities complete job applications and government forms.
• The technology can potentially reduce dependence on human assistance.
• Assistive technology: AI can function as an interface between persons with disabilities and otherwise inaccessible information and services.

What is preventing AI from becoming inclusive? 

Structural factors

• Digital systems were not designed with accessibility at the core
  
• Many websites and applications remain inaccessible to persons with disabilities.
• This forces disabled citizens to depend on others or find workarounds.
• The problem predates AI but becomes more consequential as public services increasingly move online.
• Inadequate representation in AI training data
  
• AI systems learn from the data on which they are trained.
• Under-representation of persons with disabilities can therefore reproduce existing social exclusion in AI outputs.
• Disability bias in AI models: Research using AccessEval, a benchmark covering nine kinds of disability across 21 languages, found that language models became:
  
• More error-prone
• More negative in tone
• More likely to stereotype when disability entered the interaction
• Existing social assumptions can enter AI systems
  
• AI is not automatically neutral.
• Bias in training data can translate into biased responses.
• Example: Mainstream chatbot responding to a blind person's career-related question as though the person had suffered a loss.

Cyclical factors 

• Rapid deployment of AI without adequate accessibility checks:
  
• Absence of effective fallback mechanisms: 
    A survey of 2,462 users of the NClude platform, which uses AI to help disabled people navigate inaccessible websites and complete applications, found:
• This shows that AI can improve access but cannot yet substitute for human support in all cases.
• Rising energy requirements of AI infrastructure:
  
• India's data-centre capacity is projected to rise to more than 6.5 GW by 2030, more than four times current capacity.
• This creates another vulnerability for disabled citizens who depend on electricity-powered:
• Power cuts or grid stress can therefore have greater consequences for persons with disabilities.
• Ex: Maharashtra reduced its renewable-energy requirement for data centres from 100% to 51%, illustrating the tension between rapid AI infrastructure expansion and energy sustainability.

How should India approach AI and disability inclusion? 

Incremental measures 

• Test AI systems for disability bias: Government-deployed AI systems should undergo specific disability-bias testing before deployment.
• Improve representation in training datasets
  
• Training datasets should contain genuine disability representation.
• Disability-related data should be gathered with consent.
• Retain human fallback mechanisms:
  
• AI systems should not eliminate human assistance where the technology fails.
• Users should have a mechanism to obtain staff support when AI cannot complete a task.
• Make accessibility part of digital governance: Existing accessibility requirements under the 2016 Act need effective implementation rather than remaining formal requirements.

Disruptive measures

• Give persons with disabilities a voice in AI governance: Disabled citizens should participate in decisions about:
• Training data
• AI systems
• Infrastructure choices
• Integrate disability concerns into AI infrastructure planning:
  
• Data-centre expansion should account for:
• This is important because infrastructure failures can disproportionately affect people dependent on powered assistive equipment.

Conclusion:

AI should not require disabled citizens to adapt to inaccessible systems; the systems themselves must be designed to include them.