Inclusive AI for Persons with Disabilities

Contents

Executive Summary

This multi-speaker panel discussion at an AI summit brings together government, academia, civil society, and industry to address the critical gap between AI's potential and its actual accessibility for persons with disabilities. The speakers emphasize that inclusive AI design must be non-negotiable, economically viable, and centered on the lived experience of disabled users—particularly in underserved regions like rural India where 70% of India's disabled population resides. The overarching message: AI can amplify exclusion or enable inclusion at scale, and the difference lies in deliberate, co-designed solutions.

Key Takeaways

Nothing About Us Without Us: Disabled persons must be co-designers from day one—not afterthought accessibility auditors. This is non-negotiable for both ethical and practical (quality) reasons.
AI Amplifies Both Inclusion and Exclusion: Without deliberate inclusive design, AI scales discrimination. With intentional co-design and representative training data, AI can unlock independence at scale for billions.
Affordability & Rural Reach Determine Real-World Impact: A₹5,000 voice-based app working on basic smartphones in rural India creates more actual inclusion than a £1,000 UK device. Design and deployment strategy matter as much as the technology itself.
Enforcement Mechanisms > Laws: India has strong disability rights laws. The gap is accountability: accessibility audits, non-negotiable procurement standards, grievance redressal (Maharashtra chatbot resolved 5,000+ grievances in 7 months), and penalties for non-compliance.
Intersectionality & Systemic Collaboration Beat Siloed Efforts: Single-disability organizations working alone miss overlapping needs. The Accessibility Coalition model (TAC) bringing together multiple disability sectors with private industry creates systems-level change.

Key Topics Covered

Disability Data & Invisibility: Massive discrepancies between global (200 million) and census-reported (26.8 million) figures for Indians with disabilities; gap reflects systemic invisibility and undercount
Algorithmic Exclusion Risks: Examples of AI failures (voice recognition unable to parse dysarthric speech; hiring algorithms flagging neurodivergent candidates)
Assistive Technology Access Gap: Only ~3% of people in low-income settings have access to assistive products; ~80–90% in high-income countries; affordability remains critical barrier
Accessibility Standards & Enforcement: Move toward mandatory, non-negotiable ICT accessibility guidelines; Indian standards exist but need enforcement mechanisms
Privacy vs. Accessibility Tension: Exploring the complex trade-offs between collecting data necessary for inclusive AI and protecting individual privacy
AI Bias & Training Data: Underrepresentation of disabled persons in training datasets leads to discriminatory design (e.g., facial verification failures for blind users)
Co-Design & Nothing About Us Without Us: Disabled persons must be involved in all stages of technology development—not as afterthought consultants
Employment & Economic Inclusion: AI as tool to reduce artificial labor market barriers for disabled workers; multimodal authentication and assistive tech enable workplace participation
Digital Public Infrastructure: India's existing digital infrastructure (Aadhaar, UPI, etc.) can be leveraged for inclusive accessibility solutions
Governance & Implementation: Real-world case studies of AI-driven platforms (Maharashtra's Danga Sahay portal) demonstrating service delivery, grievance redressal, and scheme eligibility assessment
Multilingual & Multimodal Solutions: Combining multiple authentication and interaction modes (voice, text, visual) to serve diverse disability populations across linguistic regions

Key Points & Insights

Scale of the Problem is Hidden: India's official disability count (26.8 million) is dramatically understated compared to WHO estimates (200 million). This invisibility in data directly translates to policy blindness and underinvestment.
Affordability is Non-Negotiable: UK consumers with disabilities spend £1,000+/month on assistive tech; Indian solutions must work at ₹3,000–5,000 price points on basic smartphones. Scaling accessibility requires economic viability, not just technical functionality.
Algorithmic Exclusion is Real & Measurable: AI systems trained on mainstream data actively exclude disabled users (voice recognition, facial verification, hiring algorithms). Failure to include disabled persons in training data and design creates automated discrimination.
Enforcement Gap is Larger Than Legal Gap: India has accessibility laws (2016 Rights of Persons with Disabilities Act) and standards; the problem is lack of enforcement, audits, and consequences for non-compliance. Procurement and contractual accountability are levers.
Co-Design is Not Optional—It's Essential: A blind engineer testing a blindfold-based scenario cannot replicate lived experience. Technology that "works" without disabled users' involvement will inevitably fail them. This requires structural participation, not token consultation.
Digital Public Infrastructure Can Be Inclusive: Maharashtra's Danga Sahay portal demonstrates AI-enabled service discovery: disabled users submit once, system auto-profiles eligibility, fetches data via APIs (Aadhaar, ration cards), and recommends schemes. This reduces bureaucratic friction.
Privacy & Accessibility are Not Inherent Enemies: The tension is real but resolvable. Example: voice-based government services for visually impaired users require voice data but enable independent service access. Society has adapted to privacy trade-offs before (smartphone cameras); thoughtful governance can balance both.
Employment Potential is Enormous but Artificially Constrained: Millions globally unable to work not due to disability but due to accessibility barriers. AI removal of access barriers (real-time transcription, image description, multimodal auth) can unlock massive labor force participation—including for parents (like Augustia's mother) excluded from credentials.
Multimodal Authentication Solves Diversity Better Than Single-Mode: Multiple authentication methods (voice + fingerprint + visual) serve broader populations than one-size-fits-all approaches. This increases both accessibility and security.
Rural Focus is Critical: 70% of India's disabled population lives in rural areas where AI solutions must function on low-bandwidth, affordable smartphones. Infrastructure and design assumptions built for urban/high-income contexts fail in these environments.

Notable Quotes or Statements

Manik Govinda (Department of Empowerment of Persons with Disabilities, India): "The question is not whether AI can help... The question is whether we can design, deploy and scale AI solutions in the way that actually reach the people who need it most. Not in London, not in San Francisco, not in New Delhi, but also in Latur, in Siliguri, in Sambalpur."
Manik Govinda: "If a voice recognition model cannot understand disrupted speech (dysarthric speech) or if a hiring algorithm flags a neurodivergent candidate as unfocused because eye movements are too fast, we are not just failing to be accessible—we are automating discretion."
Augustia (Meta, Concept Engineering): "Nothing about us without us is a phrase I've heard a lot... Building with people who are actually going to use the stuff is the best way to make it good."
Augustia: "When we develop technologies that meet the needs of what a mainstream designer might consider corner cases or the extremes, we actually make it better for everybody" (Curb-cut effect example).
Dr. Andrew Fleming (British SHI Commissioner, lived experience with dyslexia/dyspraxia): "Responsible failure is part of inclusive innovation, especially when vulnerable groups are affected."
Amar Latkar (Mission Accessibility, lawyer with visual disability): "AI has the potential to either include people at scale or exclude people at scale... The problem is not absence of law. The problem is absence of enforcement mechanism."
Manik Govinda: "AI should be AI for all—not just solving complex problems but building resilient AI ecosystems that are scalable, economical, and spread to the wide corners of the country as a whole."

Speakers & Organizations Mentioned

Government & Policy

Manik Govinda, Department of Empowerment of Persons with Disabilities (DPWD), Government of India
Tupar Mundday, Disability Services, Maharashtra State Government
Dr. Andrew Fleming, British SHI Commissioner to East and Northeast India

Industry & Research

Augustia, Director of Concept Engineering, Meta
Professor Allison Noble, Vice President (Science Policy), The Royal Society (UK)
Vince Cerf (quoted), engineer involved in Royal Society accessibility report

Civil Society & Legal

Amar Latkar, Co-founder, Mission Accessibility; lawyer with visual disability
Nipun Malhotra, Moderator, accessibility advocate

Institutions

The Royal Society (UK) — National Academy of Science
NIPID (National Institute for the Empowerment of Persons with Intellectual Disabilities), Sikandra
Tinker Labs — startup (mentioned: Annie product)
Tester Labs — startup (mentioned: KBO product)
Assisted Tech Foundation, Bangalore
Quantum Hub (event co-convener)
RTC (sign language interpreters)

Government Initiatives Mentioned

India AI Mission
Rajiv Ratory case (driving accessibility guidelines)
Danga Sahay Portal (Maharashtra unified disability services portal)

Technical Concepts & Resources

AI & Machine Learning Applications

Speech-to-text and voice recognition for dysarthric speech accommodation
Adaptive learning platforms (personalized, real-time feedback for neurodivergent learners)
Facial recognition and biometric authentication systems
Real-time captioning and transcription (multilingual)
Image recognition/description for document access and environmental navigation
LLMs (Large Language Models) for accessibility enhancement
Multimodal AI systems (combining voice, visual, text modalities)
Chatbots for grievance redressal and information access

Accessibility Standards & Frameworks

Indian Standard for Web Accessibility (ICT product guidelines)
UK Special Educational Needs & Disabilities (SEND) Framework
WCAG (Web Content Accessibility Guidelines) — implied
Persons with Disabilities Act, 2016 (India)
EKYC (Electronic Know Your Customer) biometric verification systems

Digital Infrastructure & Tools

Aadhaar (biometric identity system)
UPI (Unified Payments Interface)
API integrations for cross-government data access
Maharashtra's Danga Sahay Portal — end-to-end service discovery & application
Chatbots for service information and grievance management
Smartphones (₹3,000–5,000) as primary assistive technology platform

Methodologies & Approaches

Co-design with disabled persons throughout development cycle
Inclusive training data collection and curation
Accessibility audits and conformance reporting (mandatory)
Multimodal authentication methods
Curb-cut effect principle (accessible design benefits all users)
Disability impact assessment for procurement

Research/Reports Referenced

Royal Society Report on Inclusive AI & Assistive Technologies (2024) — case studies: UK, US, India, Kenya
WHO Global Prevalence Estimate (16% disability; projected 3.5B needing assistive tech by 2050)
India Census 2011 (26.8M official count, widely acknowledged as undercount)
Indian Test of Intelligence (NIPID, culturally adapted IQ assessment with AI enhancement)

Accessibility Note: This summary preserves the transcript's accessibility-forward approach, including repeated image descriptions and explicit mention of sign language interpretation services.