Blind Innovators Lead the Way: AI Accessibility Research Learning from Users, Not for Them

Jaylin Herskovitz knows something that most AI researchers don't: blind people are already hacking their technology.

Not maliciously, but creatively — combining screen readers with voice assistants, switching between different apps to accomplish single tasks, developing workarounds that would impress any programmer. As a researcher at Tufts University who happens to be blind, Herskovitz recognised these behaviours as expertise, not problems to be solved.

That insight led to ProgramAlly and AllyExtensions, revolutionary tools that flip accessibility research on its head. Instead of building AI systems for blind users, Herskovitz created systems that blind users can program themselves.

ProgramAlly lets users train AI to recognise specific visual elements they encounter regularly — not just "person" or "car," but "my medication bottle" or "the correct bus." AllyExtensions creates custom shortcuts between different accessibility apps, automating the complex workflows blind users already navigate daily.

The approach represents a fundamental shift from seeing disabled people as passive recipients of technology to recognising them as "domain experts in their own lives," as Herskovitz puts it. Early testing shows users can create personalised solutions that work better for their specific needs than one-size-fits-all alternatives.

Key Facts

• Tools allow blind users to train AI for specific visual identification tasks
• Based on study of "hacking, switching and combining" techniques BVI users employ
• ProgramAlly enables custom object recognition beyond standard categories
• AllyExtensions creates automation between existing accessibility software
• Research recognises disabled users as "domain experts in their own lives"
• Source: Tufts University research, presented February 2026

Why This Matters

Traditional accessibility research follows a top-down model: researchers identify problems, design solutions, then test them with disabled users. This approach often produces tools that work in laboratories but fail in real-world conditions because they don't account for the sophisticated strategies disabled people already use.

Herskovitz's work emerges from disability studies principles that view disabled people as innovation leaders rather than problems to be solved. The "curb-cut effect" — where accessibility improvements benefit everyone — suggests this approach could improve AI systems broadly.

What We Don't Know Yet

The tools require technical comfort that not all blind users possess. Training custom AI models demands time and patience that may limit adoption. Integration with existing assistive technology ecosystems remains challenging.

Early research involves small user groups, and scalability questions remain unanswered. The approach may work better for some types of tasks than others, and long-term effectiveness studies are still needed.

Sources: Tufts University · Accessibility Research Community
Published February 28, 2026 · Category: Science & Technology