AI and Accessibility: The Good and the Bad - Fireside Chat

Is AI a powerful accessibility assistant or a flawed tool that perpetuates stereotypes? Experts weigh in.

#1about 3 minutes

Assessing the current state of web accessibility with AI

The web's complexity has grown, but accessibility has largely kept pace, with conversational AI showing potential to simplify user tasks.

#2about 5 minutes

Evaluating the quality and context of AI-generated alt text

While early AI-generated alt text was poor, newer models are improving but still struggle with providing context-specific descriptions for images.

#3about 4 minutes

Understanding the risks and biases inherent in AI models

AI can introduce hazards, such as hallucinations in food descriptions, and perpetuate harmful biases by misrepresenting people with disabilities.

#4about 3 minutes

Addressing the accessibility gap in modern design tools

Design tools like Figma lack native accessibility features, causing critical semantic information to be lost during the handover to engineering.

#5about 4 minutes

Debunking the false promise of accessibility overlays

Accessibility overlays that claim to automatically fix websites fail to provide real accessibility and do not offer reliable legal protection from lawsuits.

#6about 7 minutes

Prioritizing native features over custom interface widgets

Building custom widgets for features like dark mode or text resizing is less effective than supporting native browser and OS capabilities.

#7about 7 minutes

Making the business case for accessibility beyond compliance

While proving direct ROI is challenging, accessibility can have a positive revenue impact, and legal regulations act as a necessary enforcement mechanism.

#8about 6 minutes

How assistive apps combine AI with human support

Applications like Seeing AI and Be My Eyes demonstrate a powerful model where AI provides initial assistance with an option to escalate to human volunteers.

#9about 10 minutes

A future outlook on AI's evolving role in accessibility

The future of AI in accessibility depends on addressing its current flaws, including inaccessible interfaces and biased training data, while leveraging its potential for narrow, specialized tasks.