The conversation between Mark Shapiro and Gerard Cohen explores the growing interest in using AI for digital accessibility, while grounding expectations in practical reality. They begin by acknowledging that many organizations, including large enterprises, are considering replacing accessibility teams with AI. While AI can assist with identifying basic issues such as missing labels, alt text, and structural problems, Gerard emphasizes that these capabilities are not new. Similar automated tools have existed for decades, and AI largely builds on those foundations rather than introducing a transformative leap forward.
A key theme throughout the discussion is that AI can support efficiency but does not eliminate the need for human expertise. AI can help scan websites, flag issues, generate code suggestions, and even improve areas like captioning, which has seen meaningful progress. However, outputs still require human validation to ensure accuracy and usability. The speakers stress that relying solely on AI without understanding accessibility principles can lead to flawed implementations, especially when context and intent are involved.
The limitations of AI become more apparent when dealing with complex, real-world scenarios. Accessibility is not just about checking boxes or adding technical attributes. It requires understanding user intent, business context, and the nuances between usability and accessibility. AI struggles with these areas, particularly when interpreting meaning, handling dynamic interfaces, or evaluating user experience. The discussion highlights that accessibility is deeply contextual, and AI lacks the human perspective needed to fully assess whether something truly works for users.
Ultimately, the conversation reinforces that AI should be treated as a tool, not a replacement for human accountability. Removing human oversight introduces risks, especially in legal and compliance contexts where responsibility cannot be delegated to a machine. The speakers caution that while AI holds promise and can enhance workflows, organizations must use it thoughtfully and maintain human involvement to ensure accessibility outcomes are both accurate and meaningful.