Ungenerating Barriers: Fixing and Avoiding Accessibility Issues Caused by Generative AI

It seems like Generative AI is everywhere in the digital space nowadays. Large language models (LLMs) are used to create reams of content, write code, or change swathes of web pages and software. Many people use Generative AI to write, edit, or remediate documents. This content is often inaccessible. Sometimes, the content has basic coding or formatting errors. In other cases, the content meets technical requirements - but is unusable for people with disabilities."

You may feel overwhelmed or anxious about fixing this machine-produced content. Yet there are tools and techniques you can use to address this problem - using easy-to-learn, free, and existing methods. And despite the hype and mystery, we can teach you how and why Generative AI produces inaccessible content so often, and how to use this knowledge for fixing these issues. In this session, you will learn:

-- How the way Generative AI is built reproduces common access barriers
-- How the methods Generative AI uses to produce content creates new barriers or junk contentCommon barriers in generated content, including:
-- Junk code and the overuse of ARIA
-- Missing and incorrect header structures
-- Inconsistent terms and labeling
-- Unclear or unplain language

Methods to address these barriers, including:
-- Evaluating whether or not a Generative AI tool is usable in a given context
-- "Back to basics" methods to correct junk code
-- Quick checks for fixing generated content
-- Common fixes for common barriers in AI-generated language

We will draw from our own experience as accessibility professionals working in the public and private sector with a range of clients, including in correcting AI-generated code, documents, and content.

The session will include an open questions and discussion period at the end.