I always highly welcome accessibility topics. The conclusion sums up the problem quite well: "The ideas and solutions we’ve discussed are never going to replac

I always highly welcome accessibility topics. The conclusion sums up the problem quite well: "The ideas and solutions we’ve discussed are never going to replace the need for real humans to write standards for accessibility, to do manual testing, and to educate developers and content editors on how to create accessible and semantic code" ML could improve the (quite poor) automated a11y testing rate, but never get to the 100% imo. The biggest problem to solve accessibility with ML is the subjectivity of some topics. For example, it's quite subjective whether some content is understandable to the user or not.