AI and Machine learning - mostly focused on repair use cases and needs to have clearer goals

[joshueoconnor]

Over all some of the use cases are really about repair.

What does the innovation roadmap look like? Where do we need it to be? It is better here to identify what AI does poorly, and what it does well? I think this document needs to be clearer about that and with an eye on a what the goal is. I’m not sure this doc outlines what AI/ML does poorly and therefore how we can make what it does better, rather than seeing AI/Machine learning as a panacea.

We are better off getting a clear view and presenting that - as well as by assessing the pros/cons maybe suggesting areas for further exploration, rather than a shopping list for current automated accessibility tools (which is what it currently looks like to me).

[sehollier]

The way this has been structured is based on user feedback. In terms of the innovation roadmap, the ultimate request from users is that accessibility issues are resolved by AI seamlessly without user intervention. Some AI innovations like autogenerated alt text for example, if accurate, would address this so user agent accessibility done on-the-fly would be a possible innovation roadmap, but widely applicable across other technologies as AI develops.

Most of the document focuses on AI potential and accessibility issues, but acknowledges it is still fairly general in most use cases. However, some recent developments are becoming improved such as recent integration of Google Gemini into Talkback on Android. It is much better than Microsoft AI examples mentioned here for alt text. Also, there are notable improvements with live captioning. Hence, I think we can look at things AI could do well but are still in their infancy. Sign language for example, isn't really effective yet despite apps being available and is also very localised by mostly ASL. This could be an example of something that is really good that does not exist without AI, but if AI improves, then sign language online could become effective over time.

It is true that there is a lot of focus on automation, but I think there is a differentiation between how users can benefit from AI seamlessly and how people trying to address accessibility issues can also be supported by AI.

[sehollier]

Another example of a user benefit not necessarily about repair might be connected to some of our XR guidance, where tactile representation in XR space could be generated to support people who are blind. AI could scan images to determine the texture of objects, represent them in a 3D environment, and then provide haptic feedback to the user.