Meeting minutes
Accessibility of Machine Learning and Generative AI draft.
jasonjgw: Summarizes we're awaiting Janina's scturctural and content edits in a branch
jasonjgw: Meanwhile, we discussed issues for the WebAI IG conversation
jasonjgw: Add to that the ATAG attention in AI
jasonjgw: And lastly a WebMCP question posted by Matthew
jasonjgw: The issues are germane with MCP regardless that specific proposal
jasonjgw: Do have questions and issues to raise as we go ...
jasonjgw: Anyone with questions or issues so far?
[crickets]
AI Redraft
janina: Still no text to review
Coord re WebMCP, Web-AI, Etc
<matatk> Pointer to email to APA list et al about WebMCP: https://
janina: Generally in favor of the principle; against using accessibility infra in practice (more later); desire seems to be to build a deterministic structure to facilitate AI agents and give them opportunities that are more structured, apart from an HTTP request, where you don't know if it's a human or an agent. Maybe opportunity to do different
rate allowances here.
janina: Could facilitate deterministically what's open to agents, what's available, sounds like the semantic web. But, at least initially, against using accessibility tree for this, for reasons matatk mentioned about concerns it will be misused. Glad world has caught up to this, but copy it - don't co-opt it.
matatk: Seems to me we're concerned that opening the a11y tree people will stuff inappropriate things in ARIA to begin with, and it would go on from there
matatk: To some degree we're in the same situation even if we have a parallel structure, though.
janina: Maybe the ARIA structs become trimmed down as duplicative of what's in the MCP tree and works well
janina: We should consider that too
jasonjgw: Have a series of comment and questions after reading the Explainer this morning
jasonjgw: The info that's provided is different from a11y tree, imo
jasonjgw: The a`11y represents components available in the ui
jasonjgw: this agent interface rather provides functionality that doesn't conrrespond to anything in the UI
jasonjgw: e.g. a W3C TR type page, phps hierarchically organized -- no search
jasonjgw: the MCP could provide a search even though not on the TR page
jasonjgw: So not the same
jasonjgw: the MCP implementers are deciding what's exposed to the user API, and that's different from what the a11y UI structural mirror provides
jasonjgw: Wondering, did I read the Explainer correctly?
matatk: Sounds reasonable
jasonjgw: Next, Explainer is clear this API is to be used only where there's a live browser session -- it's intended as a background process
jasonjgw: That raises a11y questions
jasonjgw: Could we have functionality exposed to the agent but the underlying UI fails a11y somewhere; User then unable to use the agent UI?
jasonjgw: A semantically rich interface to a subset but with no good a11y
jasonjgw: Think this is an a11y issue we've not previously seen
jasonjgw: Am I overly concerned?
matatk: A good insight and it's likely because people will be excited and not worry enough about all the tag alongs
matatk: It's taking us to a burden shifting situation
matatk: Shifting burden of checking the agent to someone who can't do that
matatk: An astute point we should call out
jasonjgw: Thirdly
jasonjgw: More of a question
jasonjgw: Seems having ability to interact this way is powerful especially for COGA
jasonjgw: but then we get back to the second point
jasonjgw: To monitor the agent and correct for its mistakes
jasonjgw: It shifts the cognitivwe process in is own right
jasonjgw: What will that effect be? Is it offset by the opportunity?
matatk: Vigilence of learning these skills is important to be cognizent of
janina: We're defining a new kind of AT here
matatk: A model tuned for a persons' ability
janina: I'd like to put AI on this job of developing those personalized AT
matatk: Security/privacy concerns - this would have to be done locally - but if it is observing use locally as janina mentioned, perhaps this is one way to solve the problem of raising awareness with users as to which settings they could change to make their experience much better.
janina: YHes, but could be local
matatk: My excitment about the potential here is rather the opposite of Jason's concern
matatk: It's an opportunity to do so much more as a pwd
matatk: Noting there are three aspects to review
matatk: Jason looked at the Explainer
matatk: There's the proposed Accessibility Considerations section
matatk: There's an specific issue to respond on -- pointed in the email
<Zakim> matatk, you wanted to mention Adapt's work on form field purpose and maybe splitting the info between accTree and agentTree
Media accessibility, coordination of work.
jasonjgw: Coordinating on advancing Immersive Captions to APA WG Note
janina: We were trying to get the CG into AP
janina: into APA in order to keep the same authors
janina: We got stuck with some, not all
List of proposed meetings: https://
matatk: There are several standing issues -- MAUR update funding; common taxonomy across web and broadcast;
matatk: Can talk about again
Game accessibility.
jasonjgw: Notes MA and John have begun communicating on this
matatk: Where we stand is John is developing a gap analysis on gaming in general
jasonjgw: particular focus has been placed on understanding what skills and abilities are truly essential to a specific gaming task, and what incidental
johnI'm calling it itentional and unintentional friction
jasonjgw: Is a useful understanding for other contexts too
matatk: Recalls Ian Hamilton noting games are meant to be hard
matatk: There are trade-offs in how you make these things accessible
johnTotaly agree with the standards
johnbut the necessary friction thing is missing
joNo metric on how you note it's part of the intentional challenge that needs smoothing out
joComing to identifying mechanisms to challenges that need to stay and point to existing standards on how you do the details
jostrategy to identify core friction
matatk: doesn't it vary among different people?
<JPaton> https://