Meeting minutes
Accessibility of Machine Learning and Generative AI draft.
jasonjgw: prelim discussion with janina. Work still ongoing.
… Question - reading commentary on commentary on web search and integrating ML models with agents as well, led me to realizations that others may have already had. Example of waht I'm raising, and we can discuss implications for a11y. Will cite Stacey's work as an example (with her permission)
… Google and others increasingly use ML and LLM as foundation as of what used to be web search and now an AI query. Users don't necessarily move from AI to underlying websites that are linked as sources. EX: questions on issues on issues on Staceys' or my writing, and the agent finds the info, syntehsizes with sources and the users don't move
beyond that to read the material. Second - if the DO read the material, the purchase and gaining of access could be handled by the agent or the publisher's agent, if that's the case, the website doesn't need to be accessible but the material still does. The publisher at some point, if proceeds in the direction, they don't need to have websites
anymore. People only engaging with content with agents, so all they need is an API. not a UI layer. As far as name recognition and concerns, that could be an advertising contract with the provider and ways of handling that, and ways of making the user know what...they can purchase, and addressed through the API and the underlying UI source goes
away in the end. Transforms the a11y issues - large web with large surface area that needs to be made accessible, under this scenario it goes all the way, users interacting with the agent and content provided by the agent that needs to be accessible. What we know of disappears.
Janina: I think you summarized conversations we've had all along. I think yes, that is very much a possibility of where we're going. Haven't thought maybe web pages might go away but all we need is API access for what's going on underneath. This debate is already going on. Questions being generally raised, at least W3C. ATAG is wondering about this
- is the browser going away? User agent and web content. It's out of sight. Agents on both sides of that. Very likely, these are contrsucted on the fly for particular users. Pluses and minues. Plus - really personalize the content presentation.
jpaton: something we've seen in TV manufacturers, if you had a universal API that could work with all TVs, and wouldn't need to worry about multiple apps or multiple controls. Because a toxi convo with TV manufacturers...they wanted you to see their brand and their TV interface. Similarity here? Publishers have APIs, but want to keep visibility of
the brand and push the branding colors and branding logos. Potentially a business and internal politics/business politics.
Janina: counterveiling force. Branding is a high priority value. Whatever a11y we get or don't in those branding constraints. Some content could be done differently and arguably more accessible, vendor looking for branding as top value. I've always maintained, that user is in charge of the look and feel, not the content author. You can't avoid
that. I think we've always said whatever is published at a web address, OK to be branded, that's what you copywrite, but user in charge of how it's presented to them. In Jasons' model, that's abstracted away - what we're left with - what's copywrited?
jasonjgw: model I'm imagining - good or bad scenario...branding becones a relationship between content provider and agent provider. Agent can forward that through to the user as appropriate. no longer an interaction with the underlying interface. negotiation and provider of the agents and user interacting with and resources consuming. Think that
the agent provider will win that battle, because that's what user wants to interact with for efficiency reasons or other...agent providers will win that struggle.
Janina: user should stay in charge of what agent is used...
jasonjgw: agree that could be a problem. There's no underlying UI from content provider.
Janina: what do you tell the library of congress what your'e copywriting.
Jasonjgw: underlying human resources...
… we can imagine in the future, let's say for your (stacey's) next book. The publisher could sell the rights for LLM to sell that content and be lucky to get some financial remuneration from that.
jasonjgw: i think where we're heading - a11y issues could be very different, accessing not via thousands of webpages, and content gets abstracted and gets presented from sources and presented and changes that a11y conversation significantly. I don't know how far that's going. A path in that direction could go further or less, and vary depending on
the application or the person. But that pathway is opening up. a consequence might be underlying content, info filtered through...remediation might become the preferred accessibility strategy. If they fail, there might be human intervention. At a cost? A11y in taht direction in agent center.
stacey: interesting Q? People may ask how to we know the alt text is correct? That moves across to other AI-based mechanisms: how do we know the content we are getting is correct?
janina: But then just because alt text is human generated that doesn't mean it's correct.
stacey: Summary of conversation from Stacey that happened in between the above comments (since hard to scribe and talk) -- is the content itself (being filtered through the AI) become part of the accessibility itself? As AI transposes/synthesizes the content, how do we know it's correct? And to catch up on the other content, she added content
around monetary difficulty with businesses and customers. As the UI may go away and replaced with agents models, can people afford it? Are they willing to pay for it? In addition, the increase of the large data centers and usage of water and other natural resources - where are they going? Will building those get approved? And also made some
comments on brand and IP with USPTO and what you can get IP on and copyright infringement.
Game accessibility.
Janina: John - not sure if your'e ready with anything yet...
jpaton: started gap analysis. A lot to do and a lot of headings that I'm going to need text on them. It's coming along.
jasonjgw: meeting planning - with no pressure intended, willing to discuss where it stands next week?
jpaton: yeah, I only have three working days between now and then. (bank holiday) and start of half term. Maybe the week after, I shoudl have more
jasonjgw: should we move the whole meeting to the following week? Give JAnina more time as well
Janina: if you haven't seen content from me by Monday, which is a US holiday, then let's put it off to June
Janina: Try to keep my cochair about the major work we're doing, a joint effort with APA. He still needs to be convinced. Which is why we're doing a gap analysis. Commit to sending a list of links, so feedback from Matthew on what's out there. Maybe that's everything we need. But obscure enough...if you (John), who is working in this industry,
can't find it, then how is the everyday person going to find it?
… and discussion, he was excited. Loves gaming. As a coder and enthusiastic player.
jasonjgw: make sure that gets FWD to John..
Janina: hopefully he posts it to the list
Janina: I didn't press for a time commitmment on that
jasonjgw: gap analysis takes time anyway.
Miscellaneous topics.
jasonjgw: standby on early next week or move to the following week
Stacey: Stacey will not be attending meetings on June 10 and 17 for vacation. Perhaps might not be attending July 1 (TBD)
jasonjgw: my travel plans are uncertain at the moment. TBD
Janina: if anyone any ideas in ITF -internet engineering task force (they own the left side of the :// and all of the other ones) - let me know