Meeting minutes
further work.
Coordination: CAPTCHA and authentication.
<Roy> https://
jasonjgw: Ongoing discussion on how web authentication and security keys may overcome need for CAPTCHA
jasonjgw: Any recent developments or intended actions?
janina: Email was sent to the FIDO Alliance, but no response so far
janina: Suspect that many of the old CAPTCHA techniques will be around for a while, so those points are still in scope
jasonjgw: Mentioned a new resource that we may want to reference
jasonjgw: Once we get enough new information we can be in a position to update the doc
janina: Suggest that the new US document that Jason mentioned should examined for material to update the doc
Ethical Principles for Web Machine Learning - review of draft.
scott_h: Will take that as an action item
jasonjgw: The draft is ongoing. Any thoughts?
https://
janina: There is also a WebML working group active in this space
jasonjgw: There is a transparency section that would be good to look at
janina: Since the document is still open, we should go ahead and submit github issues
janina: Knowing the basis of previous failures of AI could be useful in understanding how to improve this technology
jasonjgw: Suggest we look at this again when Lionel is available
Artificial intelligence and people with disabilities - opportunities for
janina: Also suggest we reach out to Yutta
jasonjgw: AI may often have propensity for errors that the developers had not anticipated
janina: Rooting out how AI errors occurred will be useful in correcting this in future
Raja_Kushalnagar: About the web overlay AI approach -- when errors do happen, there is no human involvement to fix AI errors
janina: Such overlay AI inaccuracies has been a known issue, which is one reason why the NFB and others have opposed the use of such tools
janina: This is another good topic to bring up when Lionel is with us
Raja_Kushalnagar: We know AI is commonly used for audio captions, for example, but the problem arises in high-stakes settings
janina: One point is that when captions are retained for future publishing, then correcting those errors should be important
Raja_Kushalnagar: It is important to know how good the accuracy is; while the live automated captions may be quick, it is hard to know when spoken information is going to become high stakes
janina: This has really expanded since COVID has pushed so many meetings to virtual
jasonjgw: The legal point is crucially important in certain settings; knowing at what point the automated captions fail to meet required access
jasonjgw: This is an area of discussion we should continue discussion and collaboration with other W3C groups
Raja_Kushalnagar: Agreed
Raja_Kushalnagar: I am happy to help with further analysis and writing
jasonjgw: There is legal research to be done as well as the technological issues
jasonjgw: There may be a point at which the AI becomes as generally reliable as human-moderated access, but this may vary by the specific vocabulary used and setting in which it is used
Raja_Kushalnagar: About Open-AI - it doesn't allow people to do nefarious things; but human moderators are not supposed to censor communication
jasonjgw: The problem is that the system may be trained to censor certain content, which can be a problem
Collaboration Tools Accessibility User Requirements.
jasonjgw: Still gathering COGA feedback
janina: They will be adding github issues
jasonjgw: Once we have the comments and address them, at that point we should be able to push out an edited draft
scott_h: Have some additional points and citations I plan to include
scott_h: Was unsure whether to wait for COGA's comments first.