Meeting minutes
Miscellaneous topics.
Janina's APA/RQTF involvement.
Janina: continuing as a contractor with Level Access, enables her to continue her work with W3C...Affiliation has changed, no longer an invited expert. She was auto-removed from her roles in W3C due to that change. W3C needs to make her a chair for APA - so it's a holding pattern. No chair authority at the moment. She is here as a member
representative for Level Access. Work effort is the same, except for managing things. So not taking on chairing statements.
Janina: I'll communicate any news as soon as I have information
jasonjgw: my understanding, is it will support your work on publications?
Janina: Yes. They're very interested in that.
Accessibility of Machine Learning and Generative AI.
jasonjgw: two on here, first is related to any updates and coordination on the doc, And the second is to continue productive conversation from last week. Let's take first one and see what we can coordinate.
Janina: status remains the same. Scott's edits still aren't in there. The blockers are now unblocked (not being on the email) - all has been managed and handled this past week.
jasonjgw: If things get difficult, you can turn to us for moral support. Keep working on the updates and get them in, essentially?
Janina: yes. Scott's first. Another branch, forward-looking, might restructure it. We need to be saying things that affect how W3C works and forward looking and specific. The Canada standards, valuable to bring to our attention. In APA they got very interested in it, and additional people in APA and W3C are noting that work, and our related
efforts. Conversations in other task forces in APA noting parts of things being worked on. Things that might belong in the paper from RQTF
jasonjgw: Note that the Canadian standards doc is the world leading attempt in this at the moment. Do they intend to move into international forum, that would be interesting to find out
Scott: value that being shared
Janina: status question on it - adopted from a standards company, it's not government level?
jasonjgw: a standards org but closely associated with Accessible Canada Act.
Janina: how can we find out what type of authority it has?
… Key phrase is "going to be"? that's the question. Does it become legally binding because the standards group...
jasonjgw: if it's under an anti-discrimination law, if they adopt a standard for purposes of discrimination act, complying with the standard in absence of contrary evidence...this is fairly normal in those kinds of bodies. I haven't looked up what role the standards have in the framework. Adoption has legal effect under relevant law. standards org
is a gov sponsored entity.
jasonjgw: different countries handle this differently. In Australia, someone can't go to court for not following the standard, but they can file anti-discrimination case. It depends on how each country sets up the role of the standards. Can be used defensively rather than a basis of a cause for legal action.
… this is a beginning of a series of docs in this area that they're creating. More detailed docs coming later on. Consider this the broad outline of what is required and subsequent detailed publications coming later on. Good for ISO or relevant body to take up this work and done in an international setting. Having one country worked on it is a
good starting point. Will have to see how it develops in a couple of that.
Machine learning and generative AI: observations and conclusions.
jasonjgw: invite Scott for a presentation on this. Also used ML systems for doc recognition. Microsoft seeing AI, used for recognizing printed material (like mail), and mostly it's not for crtitical information, as I have electronic means for that and electronic bill pay. Don't have to worry about inaccurate character recognition. It can recognize
tables...it will make the cells accessible and navigable with screen reader
Janina: interesting to make lists of where we use AI in everyday lives. We used to need a reader - a paid reader - and that person has been replaced by AI.
jasonjgw: in addition to that, this one is performing relatively well in docs that aren't in prestine condition - they don't sit flat, etc. Printing might not be perfect or envelopes with printing that's not easy to read. generally better than previous tech. Note that my understanding that speech recog has improived with modern neural network
tools, which is why auto captions more of a possibiltiy these days
Janina: wearing my ray ban glasses and opening cabinet and asking for specific item and fnding them...
jasonjgw: using AI tools ... and conclusions we'll get to based on last week. Scott- we left a spot open for you to add to our wisdom on using the tools and observations
Scott: Some of the things are quiet similar, Using AI more to recognize things around the place, getting thigns out of fridge, things that feel the same but have different labels...like two different blocks of chocolate. Use AI to identify. Interesting to see some of the audio descriptions - seeing AI has been updated. Grab frame from video and
describe. Clunky, but interesting. I find useful. Still to be some challenges. Find that from a research perspective, joini things together sometimes it's good or it goes off on a tangent. Vetry slow computer tried to compensate with a graphics card of sorts, so curious what OS could run onit, anything modern that could be installed on it, so in
trying to dig into that with AI queries and primpts I could get it after a while and it could understand what I asked it - but until I phrased in the right way it wanted to give general info. And tell me more about computers in that vintage...but trying to join those elements together...You could tell it was trying and the same query would get
different results, but just a little bit off. Use Ai in my daily life and try to find common elements between things, but if try to fig deep you still run into these limitations.
… after last years AI AI AI...it's going tochange your life (Samsung videos where AI overlords) - huge theme of products, but a return to more practicalities, how can AI support things that are more practical. Not as heavy of a focus on AI. So much AI slop on things like you tube, it's almost like AI is using AI to train AI. Ended up in a loop.
Will see what the result will be moving forward. As someone with low vision i use it to identify thigns. As a resaercher i'm intersted in seeing what it pulls together form a query and what connections it makes, and sometimes it misses in practical terms.
jasonjgw: before we have observations, i want to add one more, and it's personal significance. Modern hearing aids and other similar devices now trying to enhance audio, now have ML trained nerual networks in them. AI running on digital signal processing chip of the device to provide real time audio processes. Smallest hardward platform that can
run it. Look at promotional material for hearing aids they're promoting AI capabilities. take into account.
Scott: google advertising this in Australia
<JPaton> Link to our AI AD research https://
Janina: i'll add one from this week - will be an international medical traveler, there are lots of places to stay and might have to be a lto of stairs. Asked it to find most accessible beach access and generated a list that seems to be fairly accurate, in checking the resources. Saved me a lot of work. First of all - raised an excellent one Jason,
non interactvie use of AI in our devices that we use all the time. This is something that we're unaware of how it's giving us benefits. It's come to web browsers with automated alt text if it doesn't have it - might not be reliable but it's there. The non-interactive benefit of AI and the users are possibly unaware...Doesn't require any interaction
to do it's job. Non-interactive is a key thing in those.
jasonjgw: auto captions and generated captions. turn it on and it does it's business
Janina: noise-cancelling headphones. not sure how much is algorithms and how much AI
… we're all doing it, but it's all important becuase we're using AI to do normal, everyday boring tasks. Found that prompt formation is a key requirement for getting credible, plausible results. have to check on the resulsts. Useful to refine how we ask questions. ask it to do one thing and realize we could have asked it better. refine it or start
a new prompt that's more clearly refined. Some call it prompt engineering. something anyone can do well, so think of it as prompt formation. Everything we talked about had some aspect of that
… when we use it in our work it's like our research assistant, that does things we would probably do ourselves. The difference is it can do it in minutes or seconds instead of hours and days. And maybe find some sites we weren't aware of. As we check it's sanity, we want to check where, how, when...And I've been pleased by chapt gpt and
conclusion, where it's looked, what barrier it ran into so tried something else, gives a chain and links to confirm. I think my example last week was switches and ports on a new keyboard. We confirm it's true when we're not sure. AI is fully capable of making something up. But we check and all use the results. There's a pattern in that we want to
capture.
jasonjgw: add that, bear in mind, creating an appropirate prompt for a general purpose tool, checking its output and making decisions on how reliable it is and what use to make of the information it provided you. Required high level reasoning and analyitcal skills. ANyone who is not so good in those areas, might find themselves taking up false
information from a tool vs someone who is more capable. Cognitive dimension of ML, AI, and a11y - types of cognitieve skills relying on from the user, will differentiate who will make more appropriate use of it.
janina: validation conundrum - added to the list. Is ordinary reasoning sufficient. sometimes you need expertise to validate and clean up..there needs to be some sort of equality in that. Ordinary vs expert reasoning
jasonjgw: more qualified/expert people may be more effective? The better and more expert in the subject matter you'r working with the more advantage you can tkae of the tool. More advantage to the experts than the entry level. In various forms, a good assistant to work with.
janina: when we talk about it that way, what's the different between AI and old fashioned google search, from the user's perspective?
jasongjgw: qual perspetive. Can draw conclusions. A qual different. People who said it's a threat to web search industry were exactly right.
jasonjgw: important observations and some merit in coverage when revising...
Janina: that's the point of the exercise. Yes.
jasonjgw: when you have it at the point when comfortable, we'll all review and contribute
janina: yes, good plan