Meeting Minutes
Wendy Reid: Alright.
... Okay, thanks everyone for joining today. Just for the record, I had this open in front of me. So we're gonna do the same as discussed last meeting, 4 minutes, so we'll take a transcript from this meeting. I will take that transcript and upload it into into, ether pad for people to edit, so I will send a link out or I'll share it right now, actually, because people are welcome to edit
... I've also included in this EtherPad after last meeting. There's a section for the minutes, there is a section for chat, because we saved the chat last time, because there was actually a lot of really good discussion in the chat, and we'll do that again. And, notes. So if anyone has notes that they've taken during the meeting and wants to share them, as part of the minutes, then there's a notes section.
... If you've never used Etherpad, all you have to do, it's probably gonna prompt you to sign in, you just need to use your W3C account.
... Same username and password as your W3C account, and it should just work.
... But you don't have to pay attention to that during the meeting. I will make sure that everything gets copied in there, and then you're welcome to edit it later for spelling mistakes. The thing I've noticed most from the transcript from last time is that it gets names wrong, and it gets, acronyms hilariously wrong. So It's something to keep in mind.
... For this week's agenda
... Actually, I'll start with the first reminder, which is, if you are new to the group.
... We have an introductions thread going where people can introduce themselves. And it's been nice to get to see everyone and learn why they're here.
... I wanted to start off also by sharing some updates that I've made to our GitHub. So, I'm going to share my screen, because it's easier. I'll explain them, but I'll also show them, because it's nicer to see. So, we now have a homepage. I created just a little, little GitHub pages, home page this will be where it has a lot of the same information as the repository README, but, so there's links to our issues, our discussion, our wiki.
... You can find I will update it with meeting minutes when the meeting minutes are ready, so you can be able to read them online.
... There will be, when I have a draft ready of the, document, it'll be uploaded here.
... Same with the charter and all of that. I was really hoping today that I would actually be able to show you, the the draft version of the note, or of the document, but, I have spent several hours trying to get it into, like, new W3C style. And I have still over 100 broken links to fix, so I am working through that right now. I'm getting close. I cracked the code on what was going on with it.
... Basically, the ATAG 2.0, was generated on an old system that W3C doesn't use anymore for its documents. We now use something called well, two options. You can either use something called Bike Shed or something called Respec. I'm a Respec fan. So, I'm trying to take it over to Respec. Respec has a lot of wonderful features that make things a lot easier. But it works a lot better when you're working from a new document, not an old document. So that is the problem I'm running into now, is cleaning it up.
... I will hopefully I'll send out a link once I have it, once I'm done cleaning.
... The old way that definitions were referenced is really hard to clean up.
... So let's move to, what the meat the meat of the meeting today. So, I thought for today's meeting that we could talk about some of the open discussion topics that we have going right now.
... For people who weren't here last time, what we've decided to do is we're gonna have discussions in the meeting, but, so that people can participate asynchronously, we will have we are going to use the discussions feature on GitHub, so people can contribute, comment to each other, and we can, you know, take those discussions and talk about them in the meeting.
... So the first one that I want to chat about, this one got some traction, was around coming up with type or having a discussion about types of authoring tools. So this, you know, there ar.
... I think some of this work was done when ATAG 2.0 was created, but obviously in the intervening 10 years, there are probably new tools, or new types of tools out there. What should we be taking into account? And do we maybe need to recategorize them, or think of new categories for authoring tools.
... And we'll be using the raise hands feature in Zoom, so I can keep track of who wants to talk and when.
Charles Hall: I just like that, you used the word categories. I think we should use that going forward, in the in that discussion, channel in GitHub.
... I added the list from the 2.0 implementation report, which called them, market segments, and and I don't think that's, I think that's too narrow, or possibly, too vague, so categories, works better than market segments.
Wendy Reid: Cool, that's good to know. Yeah, I I kind of agree, market segments is a bit of an awkward delineation.
Hidde de Vries: Hi, I was thinking, there are some tools that are specifically making web content, and then there are tools that can make any type of content, like you can make a Word document and never put it on the web, or, like, edit a movie or something like that.
... Do we wanna scope it to web specifically, or is that not very important to do.
Wendy Reid: I think, because we want to align with WCAG 3, that we need to look beyond the web, but are you, like, Beyond the web, but no.
... I guess the scope is, like, digital.
... So we're not, we're not gonna, you know, talk about authoring tool accessibility for typewriters, but I think, like, we can look slightly beyond, like, I guess, like, maybe a good example is, like, mobile applications, right? Like, Twitter is a mobile, or whatever we want to call it now. Twitter is a mobile application, that you can create content on, or Mastodon, o.
... So we should include those types of interfaces in our in our categories as well.
Hidde de Vries: Or maybe one more thing to add.
... to, kind of, lock the definition into, sort of.
... I was thinking maybe we need to say something like anything that creates web content, or, like, that results in HTML, or something like that. Maybe that is a way t.
... Decide, kind of, what Belongs, or art doesn't belong i.
... In this, because that seems to be kind of scrolling through the issue, what most of the tools that we're mentioning ther.
... actually do. So something that you ca.
... After you've used an authoring tool, it ends up, or it can end up on the web.
... Is that, like, a way to scope it, perhaps.
Mary Ann Jawili:
... But, like, if I can create a video, then I would wan.
... to be able to, I don't know, add captions to that video.
... And also, I like the Twitter example, just because, like, you're creating content that will be posted, but you should be able to make that accessible for anyone who's looking at it. And then also, if you're creating.
... I guess what I'm also thinking about is, like, there's lots of platforms where people can build on top of that platform, such as Salesforce and ServiceNow. I work at ServiceNow now, so, and I'm thinking about how we want t.
... We should want to encourage people to create.
... To build stuff on top of the platform, but also make sure that stuff is accessible.
Charles Hall: I think reacting to and building on those last two comments.
... I have, in the discussion thread, a short list to get started, and every one of those thing.
... can be published to the web, despite the fact that they also exist in other channels, right? So social media was brought up, social media obviously has native apps, but they also have the web.
... We have, like, customer relationship management platforms.
... Like those that were mentioned, again, they publis.
... To native applications and the web.
... Figma has a desktop client and a web client, so, like, pretty much all of these things involve the web.
Wendy Reid: Yeah, I think that's a good distinction, and maybe we can this is basically the line, I think, and this is what Hidde, I think, was getting at, was like.
... Even if the source of, like, where you're creating the content is not on the web, if it can end up on the web.
... or be cross-published to the web, then that is in scope.
Lisa Liskovoi: I think I think that's great, but.
... I guess my fear, in terms of how we articulate it, is that we do it in a way so that it's not just the piece that's published onto the web that is accessible, so that it applies in a way where the desktop applications and the mobile applications also have to be accessible.
Wendy Reid: Yeah, I think that there's a key part of the wording in there somewhere.
Nicole Letoile: Hi, I'm new, my first meeting, so I'm trying to listen and figure out, what where we're headed and what's happening. Just some context, I spend a lot of time with the accessibility piece of online learning and learning management systems. I think I'm in the right place, authoring tools and helping instructional designers create accessible content.
... And if I think if I hear, what you're saying.
... I always, you know, if online learning is created with tools that produce content that is used through the web, it should apply, so am I are we talking about that? Is that what's happening.
Wendy Reid: Exactly.
... Yeah, I think what we're we're edging towards is that th.
... Authoring tools themselves, be they web-based or software-based, that's just the distinction I will make. Those tools should be accessible, and the features of those tools should be accessible, and then the content that they produc.
... Also needs to be accessible.
Nicole Letoile: Got it, thank you.
Wendy Reid: Yes, yeah, Charles, the content falls under WCAG.
... That's the we get the we get the fine the fine line.
... Alright, I think Where More or less agree. Oh, go ahead, Charles.
Charles Hall: Yeah, it hasn't come up yet today, but it came up in discussions prior to the kickoff meeting, and I've had many side discussion.
... With other peers in the industry of where AI fits into this.
... And it's my opinion and the result of some of those conversations that if AI is used to produce content.
... It's Function of doing so, and its output.
... Has nothing to do with the authoring tool.
... The authoring part of it is the ability for the human to provide the prompt to it.
... So, it's really just, when AI is used in the context of creating content.
... The input method for the human is the authoring tool, and no other part of it is.
Wendy Reid: I want to make sure I'm understanding. You mean, like so, like, let'.
Charles Hall: If you use a web interface to access something like ChatGPT, and you, as the human use an input in the browser.
... to provide a prompt to the AI, requesting Content be created.
... That's the authoring tool, the input.
... Everything else the AI does Is irrelevant to the aspect of authoring.
Hidde de Vries: Yeah, I get that point that you're making, Charles. I do think the kind o.
... what the eye does after I did things.
... It's still gonna end u.
... on the website, and then it is kind of going to become web content at some point, right? So you do some inputting.
... That is authoring, and then you put it on the website, and what happens in between, maybe.
... AI doing stuff, maybe organizing it, maybe doing doing things to it, but if it then ends up on a website, there's still something to say abou.
... how that has happened, so I feel lik.
... whatever AI did, it may, like.
... add headings, or remove headings, or whatever, like, it may do things that are relevant to accessibility, so I think the end result that then ends up on a website is still kind o.
... part of the authoring operation, basically the B side of ATAC.
... And I think what you're referring to is the A side of ATEC, which is about th.
... authoring, whereas, like, the B part is more about the ordered content.
... That makes sense.
Mary Ann Jawili: Okay, I think I understand what Charles was saying, I think that makes sense. I agree with the input needing to be accessible, but I yeah, I guess I jus.
... my experience with, AI is that there can be a lot of back and forth, and it's a very dynamic interface, and I would want that back and forth, the dynamic content.
... the dynamic UI to also be accessible, just to be clear, so it wouldn't just be the input, but everything else, betwee.
... Everything else that's part of that interaction, from the initial input to th.
... whatever content the per- the user wants to use from the AI.
Wendy Reid: I put myself on the queue, to say, because I think the Charles, to your example of, like, the input field and back and forth, so, not all tools do this, but I've noticed that some have started doing this. There's a tool I use called Cline.
... That started doing this in this latest update, where it actually shows when you say, you give it a prompt.
... as it's thinking, it actually shows you what it's thinking. So you can see a feed of i.
... The process that it's going through to produce the output.
... And that, I think, is an example of something where you need to know that that is happening, if it's present, and it's something that you probably want t.
... Not all tools show you, and, like, so when we're talking about the, accessibility of, like, the interface, I think, like, it's stuff like that. It's not just the input field, and it's not just the chat, it's what's happening alongside.
... Tha.
... because you almost need to be able to react to it. Like, I definitely have had interactions with it wher.
... When I see what the thinking what the thinking is versus what the output is, I have to intervene to be like, whoa, whoa, whoa, like, you're on the wrong path, don't make this change to the file, make that change to the file.
... And I guess, like, Shivaji, to your question, like, yeah, good one. I don't know if I haven't tested using Cline with a screen reader. I should maybe I'll do that today when if I if I pick it up, because I'm curious, too.
Sambhavi Chandrashekar: Yeah, I, joined late, so apologies if it's not, relevant, but the two goals for this kind of a standard is tha.
... everybody should be able to author, like Wendy, you were saying rightly, and the only people who can certify are nativ.
... Assistive technology users.
... you and I cannot really how much ever we know how to use. I mean, that's that's not the main point. And the second thing is that the content that comes ou.
... Must be usable by everybody.
... And it's also about pushing organizations that create those programs That spew out the content.
... To be aware tha.
... what they create should be readable, understandable, usable. Even during the prompt and response process, because it's a cycle, right? You put in something, you get a feedback, and then you react.
... All of that should be accessible.
... if it was not relevant to what Charles was saying, pardon me, but Charles, you really got som.
... brain work my brain working in a different direction today. Thank you.
Shivaji Kumar: Yeah, so, so when we think about platforms that create content.
... And we want to come up with a list of tools. So we are basically thinking about some of the larger, bigger fishes, right? So, for instance, ServiceNow, and Salesforce, an.
... The traditional existing platforms.
... Because if we can get them to adopt thes.
... Recommendations or guidelines that we might create, then we have a much better chance at convincing the smaller fishes.
... Is that is that correct.
Wendy Reid: I think it can go either way.
... One thing that I've noticed, speaking from my experience in the EPUB space.
... And digital publishing, is that sometimes it's actually the bigger players that are hard to convince to do to make changes, but it's the smaller, more agile players that are willing to experiment and willing to, like.
... push the envelope, because maybe they're looking for an edge, or it's something they wan.
... Can do, because they don't have to worry as much about pivoting.
... So I wouldn't, like like, yes, it's great if we can get the big players to do it, but it shouldn't be.
Shivaji Kumar: ultimate goal.
Wendy Reid: The ultimate goal.
Shivaji Kumar: Got it.
Hidde de Vries: Yeah, I agree with that, and I think there is also a market incentive for companies that are doing it, because so man.
... Websites need to comply with legislation.
... A company that makes an authoring tool that creates accessible content.
... is more likely to win with those customers. At least that's the theory. I don't know if it's gonna happen like that in practice, but we can.
... We can definitely promote it like that, I think.
... Yeah, it's, it's good if your authoring tool does meet ATAC well, because, yeah, you'll be able to get more, more customers in the sectors that need to meet the legislation.
Wendy Reid: This segues really nicely into what the second thing on the agenda was, which was, why wasn't ATAG2 more widely adopted.
Charles Hall: Just real quick to close out the last one, it sounds like we have consensus that AI belongs.
... being, covered by ATAG in some capacity, but it doesn't sound like we have consensus on how. So, could someone take a stab at adding a how in the in that discussion thread.
Shivaji Kumar: I can give it a shot.
Wendy Reid: Oh, that would be great. Thank you.
Shivaji Kumar: Sure.
Wendy Reid: So, yeah. Why wasn't ATAG too more widely adopted.
... I see you in the chat, Barry. Would you like to say.
Barry Feigenbaum: You know, I have a lot of history in this area. I'm a little bit out of date, but I have a tremendous amount of history of this.
... And My personal opinion, and I think it's well supported, is tha.
... Accessible content will be created when accessible content takes no extra effort to create. Otherwise, it will never be created, simply because most people and most businesse.
... don't care to spend the time or spend the money it takes to make it happen. There are a few, like, ones that have to meet government standards or something, they have other motivations that sort of motivate them to do it. As long as those aren't holding out for them, it really is irrelevant.
... And for the vast amount of content, especially more recently, that exists, that motivator simply isn't there. So, if you just don't personally want to do it, and are willing to spend the extra time to do it, you won't do it.
... And most businesses won't do it, simply because it's not the return on investment isn't there. So somehow, and again, I don't know exactly how to do it, but we have to find a way, in my opinion.
... to do it automatically. Now, I think we have a great tool that's emerging now, which is artificial intelligence, that allows us to find some way to actually do this, to make either the generation or the presentatio.
... do the job for the author, not have the author do the job. And somehow I mean, the author may actually, you know, validate the work done, or somehow, you know, tweak it, make it better, but most of the work needs to be don.
... for the author automatically, not that the author hasn't taken the effort. And that's where our I think our challenge is. Authoring tools need to b.
... embedded in the either content generation or content presentation, probably both, mechanisms, such that they just happen, and they just work. I try to use a point in a comment before, like.
... this is notional alt text. Well, yes, some content author describes a picture that they or an image they stuck in a document. Well, they're describing it based on a completely.
... unknow.
... motivation for the human being read looking at the picture, and so they're describing it in a very generalized way. Some people may want a very summary thing, like, oh, it's a picture of a person. Other people may want, you know, 10,000 words describing every little thing in there. You don't know, as an author, what the reader wants.
... Only the reader knows what they want. You know, whether they're going to spend 10 minutes on a picture or 10 seconds on a picture. We don't know. So it's very difficult to author it. We need to find some way t.
... you know, know the user, and adjust to the user, and all that stuff, and that's in the presentation engine, not in the authoring engine, if that happens. So there's some problems here, at least in my opinion, about, you know, the goals, at least the old goals of ATAG.
... they just didn't quite solve the problem. We need to look at it a different way.
Wendy Reid: Thanks.
Shivaji Kumar: Can I chime in.
Wendy Reid: Well I have a bit of a queue going.
Mike Gifford (CivicActions): Being unmuted is really important, but, thank you. Having good defaults are really important, being able to go off and make sure that your, whatever systems you are have this built in, and this is what we try to do in the Drupal community as much as we can.
... And that includes looking at the authoring experience. But I think there's more than that that is required in order to make this work. Partly, it comes down to investments. When you're dealing with open source tools.
... governments need to invest in this, and be able to say that they're going to either pay a premium to ensure that people are building it into the systems that governments are buying, or other organizations as well, or it has to be something that is required as part of the procurement process. I've seen lots of RFPs, almost none of them mention ATAG, and if they do mention ATAG, it's kind of like .
... They're clueless. They have no idea how much additional they want the benefits of ATAG for no additional price.
... And that's just not possible. You have to pay more money in order to have these additional features, and no amount of AI magic is going to go off to change that, but you can provide carrots and sticks and guidance for legislators, and for that matter.
... businesses to be able to take on these challenges. And there are things like the European Accessibility Act that the only way we're going to actually effectively deal meet the challenge of the EAA is be able to support offering tools better. That's all.
Laura Staniland: Sorry, I was running errands. I'm gonna pull over. I was wondering if an authoring tool would be something like AutoCAD, or, there's a lot o.
... programs where people are 3D printing things.
... And when I tried to do a project once, An.
... I tried to look at the accessibility of the off like, the 3D modeling tool. It wasn't very accessible. So, like, there's a library of accessible media in Pennsylvania, they have a tactile as an example, they have a tactil.
... program, but they have somebody from the library who is cited, becaus.
... A program isn't fully accessible to someone who'.
... you know, got a vision impairment, it would be really great if these 3D modeling programs, or Photoshop, whatever, were accessible. Somebody said the user experience for th.
... you know, the blind person, whatever, needs to also be accessible, and that was one of the biggest barriers I found, and I think, you know, probably they've found, is that right now, things like CAD, AutoCAD, Autodesk, whatever, the.
... They are not fully accessible. I don't know if that'.
... Considered, like, an authoring tool, or that would be in some sort of, like, media working group.
Wendy Reid: That's a really good question. Might have to think about that.
Hidde de Vries: Yeah, so I think the kind of two reasons that ATAG didn't get adopted as much as we would hope. One is, I think it's very big. There's a lot of requirements, like.
... The first SE is literally to meet all of WCAG.
... That's A111.
... And then you haven't even gotten started with the rest of the document. Like, there's a lot of requirements kind of hidden inside of all the requirements, and they all make sense, like, from our point of view.
... But then it does make it very hard for an authoring tool vendor to actually go and comply with all of it, at once. And I think the second is that it's no.
... super practical, many times. I used to work for CMS, and it was really hard to convince managers t.
... spend time on something like this, to say, like, we've got this document, let's try and meet all of it. There was a lot of, kind of, questions around how do we practically implement a lot of these things. Some of them are hard to measure as well.
... Like, there's one, for instance, that is called Help Authors.
... decide.
... and it's super unclear whether you've done enough to help them decide or not, or, like, there's a lot of questions to ask, like, while you're doing the implementation. So I think ATAG coul.
... In our next version, could benefit fro.
... more practical guidance, more kind of information on, here are some examples of how the CMS has implemented it. So I'm hoping we can work together with some of the open source ones, maybe Drupal, through Mike, to show, like, this is an example of how is it solved here.
... To make it easier for folks to actually perform. So there's practical reasons, and there is size reasons, and I don't think we have fixed the size reasons, because a lot of it is really important. But we can probably work on the practical part.
Shivaji Kumar: Yeah, so, mine is sort of a prett.
... It's a curiosity-based question, especially for those who participated in the previous versions of ATAG. And the question is.
... And in addition to regulation-based arguments.
... Did you guys also consider market-based arguments that, hey, if you make your tools accessible.
... then you would have a much better reach to the customer base. How did those conversations go, if they did at all.
Wendy Reid: It's a good question.
... I do not have an answer for it, but, I was thinking in this, too, I think there's, like.
... some of this is timing, right? There wasn't the same landscape.
... 10 years ago for ATAG2 that there is today, so there's that to consider in its lack of adoption. I do think that there is als.
... You know, when there's a lack of clarity, or it's difficult to determine how to implement something, it's easy to just say, let's not do it, because you don't know what you're doing.
... And I think, like, Our opportunities now, too, ar.
... we can think about how to integrate these new tools, we can think about how to make authoring a little bit easier, also just for content creators in general, because I think, like, one of the fears wit.
... Barry, I think you made a comment a while ago that I thought was really interesting about how different types of content have different needs, or people have different needs from similar types of content, and so writing one image description isn't always necessarily sufficient.
... Because users may need different, you know, shorter description, longer description.
Barry Feigenbaum: You know, a different type of description.
... how do you get that? You know, one person can't write all those descriptions necessarily. We have tools for that now, that could do that.
... so I think, like, we also have, like, a different technological landscape to look at, which might make some of thi.
... less scary to implementers. But then we also have, like, the legal side, which makes it more important for implementers, to consider.
Mary Ann Jawili: Hey guys, I put in the chat that, so, I used to work at a company, just very previ- very recently, and just my experience I don't know why ATAG, wasn't widely adopted.
... But as far as, a company goes, like, in my experience.
... There's just a big brush to, like.
... prioritize all these other features and not necessarily, supporting accessible content. And if that if there was a need to if they did work on accessible content, it was just more like, oh, customers have been requesting this.
... So, that that was kind of the motivation, and customers were oftentimes motivated by, like, their legal obligations. So, I just think it does go back to legal, but it it's kind o.
... It in my experience, it's because customers have been putting were putting pressur.
... I would also say.
... sometimes it was, like, the bare minimum, lik.
... Okay, you can set headings, I'm just making stuff up at this. Oh, oh.
... There was something around, like.
... tables, but, in any case, like, maybe like, w.
... there was not a strong interest in making it a in facilitating the creation of accessible content. As long as you could do it, if it was a bit tedious, then that would be okay, if that makes sense. But we didn't want to necessarily invest to make it really easy or automatic.
... For people to create accessible content.
Mary Ann Jawili: Oh, yeah, and EPUBLAND. Yeah, so we were working with, ABC, recently, with InDesign.
... Yeah, there were so many tough conversations, like, well, technically, you can do it, it's just a few hoops you have to jump through, but, yes, you can do, tables, and oh, like, with tables, column headers.
... well, you can't do I don't know, there were I can't remember it anymore, but, like.
... There was a need to create, indicate either row headers or column headers, like, programmatically, bu.
... Since you couldn't even create row headers for, like, people who didn't have visual impairments, then we wouldn't bother to create to enable that functionality for programmatic row or column headers, if that makes sense. So, anyway, you're naughty. Okay, good, I'm glad you understand.
Wendy Reid: I remember, I remember. Alright, with our last couple of minute.
... This conversation has flowed so nicely.
... The last topic that I thought would be worth spending a little bit of time on today was use cases for authoring tools.
... And so I encourage everyone, and thanks, Mike, for providing the first example, thinking about, like.
... taking one step back, you know, erase ATAG2 from your memory for a moment, or if you didn't have it in your memory, don't worry.
... What are as someone who either implements authoring tools, or uses authoring tools, or uses the output of authoring tools.
... What are the use cases today, and what are the use cases tomorrow? Because I think, like, I love a use case exercise.
... As a starting point for looking at, future work, because it gives you an opportunity to think, you know.
... It, like, there's no harm in saying something that already exists, because then it's just like, sweet, we get to cross it off the list, it already exists.
... And when you're talking about something that doesn't yet exist, then you get to have the fun of thinking, okay, well, how do you make this happen.
... S.
... I gave, like, a rough structure, and Mike was helpful enough to give some examples of what that structure looks like, but the use case structure is generally, as a person who does something, I need to be able to do this.
... And a description of what you want to do.
... And so, thinking a.
... people who I guarantee we all use authoring tools every day, and we all author content.
... And we know people who do that.
... What are the use cases fo.
... this new version of ATAG? And think I think, especially thinking about AI, because this is something we want to include. What are the use cases for AI.
... using AI both as a user and as, like, possibly as a receiver of AI-generated content? This is a big question.
... I'm gonna apologize in advance, my background is in publishing, and I chair the EPUB working group, so I EPUB is on my brain all the time.
... like, this is a conversation we're having in EPUP right now, about, you know, how do you tell users, or should users know when content is AI-generated? And how do you communicate that effectively to users, in a way that's no.
... Intrusive, or annoying, or, you kno.
... And, like, effective. And, like, what level and, like, even an additionally fair question of what What level o.
... interference interference is maybe not the right word. What level of participation does the A.
... and the human have, that takes it from AI-generated to human human-generated.
... Like, where's the Where's the line on the content where you no longer Have to worry about that.
... These are all the kind of things that I want people to keep in mind.
... Totally agree, Charles. I think all AI should be disclosed.
Hidde de Vries: To me, I think one of the major use cases for ATAG, and I think also for the AI angle of it, is t.
... Help people create more accessible experiences, so to hav.
... Heuristics that can help people figure out where they may be making something that's inaccessible, guide them towards making it more accessible.
... Those kinds of things.
... come to mind, like, I feel like that is something where I think that can really shine.
... And where we'll also get a lot of adoption, because that is something a lot of people need to do today, and if it makes that easier for them, that's great.
Wendy Reid: Something this conversation actually.
... Had me thinking about was, you know, we haven't defined fully what our deliverables are.
... And I wonder if one of the deliverables we may want to consider is producing a set of sample context files for AI tools around the generation of accessible content.
... So, like, most AI tools, you know, be that ChatGPT, Copilot, Cline.
... A lot of them use there's more and more names for this now, but, like, they're essentially context files, a set of instructions that you can give and say, hey, I'm working on content. Whenever I do this, I want you to look at this file before you start working.
... And.
... It's really easy to write a context file. It's generally just a very simple set of instructions, and you're just giving i.
... You're giving it kind of, like, a persona to take on, and parameters to follow.
... And while, obviously, like, anything we create wouldn't be, like, you have to use this, but just to get people in that headspace of, like, oh, like.
... as an as an, you know, content author, I should do the following, and it's just a list of, you know, instructions that maybe they can build off of for their ow.
Charles Hall: I know we're coming up on time, and we're still talking about use cases, but I think in order to get some of the use cases we'll need, we also have to talk about.
... scoping a definition of content. For example, the definition of content in WCA.
... Includes the information that's communicated to the user, but it also includes the code or markup that makes that possible.
... So, would an authoring tool be a code editor.
... If content is code.
Wendy Reid: I mean, I think thinking about Hidde's example of, like, getting prompted as you go, like.
... we actually have, like, an existing example of that in, like, most, developer environments, right? Code linters.
... that hint to you as you're working, like, hey, that property you're referencing doesn't exist, or, you know, you're you know, that you didn't close that tag there.
... Can we have that for content generation. You're writing a paragraph, or you insert an image, and it's like, hey, you didn't put alt text on this picture.
Hidde de Vries: Yes, I think so. I think we could have that. There's also a bunch of those kind of things that we can obviously do without AI as well.
... I know of one CMS that will not let you save the page if you haven't added alternative text. There's obviously all sorts o.
... things around that, because then people put in nonsense, and then they've met the requirement technically and all that. I've also built a plugin for the CMS I was working at, with, like, some rules that would automatically check if your list was a list and all these kind of things. And that's basically three lines o.
... JavaScript that you can use to do that, which I think, thinking from a sustainability angle, that could be much more effective to have, like, thos.
... few lines of JavaScript, as well as a context file, I like that idea.
... But it triggered me to think about maybe we can, in addition to a context file, also say, and here is a few JavaScript functions you can insert into your editor and have it checked.
... some some basic things that are checkable, that way. I'll pop a link int.
... I think it's, like, 4 or 5 jobs profiles. I know at the Dutch government, some colleagues of ours are working o.
... building this, and I think they're doing that open source.
... But they're building, basically JavaScript that can go into a CMS and then tell you what's wrong and, warn you when you've forgotten something, so that's something we could probably also think of as a deliverable, to have some non-AI solution.
... That is kind of less smart, but also helpful to find some of the things that are easy to find.
Wendy Reid: I think there's a also, I think, and this is possibly something we can put in the document itself as we're talking about AI, because I thin.
... the hype cycle that we're in right now has people kind of over-indexing on AI and using AI for everything, even things tha.
... It's not the best at, especially when it comes to content.
... And I think, like, reminding people as well to kind of push pull back in the other direction to say, like, hey.
... your your instinct is right that this thing can be automated, but AI is probably not the best way to automate it. And, like, you know, using things like code linters, I think, is a good example of that, where, like, code linters aren't AI, they're they're buil.
... Programmatically. Or, like, things like.
... things that pull like, pull things out of content in a systematic way, like, that that's a Python script, or, you know, can be done with with much more automated processes that are probably actually going to be much more consistent than AI, because AI tends to lose it a bit when you give it too much to work with.
... So I think that that's, like, a good actually, that is probably a good use case of, like, you know, as .
... trying to think of how to, like, phrase this, but, like, as a content creator, I want to know when when I should use AI, and when there might be other options out there.
... or as a user, I want accuracy in my authoring experience.
... I want to trust my authoring experience.
... As usual, the chat is wonderful.
... All right, we're just at the top of the hour, and I don't want to run over. This has been a really good meeting, so thank you everyone for participating.
... so far in the feedback, it looks like this time slot works pretty well for everyone, so I'm gonna keep this time slot going.
... I believe at this point, I'll have to double-check my calendar.
... I think I only scheduled meetings till the end of the year, so what I'm gonna do is so we will have one more meeting in two weeks on December 19th.
... And the.
... I will fix the meeting I'll set up the meeting cadence so that it, continues into next year, obviously.
... I will likely, probably not have us have a meeting on January 2nd, because that would be the next time slot, and I'm gonna guess that most people wil.
... probably be not in office, or not in the mood to be in office, even if you're physically there.
... And Barry, I'm jealous, I hope you have a lot of fun.
... But yeah, so we'll have a meeting next meeting on December 19th, and then we'll probably, not have a meeting again until mid-January. I will send ou.
... materials after the meeting. I will let everyone know when I finally finish getting through the edits of the document, so people can look at it.
... But yeah, please keep participating in the discussions. Feel free to open new discussions if there's questions that you have, or things that you just want people to noodle on.
... And we will all we will talk in 2 weeks.
Shivaji Kumar: Thank you. Have a great rest of your day.
Lisa Liskovoi: Thanks, everyone.
Chat Transcript
Wendy Reid: https://pad.w3.org/p/ATAG_CG_Minutes_-_Dec_5_2025
Wendy Reid: https://github.com/w3c-cg/atag/discussions/3
Charles Hall: That seems expected given the age of the Rec
John Jameson: And I’ll introduce myself here too since I’m new — I’m an accessibility tester/developer at Princeton University, and Mike suggested I swing by since most of my work is on writing live validation and inline tips for authoring tools (e.g. Editoria11y).
Hidde de Vries: we just through earlier in the year for WCAG-EM, it was fun
Mike Gifford (CivicActions): https://github.com/w3c-cg/atag/discussions
Wendy Reid: https://github.com/w3c-cg/atag/discussions/12
Adam Chaboryk: Good morning!
Mike Gifford (CivicActions): @Adam Chaboryk feel free to add your info here https://github.com/w3c-cg/atag/discussions/3
Mike Gifford (CivicActions): Twitter is really terrible. We should include that in the spec.
Charles Hall: Or use a web API
Mike Gifford (CivicActions): That is a terrific starting point, but that means almost any digital content. It is a big goal.
Charles Hall: The content falls under WCAG though
Nicole Letoile: Agree
Hidde de Vries: ATAG is like a superset of WCAG, one of the SCs is to meet WCAG
Hidde de Vries: (to +1 what Charles says)
barry: This may not be well received, but IMHO putting the responsibility for a11y on authoring tools is not likely to be generally successful. economics fights us. It is better to try to get presentation tools be able to react well to "normal" (vs. augmented) content. AI is a potential means to make this happen. much more pervasively. Consider alt text for images. There are often many fomrs of it that are needed based on the readers needs. It is unreasonable to expect the author to provide many forms to cover this. Instead the renderer (engine) needs to understand the reader's needs and make the appropriate alt text real time.
Shivaji Kumar: Both content and interface need to be accessible, similar to other tools
Charles Hall: But the output is subject to WCAG
Mike Gifford (CivicActions): How is the human in the loop?
Charles Hall: doesn’t the back and forth occur in the same input field?
Hidde de Vries: Replying to "But the output is su..." but also has a relation to the B-part of ATAG https://www.w3.org/TR/ATAG20/#principle_b1
Charles Hall: But is output from someone or something else a task of authoring?
Shivaji Kumar: Is that feedback picked up by AT?
Mike Gifford (CivicActions): Certainly the input propt and interface with AI needs should meet WCAG standards.
Charles Hall: Replying to "Certainly the input ..." Agree. But the input can be used to subsequently create content. Much the same as user generated content interfaces.
Charles Hall: I have no regard for the size of the fish
Mike Gifford (CivicActions): I’m not concerned with the size of the fish. I’m concerned with the incentives. Things like procurement.
Mary Ann Jawili: My immediate past employer was a company, and I was definitely thinking about ATAG while I was there.
Wendy Reid: https://github.com/w3c-cg/atag/discussions/5
barry: If we want to make accessible content pervasive, we need to find a way to make authoring it essentially free and no extra effort, else it will never become widespread.
barry: ATAG 2 was too expensive to implement, thus is was not widely done.
Charles Hall: One reason is also that it was never regulated as required.
Lisa Liskovoi: I don’t think it’s realistic for accessibility to “just happen” or “just work”, even with AI. AI is not a magic bullet.
Charles Hall: But the image description should align to the intent of the author just like the rest of the content.
Charles Hall: Opinion (aligned to ATAG 2 scope) when the output content is only for the author and not made available to others, the authoring aspect is out of scope.
barry: Yes a11y will never be entirely free, but the closer we can get the better. ATAG 2.0 is IMHO generally too expensive/hard, and anything like it will not be much more successful. IMHO almost any static enablement will not be sufficient.
barry: I will end my soap box now.
Mike Gifford (CivicActions): And the back-end interfaces are so much more complicated than the front-end components.
Sambhavi Chandrashekar: Replying to "I don’t think it’s r..." Conformance with at least 60% of content accessibility guidelines is achievable without human intervention. We could push for that and for alerting humans in the loop to intervene to tweak the remaining.
Mike Gifford (CivicActions): Evaluating Drupal for ATAG was very difficult. So much was judgment.
Mike Gifford (CivicActions): Replying to "I don’t think it’s r..." Better use of automated tools (and some form of AI) is important.
Charles Hall: Replying to "I don’t think it’s r..." Which seems relevant to WCAG but not the author
Mike Gifford (CivicActions): Replying to "I don’t think it’s r..." Why not for the author?
Charles Hall: Those have always been solid arguments for WCAG. But ATAG is about who gets to participate in making the things.
Hidde de Vries: to Shivaji's q: market arguments were definitely part of how we tried to promote ATAG at the time
Mike Gifford (CivicActions): Replying to "Those have always be..." Part A, yes.
Lisa Liskovoi: Replying to "I don’t think it’s r..." I agree that some things can and should be automated, but that’s a way (and should only be one way, not the only way) to meet requirements. Personally I don’t think we should be relying on automation, or thinking about what should or can be automated in how we think about requirements or how to meet them.
Wendy Reid: I have stories from EPUB land…
Sambhavi Chandrashekar: Replying to "I don’t think it’s r..." “Which seems relevant to WCAG but not the author”
Wendy Reid: https://github.com/w3c-cg/atag/discussions/15
Sambhavi Chandrashekar: Replying to "I don’t think it’s r..." Charles, I see your point. We are targeting the system and not the author - I think.
Sambhavi Chandrashekar: Replying to "I don’t think it’s r..." Per ATAG, content created must be accessible.
Lisa Liskovoi: Replying to "I don’t think it’s r..." there’s some nuance there, there is content “at the end”, and there is content that is part of the authoring experience. WCAG covers the first, but not the second.
Lisa Liskovoi: Replying to "I don’t think it’s r..." at the end = after it’s published
Hidde de Vries: love how iA Writer indicates AI written content
Charles Hall: Disclosing AI generation is an ethical mandate.
barry: yes
John Jameson: This is what I do
Sambhavi Chandrashekar: Canada published the Accessible & Equitable AI Systems standards on Dec 3rd.
Mike Gifford (CivicActions): So much can be done with code, but not AI, Certainly not Generative LLMs.
Charles Hall: Most software evaluation and rulesets do not require AI
Hidde de Vries: https://github.com/hidde/porta11y
barry: if a tools requires (say) alt text, what determines if it is sufficient? The author may be a weak judge.
Sambhavi Chandrashekar: Replying to "Most software evalua..." Popularly, there’s conflation between automation and AI-based processing.
Charles Hall: Replying to "if a tools requires ..." In my opinion, the author is the only judge. It retains human agency over intent.
barry: We seems to belive authors are experts. This may nt always be true.
John Jameson: At the moment Sa11y and Editoria11y just have a long list of “suspicious” strings often found in bad alts.
John Jameson: It’s no AI, so it doesn’t really check if the alt is good, but it is very good at detecting PEOPLE who need to read the inline hints on how to write good alts.
Sambhavi Chandrashekar: Replying to "We seems to belive a..." Barry, the business reason of ease of implementation you made is very critical.
Sambhavi Chandrashekar: Replying to "We seems to belive a..." … to achieve any degree of success with a new standard.
Lisa Liskovoi: Replying to "We seems to belive a... yes, and… that’s why we NEED regulation for adoption
Hidde de Vries: 'i want to trust my authoring experience ' - wendy (capturing in case the captioning didnt catch it, great way to put it )
Laura’s iPhone: I’d love it if there was a warning about flashing lights in coding liners, generative AI, etc.
Jeroen Hulscher: Thank-you!
Nicole Letoile: Thank you