Ai-Media Transcript -THE_STATE_OF_ACCESSIBILITY ()
FEB_25_2025


             Greetings, everybody. Hello, I'm Ryan Bateman from Deque. I'll be moderating today's session titled The State of Accessibility with Preety Kumar and Dylan Barrell.
I'm going to take care of just a few housekeeping things before handing it over to Preety. First, the session is being recorded and will be hosted on demand for registrants immediately after this live session. Slides for today's session are available on the web page below the video feed.
If you require live captions for today's session, you can access them in the YouTube video player. There's also a transcription link on the sessions page below the video feed if you'd like to digest your captions that way. We also have ASL brought to you today by the lovely folks at WCAG.
We'll try to save the last ten minutes or so for a session Q&A. Please take your questions to the Q&A tab that's located next to the video stream player. Lastly, you might notice I'm wearing a stunning purple t shirt that says Axe-con is for lovers of accessibility with the words with the word lovers in a script type.
Folks who fill out the post-event survey will be eligible to win one. Your feedback is super, super important to us. It helps.
It helps us improve this conference every year. So please look out for that post event survey after day three. But for now, let's get this thing started.
Preety, the floor is yours.
>> Thank you Ryan. I look forward to this day all year long. What if I told you that 100% automation in accessibility testing isn't just possible?
It's holding us back from by not being there already. I know this journey well because I've lived it from the very beginning. When I founded Deque, automated accessibility testing simply didn't exist.
0%. Every check had to be done manually. It can't be automated.
They said accessibility is too nuanced. Too many variables. Even my friends, like the late Jim Packer, the father of digital accessibility and others in the industry, thought I was chasing an impossible dream.
And this was when the web had only 17 million websites. But they persisted with with Act score, we reached 57%. Automation by volume, a milestone those early skeptics had insisted was impossible.
Each percentage point meant thousands of developers could catch issues earlier. I have been. When it's going to be ready.
Funny how never becomes hurry up when you prove something works. Now in 2024, with over 1.1 billion billion with a B websites, it seems like we are right to worry about automating the other 43%.
Well, today I am telling you that I and everyone at TC is committed to achieving 100% automation. Or so close that the difference won't matter for digital accessibility testing in the very near future. Today, we stand at the precipice of a transformation.
The European Accessibility Act and Artificial intelligence advances are converging to make comprehensive automation not just necessary, but possible. For developers, this means catching issues in real time, no matter how the code is being written. For testers, it means focusing on creative challenges rather than repetitive checks.
For accessibility experts like me, it means elevating the work to have more strategic impact. This isn't just about efficiency. It's about finally building the accessible digital world we've been promising for decades.
And this time, we have both the imperative and the means to deliver on that promise. I've seen this industry grow from zero, barely existing to where we are today. For years I've said if only we had the legal requirements with teak, and if only we had the technology.
Now, with the European Accessibility Act and Artificial intelligence arriving simultaneously, I've had to find new things to complain about. It's been challenging. Trust me when I say the next leap forward isn't just another incremental step, it's the transformation we've been working towards all along.
So it starts with the European Accessibility Act, which brings a whole new level of urgency. The European Accessibility Act isn't just another regulation. The European Accessibility Act is poised to transform digital accessibility the same way GDPR revolutionized data privacy.
Remember how GDPR fundamentally changed how every organization handles personal data? The European Accessibility Act will do the same for accessibility. You know, in these early days, my European team is telling me that many companies that they talk to in Europe are treating it like a surprise quiz surprise exam that they haven't started studying for.
But just as GDPR made privacy by design and by default a global standard, the European Accessibility Act will make accessibility part of every digital products DNA. Think about what happened with GDPR. Organizations scrambled to comply.
Privacy tools and automation flourished. New roles emerged. Now, companies that waited too long faced massive fines and reputational damage.
Fortune 500 companies, I think, spent $7.8 billion in preparation for GDPR, spending an average of $1.3 million a year to maintain compliance.
Noncompliance fines have amounted to over $4.8 billion. 80% of organizations have altered their privacy.
Its scope is just as broad. No way broader, affecting millions of websites, mobile apps, digital products that must begin being accessible by 2025 and must be completely accessible by 2030. In five years.
And like GDPR, the impact will extend far beyond Europe. Global companies won't maintain separate, accessible versions for Europe and inaccessible versions for the rest of their of the world for their products. They'll standardize on accessibility just as they standardized on GDPR level privacy practices.
The penalties are similarly serious, and the requirements are just as clear. GDPR was about protecting data. The European Accessibility Act is about protecting human dignity.
And here's the critical difference. While GDPR could largely be addressed through policy changes, changes to a couple of web pages, and careful data handling, the European Accessibility Act demands continuous technical validation of every digital user interface. When every code change needs to be checked for accessibility impact.
When every new feature needs to be verified before deployment. That's when 100% automation becomes not just desirable, but essential for survival in the digital marketplace. GDPR changed how we handle data.
The European Accessibility Act will change how we build everything. GDPR gave us those lovely cookie pop ups on every website. The European Accessibility Act might actually give us websites that function for everyone.
Revolutionary concept I know. The European Accessibility Act makes digital accessibility necessary. GDPR fines made companies care about privacy.
The European Accessibility Act will make companies care about accessibility. Money, teaching empathy since forever. So that's why necessary.
Now let me tell you why it's possible. Let me share two breakthrough innovations in artificial intelligence that are transforming accessibility testing. First, 100% automation of routes.
Consider all text analysis. Traditional tools could only verify if the alt text was missing. Now our artificial intelligence actually understands images and can automatically determine if the alt text accurately describes the image in its full context.
What took me as an expert minutes to do in the X platform does in seconds. It doesn't even stop to sip that coffee. This is part of the 100% automated rules in the X platform.
Building on this, our intelligent guided tests work hand in hand with human experts using artificial intelligence to accelerate the path to complete automation. We're not just automating tests. We're automating understanding.
The second breakthrough is a genetic AI. This represents artificial intelligence that truly. Understands context and purpose operating.
Shifts in what's possible. Beyond the fascinating technology, the real impact is profound, comprehensive, consistent, automated testing that frees human experts to focus on innovation rather than repetitive tasks. EAA makes digital accessibility necessary.
AI makes it possible. That's where the artificial intelligence revolution meets this European Accessibility Act regulatory imperative. The convergence that I mentioned just a few minutes ago is perfectly timed.
AI's capabilities are maturing just as the European Accessibility Act makes comprehensive testing essential. Maybe sooner. What we are going to do is find all defects automatically.
Not just find them, prevent them. These two revolutions in our industry, the European Accessibility Act and artificial intelligence, are incredible drivers and enablers, but are super powerful when they build on a wonderful platform. Acts we are building on top of the most widely used accessibility platform in the world.
It will be really wonderful. Super cool. Really really cool actually.
E brings the must do. It brings the can-do. Deque brings the watch.
This. My job was what if this? What if that for 20 years and Dylan and his team's job was figuring out how to without visibly rolling their eyes at my requests?
Dylan and his team have been remarkably successful at half of that job. Dylan will now explain how we turned this into actual technology. I suggest paying more attention to him than you did to me.
He deals in reality. Over to you, Dylan.
>> Thank you. Preety. The Deque vision.
Our ten year big, hairy, audacious goal has been to achieve the situation where 100% of accessibility, development, testing and fixing could be achieved with zero accessibility knowledge. And as Preety pointed out, we're even more excited about our ability to help you get there soon. As Preety mentioned, there are two big waves that are changing our industry right now.
Let me tell you what Deque is doing and how this is affecting development. The European Accessibility Act, or EAA, has really changed the trajectory of accessibility in a way that we believe will lead to a significant and, more importantly, a sustainable step change in adoption and implementation of accessibility. When crafting the underlying directive, the European Union was really only interested in a sustainable process, and they've structured the directive in this way.
It requires organizations to take the approach that has been advocating for the last ten years of shift left, otherwise known as sustainable accessibility. It requires organizations to embed accessibility into their processes. Why do we think that this will lead to a step change?
Because it applies to any organization headquartered anywhere in the world that sells into the European Union market. We've already seen many global US based companies investing significant amount to get their houses in order. This can only have a positive effect on the accessibility of the offerings around the world.
To support this, Dcrs invested in the last year in translating and localizing our products for the EU markets. And I'm very pleased to announce that you can now take all 27 courses available on Deakin University in French, German, Spanish, Italian and Dutch and Japanese as well. Even though it's not European language, this means that more people can learn about accessibility in their own language.
Broadening the reach for accessible knowledge. You could say making it accessible. We've also committed to supporting French, German, Spanish and Italian in the user interfaces of all of our products, many of which are already released or will be released very shortly.
I'm European, and I know how much it means to Europeans to be able to do things in their own mother tongue. And I'm especially proud of this commitment. Racing this change with gusto.
We once thought that the Turing test would be enough to tell the difference between a human and a computer, but it turns out that the Turing test is not a good enough measure, because although AI chatbots can now pass that test with flying colors, we all know that they are not quite up to the capabilities of humans in many areas. So all sorts of new benchmarks are being created and then very quickly bested by new advances. Just to illustrate how good the AI models are getting, according to OpenAI's latest releases of the O3 model, O3 is now better at competitive coding than all but 75 of the very best human competitive coders.
The three benchmark scores of the best AI models are above 70%, and this is for real world coding tasks, which include standard tasks that would normally be performed by a human coder. The advances in efficiency have gone so far that Salesforce recently announced that they are not going to hire any more software developers in 2025. And with advances like clothes, computer use, AI is moving into the realm of testing in terms of adoption.
According to Forbes, Google says that 25% of its code is now generated by AI, 52% of freelance designers use AI to help them design. And according to GitHub, 92% of US based developers use AI. And the new models are being researched and tested continuously, including a recent research paper that uses a non token based reasoning approach combined with a token based reasoning approach.
And this allows the models to think in terms and concepts that are not easily represented in text. The results are much faster and better reasoning with smaller model sizes. And many of us have heard of deep tech and the angst that its announcement unleashed in the AI and even in the international relations realms.
And one of the big advances underlying that announcement was the supposed emergence of reasoning without human feedback, using a technique called reinforcement learning with verifiable outcomes or just plain reinforcement learning. So the changes are discontinuous and they are unpredictable, even for the best experts in the industry. Take a look at Deep Sea, for example, and we could see artificial superintelligence next week.
Or it might take many years. But one thing is certain the AI capabilities we have today are the worst they will ever be in the future. And yet, with all of these advances, because of the limitations of the current transformer approach, the human coder in fact the human when you are thinking about design as well is still essential.
And the code or the artifacts that are being generated is still not accessible. We still need testing for functional and nonfunctional requirements, and we need human oversight. What we are doing at Deque to accelerate the creation of accessible digital experiences is building on the foundation of using AI in accessibility testing that we pioneered in 2018, when we developed proprietary machine vision models.
As this technology has advanced over the last two years, we have upgraded our capabilities to take advantage of these advances and have two major thrusts. The first is Access Assistant and the second is doubling down on automation in the access tools. Access assistant, which we first announced and started limited customer availability last year at Axe-con, has advanced.
We've advanced access assistants capabilities significantly, and we are announcing general availability of this here at Axe-con. Not only can access System give you Deque level expert answers to a wide array of questions on accessibility, from policy to detailed technical questions on very specific technologies, but Access Assistant can also answer all your Access Tools related questions, because it has access to all of our documentation. In fact, what Access Assistant is leveraging is what we call the Deque corpus.
It is all the high value accessibility and product documentation that Deque has built up over many years. Access system can also explain accessibility defects found in our tools and suggest fixes to these issues, including source code fixes. If given the right context, Access Assistant is your digital accessibility expert, your tools expert, and your helper in implementing and fixing accessibility issues.
Its on demand knowledge. In addition to the chat interface integrated into D-q, University Access Assistant is now able to join conversations inside your Slack or Teams workspace. And provide insights with the context of the chat thread.
The chat context to answer the question and any of your follow up questions, all with guaranteed high quality answers. Because the answers are constrained to the Deque corpus of knowledge. This is an example of Dca's philosophy that was validated by Forrester last year, that these tools are the best tools for use by people involved in the content creation process.
We meet you where you are. I'm very pleased to announce today that the Slack and Teams integrations are available for limited customer use right now. Please contact us if you would like to participate.
Access assistant is also being integrated into the tools to help with understanding and fixing accessibility issues. Please attend Harris Schneiderman's talk on Wednesday to see what we will be releasing related to this shortly. But as we have said earlier, the development workflow is changing as AI advances.
It is now possible to have Agentic systems that use Llms and other tools via protocols like for example, the Model Context Protocol. What the Model Context Protocol allows is a standardized way to make tools available that the agent can decide based on the context to pull in when they are needed. The agent takes instructions from the user and works independently for potentially a long period of time, and then comes back and presents its findings or results.
Sometimes it comes back and asks for more clarity. Ask clarifying questions, but at the end it presents its results and the user can then provide feedback and the agent will go away and modify its work with the updated input. This carries on in a loop until the user is happy with the results.
During this loop, the agent is generating code using AI reasoning and pulling in the tools via the model context protocol that it needs to efficiently achieve the outcome of the user. The results of using AI agents, as opposed to just AI models, is results that are much more mature, accurate, and closer to what the user wants. Also, the external tools allow for solutions to be generated that are less prone to the AI hallucination and variants, but without some help, the user still needs to be an accessibility expert in order to produce accessible outcomes.
With the off the shelf solutions. But when you introduce the platform's capabilities into the system, this is where the real power of the AI coding agents is unleashed. From the perspective of accessibility, the coding agent has access to Access Assistant and to access tools via the Model Context Protocol.
These register the access platform in various roles. Roles. As a specialist in accessibility, accessibility requirements and accessibility tools that can be leveraged whenever needed.
The developer does not really need to do anything differently. They use the coding agent to implement the feature they are working on, and the coding agent talks to the platform whenever it deems that to be necessary. For example, the agent may call Access Assistant to get requirements for the accessible implementation of user interface features or specific user interface components.
Then it writes the code to meet the functional requirements and the accessibility requirements returned by Access Assistant. After generating the code, the agent uses the tools to validate the code and fix any issues. The Ex tools find and all of this happens automatically, but under the supervision of the user.
But Dylan, you say ex can only find 57% of the issues. Well, I'm very pleased to announce the second major thrust that we have to increase efficient, accessible development. We are doubling down on the AI empowerment of ex tools.
We are announcing here today two new capabilities advanced rules and AI assisted testing that, together with our existing capabilities of the intelligent guided testing and the ability to record and replay test sessions, drive our fully automated testing up to and beyond. 80%, of course, of the issues that development teams make by volume. These testing and validation capabilities are equally useful to a development team member in a more traditional workflow, or to the coding or testing agents in a more modern workflow.
We are meeting you and your organization where you are today, with the tools that make your team members more efficient and more effective, while also ensuring that as your workflows change, the advances that you have made in accessibility can be maintained and advanced to. Please attend my session on the future of the access platform later today. To get more details on this.
And other advances we are making to access.
>> Turns out if you complain enough about manual testing, someone will build an AI just to get some peace and quiet. You're welcome my dearest digital accessibility industry. Don't miss Dylan's session, where he'll dive deeper into how he's building the future of accessibility tooling in the Ax platform.
Now, let's talk about what this means for our industry. You're witnessing a fundamental transformation in how accessibility work is done. Think about the pattern we've seen with accessibility innovations of the past.
They start as a solution specific herb for specific disabilities and then consistently ripple beyond their original purpose to benefit everyone. A decade ago, I remember we explained how captions weren't just for the deaf community. They helped everyone in noisy environments.
Today, those same accessibility principles are driving artificial intelligence advances. Captions generating videos. All text creating images.
Screen reader workflows. Inspiring voice interfaces. Your work isn't just following the AI revolution, it's enabling it.
The demand for accessibility expertise isn't just growing, it's about to explode. The European Accessibility Act alone will quadruple the market need. And unlike the manual methods of the past, you'll be working with intelligent tools that handle repetitive testing while you focus on the creative challenges that make accessibility work meaningful.
For our European partners. We are building an ecosystem where you can use our platform to address the urgent need for EAA compliance. Now, while our tools will automate compliance testing, the crucial usability testing still requires human expertise.
Please contact us to explore. How to deliver your services alongside our products and extend our joint mission's reach. The tools that Dylan highlighted aren't just about making accessibility easier, they're fundamentally changing how digital products are conceived, designed, and built.
This isn't an evolution, it's a revolution. This transformation isn't happening in isolation. It's powered by this incredible community.
With 2.5 billion downloads of Axe-con. Actually, I stopped counting so it could be more like 18 million downloads a week and nearly 30,000 of you attending Axe-con.
This accessibility community is stronger than ever. For those of us who've been at this for years, your expertise is more valuable than ever before. For those of you just joining, you're arriving at the perfect moment to shape how the digital world will work for everyone.
And here's the best part the impact of your work will extend far beyond compliance or automation. Every barrier we remove, every test we automate, every product we make accessible. It all adds up to real change in people's lives.
That's what drives us forward. You know, someone recently asked me why we named our tool AKS. They wondered if it was just because we wanted to chop down accessibility barriers.
Yeah. Partly true. Other reason was because accessibility starts with A and X just sounded cool in 2015.
But between us, I think axes are just more practical. Chainsaws are flashy. Sure, lots of noise, not to mention being a little smelly.
Have you tried to build something precise with a chainsaw? They are better at cutting things down an ax though. That's a tool you can count on.
Precise, reliable, noise free zero false positives. Ax gets the job done without the drama. That's really what this community is about.
Not making noise, but making progress. Building something that works for everyone. One precise fix at a time.

>> And I just want to call out that we had on screen briefly for a second there. Before we jump into questions, I want to congratulate Procter and Gamble, Starbucks, American Express, Doctor Gregg, Doctor Greg Vanderheiden as this year's Acts Award winners. As an annual tradition, the Ax Awards recognize significant contributions to digital accessibility in the areas of scale and culture and long standing commitment.
So we're going to continue that celebration of those great contributions in discord and in social media. And now we'll get to a couple of questions. Overwhelming most popular question.
Preety and Dylan, for you. Could you say a little bit more about what 100% coverage means?
>> 100% automation and coverage by volume of accessibility defects found in the world. That's what we mean. We are intending to extend and cover whatever automation can do, because we think that is necessary at this point.

>> All right. Thanks very much. And then another question is, is really about the scope of EAA.
I think some folks could use some clarification. If you can shed some light on the scope and applicability of EAA. Why is it something that the majority of our audience should be considering?

>> Yeah, the European Accessibility Act covers, you know, it's very clear that anybody really making content applications, anything available to European consumers has to address accessibility as a human right. It doesn't matter if you are located in the US, Canada, Singapore or wherever. It doesn't matter as long as your website or your product.
Digital product is made.
>> Preety I think we might have lost your audio there for a second.
>> Yeah. We definitely did. Well, let me carry on from where Preety was.
And she may join us. Hopefully we get her audio back. But as you were saying, as long as your product is consumed by people in the European Union, you are required to make that product accessible.
And there are there are very specific domains that they're focusing on. They're focusing very much on transport, very much on financial services, education, all the things that really make a big difference in people's lives, in particular digital lives. And it does cover kiosks as well as mobile applications.
So it's a very broad it's a very broad law and law and, and you know, for those of you who've been keeping up with GDPR, you'll know that last year, Amazon, just for what they did in the Netherlands, was fined almost $400 million for not complying with GDPR. There are similar sorts of teeth associated with this, so it doesn't matter if you're a company that is headquartered, you know, in the US or in Japan or in China, if you are selling into the European Union, you are subjected to the same laws that the European companies are who are also selling into the European Union. And that's really why we think it's relevant not for every company, but for a very, very large number of, of companies based in the US and other countries outside of, of the European Union.

>> Thanks, Dylan. I think we could also I know Preety began to address this towards the end of the presentation, but I think it might need some reiteration for the group. Could you talk about the role of the use of AI automated testing in combination with accessibility experts and using their their expertise leveraging people with disabilities?
Of course. Could you explain a little bit about what does that look like in the future?
>> Well, it depends how far into the future you go. You know, so this year, I think, you know, there's, there's we're not going to get to even though Preety likes to say we are going to get to 100% this year. I think we're going to get we're going to make significant progress this year, but we're not going to quite get there.
And so there's always going to be that, you know, that that gap that needs testing. As Preety pointed out at the end, usability testing is still one of the areas that that there hasn't been a lot of progress in terms of. What I can do.
I know there are companies that are working on that, but the. It can utilize the their expertise. And then as I pointed out in my part of the talk right now with the kind of state of the art, you know, the human still needs to be in control.
The human still needs to make the ultimate call as to is this good enough? Is this does this really meet my requirements making you know, testing doesn't go away. So all of those are areas where people with, with, with current skills can.
Can be involved and they'll just be able to do more. I mean, the way I see this is very similar to when we first developed the first color contrast, automated color contrast testing. Before that, you know, in every single page with every different color combination, you used to have to go and take a color picker, figure out the color of the pixel, and then go and do a complex mathematical calculation just to figure out whether it passed color contrast or not.
And that's very tedious. It's not really using the skills of an accessibility expert at all, but by by automating that in Xcor and taking a lot of that away, we turned that into, into an instantaneous click of a button. And that freed up the accessibility experts to then focus on the things that that were more important.
Also focusing on policy and that sort of thing. I Preety you back. Yeah.
I'm back.
>> Sorry. My audio.
>> Conked out for a couple of seconds. I think it's actually our responsibility as subject.
>> Matter experts to provide the oversight to be in the human in the loop, and not let AI rely on AI without giving that oversight. It's our responsibility to do that because they will always be human. Oversight will always be needed.
Always. And I think that's our role. Our role is to make sure first we're providing that oversight.
And second, we're looking down the horizon, addressing the next challenge where our creativity is needed.
>> All right. Thank you so much. I think we've got time for one more question.
And I think it's in the same vein. Right. A lot of questions on the subject relative to the use of AI.
Yeah. How can we ensure that it's being employed in an ethical way. Right.
We've heard in past years. Right. We've talked about, you know, in cases of A.
I. Maybe amplifying and causing problems, especially amongst the disability community. How are we ensuring that this is used ethically going forward?

>> Yeah, I'm going to let.
>> Dylan answer the technical part of it. Let me just give you a quick perspective. I think biases in artificial intelligence are going to be there because it is reliant on the people that actually created these large language models.
And that's where I find that it's really important as we integrate these tools, that we are doing it on a corpus of data, like the Deque way and the data that we've developed for 25 years, which doesn't have those biases built in that we built. On top of that, we add to the data that is out there that may have some of the biases. Dylan, anything to add?

>> No, absolutely. So first of all, I will say that that it's a double edged sword, right? It has huge advantages for people with disabilities.
I mean, things like be my, be my, what be my eyes has done have just really sort of created possibilities for people with disabilities that weren't there before. So, so it's definitely a double edged sword. But there's biases in the data.
And I mean, there's all sorts of social biases in there. You know, when it comes to race or or gender or whatever it is, those are in the data. And in the same way, there are biases in this underlying data that relate to accessibility.
Accessibility is a is a minority concern if you if you want to look at it that way from a UI development perspective, it's been in the minority. And so that's why that's one of the reasons why these models are not good at creating accessible experiences, you know, a consistent basis over and over again. Right.
That's so and I think that's that's also why we've taken the approach of building on the model context protocol as one example, where you can use more heuristic approaches to augment that extremely that extreme power that the AI does bring us, and to augment that with consistent heuristic approaches that can can really augment that. So I think, you know, we're always developing ways to, to, to do post what's called post-training and find ways to improve the. Models output.
But I think that those those approaches.
>> Engage in Q&A and chat. Really appreciate it. I hope you have a fantastic rest of Axe-con day one.
Cheers, everyone.
>> Thank you.
>> Thank you.