Season 01 | Episode 07: Accessibility and AI

David: Welcome, I’m David Best and this is Practical Accessibility Insights. The podcast brings you tips and tricks on making your workplace more digitally inclusive. With me is my co-host Sandi Gauder. Hi Sandi, how’s it going today?

Sandi: Hi David, it’s going well, no complaints.

David: In this episode, we’re going to be exploring the whole area of AI accessibility, the good and the bad.

Have you had any experience with employing any of the AI tools within your workplace?

Sandi: I actually have. I’ve been experimenting probably the last couple of weeks with ChatGPT. I’m building a new WordPress theme and I’ve been talking to Chat about, you know, solving some styling issues or scripting issues. It’s been interesting, it’s been helpful, and you really do have to watch what it gives you, because I’ve noticed that some of what it provides isn’t necessarily accessible.

So if I remind him I want accessible content or code, he’ll fix it, but you definitely need to know what you’re doing. You can’t just rely on it without checking the answers that it gives you. I can see the benefit of it, for sure. I can see how it can speed up the process in different ways.

David: Yeah, absolutely.

And I think everybody’s sort of dabbling in it and trying to get the best value out of it. The challenge, of course, is the dangers of AI and not really understanding what you’re doing. And, you know, it could create more work for yourself. So that’s what we’re going to learn today. So why don’t you go ahead and introduce our guest for today?

Sandi: I would be happy to. So today we’d like to welcome Juan Olarte to the show. Hi Juan.

Juan: Hi David. Hi Sandi. Thank you so much for having me in the show today.

Sandi: Thank you for joining us. I think we’re going to have a really interesting conversation today about AI, but before we do that, can you just give us your elevator pitch?

Tell us about you, who you are, what you do, what company you work at, all that kind of stuff.

Juan: Great. Thank you. My name is Juan Olarte. I’m the CEO and co founder of a company called Digita11y Accessible. We help organizations integrate accessibility into the development lifecycle of their products. We do it with what we like to call DAS, which is Digital Accessibility as a Service, and we do that in three main areas: technology, services, as well as subscriptions.

At the back of it is our technology, and our technology is AI based. So, one of the things we want to make sure we do is not only to include accessibility, but how AI can make equal the playing field for everybody.

I’m originally from Colombia, I’ve been in Canada for the last 26 years. For the last 24 years, I’ve been working pretty much in accessibility as well as development.

Sandi: So you know what you’re talking about when it comes to digital accessibility and AI. It’s a great combination.

Juan: I like to say that I’m always learning. I can tell you no matter how much I learn, every single day is like, Oh, I didn’t know about this.

And you keep learning.

Sandi: Sure. That’s what makes this, this work that we do in the field of digital accessibility so exciting, I think, is because it’s constantly changing, constantly evolving, and you, you’re constantly learning. You can’t not learn every day. It’s, it’s really good.

So how would you define AI in the context of accessibility, even the whole idea of AI overlays?

Can you just tell us a bit more about those kinds of things?

Juan: Definitely, and actually there are a few things in there. So when it comes to AI and accessibility, I see AI as a big enabler for people with disabilities, and I will touch more on that later. When it comes to overlays and AI, a little skeptical.

There are a lot of companies that came to use AI for their overlays, but the reality is that they don’t really work. When you look at an overlay, basically what it does is try, it tries to inject HTML, JavaScript into a page and then changing the HTML tags. Oftentimes, it also interrupts or create, creates obstacles with assistive technologies.

So it’s very difficult for overlay to really try to make something accessible. In fact, many times it just creates obstacles for people with disabilities. When it comes to AI and accessibility, I’m very, very optimistic. I believe people with disabilities are going to have a great opportunity to level the playing field in any kind of business or subject that they want to go for as long as they use AI.

Now, when it comes to it, it really, there are two, two different things. A, how accessible an AI tool is going to be for the person with a disability. But the most important, in my opinion, is going to be how that person is going to be able to interact with the tool, meaning prompting, layering, creating the specific personas.

Sandi: So, AI would become almost like another assistive device, like a screen reader, is that what you think? That’s where it’s headed? Okay.

Juan: So, so I believe, I honestly believe it’s going to be an assistant. Currently, you mentioned about ChatGPT. There’s a few things with ChatGPT, Claude, or any other models.

Right now, you’re able to interact with it, either inputting text or talking to it, which is going to be great. You have some models like ChatGPT that have vision. I believe in the future you’re going to be able to interact with all of those three, three specific inputs, regardless of ability.

So if you’re not able to speak, you’re going to be able to type. If you’re able to speak, but you’re not able to see, you’re going to be able to provide auditory information and receive the same. And if perhaps you’re not able to talk, maybe you’re deaf, vision is going to allow you to still use sign languages and the AI is going to interpret that information and then provide you with the answers.

Now what we’re going to see over the next and I want to say I don’t want to say years because right now with AI is changing on a monthly sometimes on a weekly basis.

It’s going to be of AI systems to do actions for you. So right now they’re very good at predicting. Let me take two steps back. Right now, when you interact with an AI, it’s going to try to predict the next word. And it’s at it’s core what AI does it was trained on everything on the internet, good and bad, and it’s going to try to create a prediction.

Just like when people are driving, they’re going to predict, should I go left or should I go right? What is going to happen over the next few months you’re going to see more capability of AI to be able to reason as well as to execute actions. So if I’m a user, I can come to my AI and say, can you please tell me what kind of work do I have over the next week? Can you help me create an appointment? But also can you help me prepare for that?

David: I think that AI is certainly changing our lives. The artificial intelligence built into products is making things like GPS guidance, even screen readers have built in AI now to describe pictures and what have you. The thing I wonder, though, how effective is AI in making websites and social media communications more usable, more effective for people who have vision impairments?

Juan: So it really comes down to how you’re going to be using AI. Let’s say you’re going to be creating content, or in Sandi’s case, she’s creating a website. You can ask ChatGPT to help you code it. Now, before you try to give it specific prompts, you have to be able to create a persona. You can say, I want you to act as a web developer with this much experience in front end, back end. And also you have to ensure that you provide accessible content.

Then the next one is you’re going to start providing prompts and then you’re going to be receiving information. Now, that is going to be great when it comes to creating content, whether it’s going to be for development.

Now, when it comes to a person interacting with AI, I believe we’re going to be able to see, with more and more advanced reasoning models, they’re going to be able to understand where you are and what you’re trying to accomplish, and they’re going to be able to guide you through it.

So imagine if you’re not able to see a website, but you’re going to go through your AI assistant, and you’re going to tell it I want to be able to go and check my inbox and I’m actually looking for this person. The AI is going to go and it’s going to open your inbox and it’s going to tell you, you have 10 messages. Out of those 10 messages, 3 are junk mail, the other one is for the person you would like, you were expecting to receive some information.

Would you like to answer? Now, the key portion of it is that the AI is going to learn from you and from your interactions. So every single time you talk with it, you give it more instructions, it’s going to know how to act on your behalf.

David: Yeah. So small business owners are dependent usually on third party developers to build a website for them, and I understand that you’ve developed a website auditing tool, Digital A11Y, I believe it’s called. Can you tell us about that tool and what do small business owners need to know if they want to employ a service like that?

Juan: Correct. So let me start with Align. We call it Digita11y Accessible and the product itself is Align.

At the core of it is simplifying the identification as well as the understanding and creating tickets about accessibility. So if you were to be doing an accessibility assessment, you’re going to identify the specific issue, but you’re going to have to provide the information.

Our AI model, you’re going to provide visual information, or you can provide the code and tell it, I went to this website. I tried to interact with the accordion and the accordion doesn’t have the expand collapse functionality. This is the code, right? The AI has been trained into taking that input, put it into a friendly ticket. You’re going to review it. And then once it’s ready, you can send it to Jira, so forth and so on.

Then if a developer receives it they’re going to be able to ask questions, but also understand why and how that specific ticket is making or creating obstacles to people with disabilities. Now, that’s our product. But when it comes to how users or small businesses can utilize AI, I, again, I’m super optimistic.

I think long days in terms of like having a bigger company to where you have to hire developers are going to come to an end. Right now, you’re going to be able to talk to an AI and say, Hey, I want to be able to create an interface, or I want to be able to create a login page. These are my requirements.

And the key portion in here is going to be requirements in terms of like, what are different fields going to be like, if they’re going to be accessible, how accessible they’re going to be. Remember, AI is just a tool and is going to help you develop that, but also AI can help you ideate or brainstorm any kind of issues you have for your organization.

As long as you know how to prompt it, it’s really going to help you get there.

David: That sounds good. I think a lot of small business owners would be very excited about the possibility of not hiring external website developers. I, I think that would be a great advantage. What are the dangers though, that people should be aware of when they’re looking at procurement of these kinds of tools and trying to make them so that they are as diverse as possible for clients and employees?

Juan: When it comes to AI it can provide false information and if you don’t know about it, you’re just going to try to do it. I give an example. When we were training our AI to go through accessibility tickets it will try to make information that was not factual. So we’ll have to go back and we have to retrain it. We can we even ask why do you provide this information and the AI doesn’t know doesn’t know why. But one of the things we need to be very careful with is the information we get from there. If you’re not a subject matter expert, you’re going to have to double check.

Now that with that said, what I like to do, I may have two different models, models. Let’s say we have Claude and we have ChatGPT. I give them the same task and they’re going to give me some information. When I don’t know something. Let’s say I’m trying to develop a webpage and I’m going to have an expand collapse FAQ question.

I can ask, can you please tell me, how can we make this accessible and make it clear to me as to how people with disabilities are going to interact with it. So even before it provides the code, it’s going to tell you, okay, collapse, expand accordion to be accessible and needs to be keyboard accessible.

You’re going to have to use semantic HTML elements. This is how you’re going to do it. And then it’s going to start providing that information. So A is going to be education. Once you understand, you can actually move into creating it. What I believe AI is going to do is really going to simplify the identification, understanding of knowledge that we can use in processes.

Sandi: I mean, that all sounds amazing to me. But I wonder if, is there going to be knowledge that gets lost because we start relying on these amazing tools to do the work for us?

Juan: Perhaps not so much knowledge but how easy you can access it. So if I were to look at it, we’ve been doing accessibility for a long time.

We know exactly, I don’t want to say exactly, but if we come across with specific elements and it’s not accessible, we have an idea as to why it’s not. Somebody who doesn’t have that specific skill or knowledge or experience may not fully know, know that. My fear is that people will not try to learn themselves and go beyond versus from what an AI gives you.

AI is just going to be a tool, it’s going to help your vision sometimes, but at the end of the day, you need to be able to become proficient in the specific subject matter that you’re trying the AI to help you with.

Sandi: So if I’m a physician, a surgeon perhaps, and I’m, I do knee surgeries day in and day out, and assistive technology is somehow going to guide me on how to do knee surgery. I still need to know what a proper knee surgery process is so that if AI tells me something that isn’t quite right, I know that it’s guiding me down the wrong path.

Juan: Correct.

Sandi: Is that what you’re saying?

Juan: That’s pretty much it. So the way I like to look at AI is going to be as tools that are going to improve your productivity.

So if I’m able to identify a specific issue or a process I can tell it in a few prompts to do it and execute it for me but I still have to go through that chain of different processes to understand whether or not what it created was the correct thing. So let’s say I’m creating content. I’m going to create a blog about accessibility.

I’m going to identify a specific problem because I know it’s a problem and I know the solution. I can give it the best information. I will have to, not only that I give the information, I have to give it context. I will have to give it examples. But rather than me writing the article, you will write it, you will write it for me in less than two, three minutes as opposed to me trying to write it will take me maybe three, four days, definitely I’m not an author. But once it gives you information, you’re like, oh, oh, this is good. But maybe a specific text in here is not, it’s not factual. And then you train it and you give it information and you tell it why it’s not factual.

Think of an AI as a little kid and it’s super smart, but it’s trying to learn. Every single time you tell it this is why it’s not factual, it’s going to learn from it and it’s going to provide better content nature. Now, one of the things I like to do if I am getting content that is not the correct one, I will straight up ask, can you please tell me why is that you’re not able to deliver my request?

And if so, if I need to change my prompt, please tell me how to do it? I’m telling you oftentimes is going to give you a great prompt that you can reutilize.

Sandi: When we’re thinking about AI, we need to think about it as a productivity tool, more than anything, is what it sounds like to me. Like, it’s going to make our work faster, more efficient, more effective, and certainly for people who, might be experiencing barriers, can’t do some of these things, it’s going to be a productivity tool for them as well, because those barriers are going to get removed.

Juan: Correct. So if we go beyond assistive technology and AI being an assistant. So right now, a regular interaction is going to be me going to ChatGPT, perhaps providing a persona, then prompts, providing some training, and then asking him for different requests so he can help me create content. The way I envision it to work is going to be that specific amount of work that can be very time consuming is going to be pre loaded.

So the AI is going to have that context. And then the user is going to provide better information as to how they like to do things or the work they need to do. The AI is going to learn from that and it’s going to help them execute things a lot faster. So let’s say you have somebody with cerebral palsy type of disability and they are going to be doing some work on a website that requires them to go through like 15 different clicks.

Now you can train the AI to say these are the things I need to do so that person is going to do it, so it’s going to do it a lot faster for them. And then when it comes to understanding the information, it can even simplify or provide better explanations as to why things need to be done in a specific way.

And it will help them execute it. That’s what I think the real power with AI is going to be. And then we’re not quite there yet. That’s actually, we have AI and we have something called agents. And I believe that’s going to be the next thing where we put AI with agents and they’re going to help us execute things on our behalf.

David: Sandi indicated that AI tools are really a productivity tool for us to gain more efficiency in the workplace. However, one of the big challenges we have is accessibility. And even many of our website auditing tools are not keyboard accessible or screen reader accessible. Many of the AI tools I’ve played with on the internet have very challenging keyboard interfaces.

So the question I have is, what impact is AI going to have on the employment abilities for people with disabilities?

Juan: That’s a very good question. And if I were to break down into a few answers, I completely agree. Any interface we create for an AI tool must be accessible to multiple types of disabilities.

Whether it’s going to be keyboard, whether it’s going to be states for screen readers. If you have a screen, if you use a screen magnification, then you’re not going to get alerts somewhere where you cannot see it because they’re out of the view. So we need to make sure that we provide that. When it comes to employment, once we get an interface, and this is something I’ve been thinking a lot, having a disability, I understand how difficult it is to try to find a job.

I’ve been lucky in my life that I was able to find meaningful employment, but I understand not everybody’s there. So what are some of the things that we need to do? I believe there’s two things that we need to try to accomplish. A, AI tools to be accessible. So we’re going to avoid those specific issues that you were explaining.

Second one is going to be multimodality, meaning the AI is going to be able to understand input, whether it’s going to be text, voice, or vision. Once then we have those things, we’re going to be able to communicate with the AI. So the real problem or the challenge is going to be people with disabilities understanding and knowing how to interact with the system.

I’ve been teaching AI, and in fact, I was teaching AI to a group of people with disabilities. And one of the questions was, one, why would I want to use ChatGPT versus Google, if Google, if ChatGPT doesn’t give me the correct information? And she’s completely right about it. But I said, you have to, you don’t have to think about it like that.

If you find information and you want to summarize, you give it to ChatGPT. So it’s going to help you understand. Now, if you really want to do search or research, you can use tools like Perplexity that is going to go search the internet, provide a summary. And then if you want to, you can go into the website, but you no longer have to go into the website because you can say, what are the five things that I need to know from here and how is this important to me?

So it’s going to be your first tool for research and then for understanding a specific topic. And by the way, ChatGPT also have search right now. I still like the Perplexity one a little bit better, but it is quite good.

So understanding how to communicate with AI, the prompts and how it can help you complete tasks is going to be super important for anybody with a disability who wants to find meaningful employment.

David: For a young person living with vision loss and just starting out in a career, what would you advise them to focus on in developing skills and understanding tools that would help them?

Juan: Learn how to prompt and interact with different AI tools. And then from there, you’re going to be able to open a big door for multiple type of jobs that you may be able to do. One thing I like to do, I try and not always do it, but I try to listen to an audio book a week. Whether it’s going to be for business, for sales, how to better my company.

But at the same time, there’s so much information I don’t retain. But what I like to do on a Sunday, I just go to either ChatGPT or Claude and say, and I tell it, I was reading this book. Can we go chapter by chapter in terms of some of the things that I should have learned from it? And if there are any specific exercises, please help me identify them.

And then we can go through it, right? And at the end of the day, it’s okay. Can you please provide, provide a summary for me? And can you help me create an action plan? That can be applied to anything. So if somebody wants to do any kind of data analysis job, they may not know how to do it. They can actually start utilizing AI to source material and what’s written for that. Help them understand how to use it and then help them implement it within their work.

Sandi: It’s like a personal assistant then.

Juan: Correct. Correct. I mean, and the only thing is that we’re not quite there at the personal assistant portion yet. I believe we’re going to. In fact, you know, in my company, that’s some of the things we want to do.

So our interface right now, it helps us create tickets. We have an AI model that is going to allow us to ask questions. We can create basic actions. For instance, I can tell it, can you please create a ticket for me? And it’s going to create all the information. However, I believe the power is going to come where you have agents.

In fact, you’re going to have multiple agents. The way it’s going to work, there’s something called orchestrator. So it’s going to be a main agent and it’s going to take the input from the user and then it’s going to say, hey, maybe this user wants to know about cars and the other one, the other user one or about cars or about planes.

So what I’m going to do is I’m going to ask another AI that specializes only in cars, that specific question. And then once that it gives it back to me, I’m going to preview it and give it back to the user so that orchestrator will have access to multiple agents that can talk to each other. And again, I help you complete something a lot faster.

Sandi: Is there anybody out there creating standards around AI and AI accessibility? Or is it just kind of like the wild, wild west out there?

Juan: So, it is kind of the wild, wild west. I know there are some companies here in Canada to create standards for AI. I know in the United States, in Europe, they’ve tried to talk about it.

The only problem I see is that AI changes so fast. We’re not talking about years, we’re talking about weeks. So whatever was good this week may not apply next week. That’s the real challenge we’re going to have. Now, one of the things I would like to, to underline or try to help people or make people think is how can we actually use AI for good and really trying to think of the user when it comes to developing a specific application.

That would be the only thing that I can think of in terms of standards.

Sandi: So do people need to be concerned about security if they’re using AI?

Juan: Definitely. Security is super important. It really depends on how you’re going to be able to interact with the system. One thing, AI really thrives on data. So the more you train it, the more it’s going to know.

But the real concern is going to be in terms of how much personal information you’re going to give it. So if the user provide information in terms of assistive technologies, where I live, everything like that, you have to be careful in terms of how the data is going to be used. Now, I believe in the future you’re going to have models that are going to be running from your computer that you can train and then you can try to mitigate some of that risk.

But at the end of the day, that could also get expensive. You may have to have a better computer to run it. And therefore I’m just going to go to a, to a provider like ChatGPT or a Copilot and it’s going to help me do that. But always be very, very careful. And you can also go on the other way where like people are trying to deceive, could be like users from a bank or anything like that utilizing AI.

So I always encourage people to be very careful, especially when it comes to data.

David: When you think about security, the level of trust is one of the major factors involved in that. So how do we know that we’re talking to the real Juan and not an AI persona?

Juan: I think right now you may be able to, to, to kind of like understand what it is.

I can tell you a few weeks ago, I received a call. It looked a little bit suspicious to me. It was still a survey, right? I said to the person, I think you’re an AI. And then it came back to me like, no, I can assure you I’m not an AI, but then I was able to trick it. And then it repeated questions to me one after the other one, like, oh, I know it’s an AI, but right now they’re getting quite good and you really have to be careful.

So if you get anybody calling you and say, oh, please give me your social information, just be very careful. Remember not to do it and then have some common sense.

David: So, so we’re talking to the real Juan.

Juan: Yeah, yeah, yeah.

Sandi: Oh, that’s good to know.

Juan: I, I tell you what, we have actually been playing with models to copy my voice as well as video and they’re quite good.

They’re not, not, not yet, not there yet, but they’re quite good.

Sandi: That’s so scary though. I’m not quite ready for that part of the AI stuff.

Juan: I know, it’s changing so fast.

David: Juan, I want to thank you so much for joining us and giving us some insights into the challenges and benefits of the artificial intelligence when it comes to accessibility.

How can people reach you if they want to learn more about what your company is doing and maybe some other resources they can follow up on?

Juan: Definitely. So they can always search me on LinkedIn, Juan Olarte. I’m always more than happy to provide information to people. If they want to go to our website, it’s digita11yaccessible.Io and the digitally is with an A11y. The two, two ones, and then we’re here to, to help.

One of the things we try to do is always provide information. Recently, we just started something called the Access Innovators Network. And the, and the main purpose of it is to teach entrepreneurs and people with disabilities, how to use AI so they can enable themselves.

Sandi: So <span style=”text-decoration-line:underline;text-decoration-style:dotted;color:rgba(0, 0, 0, 0.5)”>Juan,</span> one last question for you. Do you have one piece of practical advice you can leave our listeners with around AI and accessibility?

Juan: For sure, try to educate yourself in terms of what AI really means. Doesn’t mean you have to understand how LLMs models work, the whole technology, but mostly how can you interact with them.

Try to understand different models and ask it, have conversations with it. Can you tell me how accessibility can be utilized with AI? Trust me, it’s going to give you a lot of good information. But first and foremost, I will, I will say is always education. Learn about AI, whatever you can, it’s going to be great and it’s going to help you.

Sandi: Well, thank you so much Juan for joining us today. It has been a real treat and I’m sure we could chat about this for hours, but we’ll have to save that for another day.

Juan: Thank you, David and Sandi. I really appreciate both of you having me here.

David: The views, thoughts and opinions expressed on this podcast.

Are the speakers own and do not necessarily represent those of the podcast team and partners. This podcast is for information and learning purposes only. The Practical Accessibility Insights podcast is hosted by CMS Web Solutions. The intent of this podcast is to raise awareness for practical advice and strategies for making digital and physical environments more accessible to everyone.

Thank you for joining us in this time of discovery and sharing for a more inclusive society. If you have questions or comments, you can email us at info, I N F O, at david dot best or Sandi, S A N D I, at CMS Web Solutions dot com. For more information and resources, go to www.CMSWebSolutions.com or www.BestAccessibility.consulting.