Accessible AI Is Better Business, with Rachael Mole - Moleworks [084]

Narrator: The Business of AI Podcast, exploring how businesses are using AI to build services and tools, transforming organizations and delighting consumers. Produced by UKAI and hosted by Tim Flagg, AI is our business.

Tim Flagg: Welcome to another episode of the Business of AI podcast and I'm delighted to welcome Rachel Mole from moleworks today. Rachel, welcome to the podcast.

Rachel Mole: Thank you so much for having me.

Tim Flagg: It's great to have you here. So let's start off by finding out a little bit about you. Could you tell us about your background and how you ended up doing what you're doing today?

Rachel Mole: So I'm Rachel Mole. I describe myself now as an independent AI ethicist. Um, mainly because I focus on the user experience of AI and how it um, is interacting with disabled people's lives. Fundamentally how uh, access, uh, matters when it comes to accessing society as a whole. And as we know, AI is now in pretty much every single facet of society. Um, So I run MoreWorks Solutions, uh, which is an operations management consultancy based in Yorkshire. But I also sit on several government advisory panels, uh, including DIP tac, the Disabled Persons Transport Advisory Committee, and the Automated Vehicle Expert Advisory Panel and Access Advisory Panel M. So though my independent work mainly focuses on what ethical AI looks like in practice for disabled people, um, I also have a personal passion for employment as a, ah, particular area of interest. Um, I'm a Churchill Fellow. So my research with that focused on how building work cultures for disabled people drives wider innovation in organizations. Uh, and I also published a white paper late last year which was the largest collection of research legislation and um, policy on reasonable adjustments of which AI and the future of adjustments was a hugely core theme. Um, and with that a, ah, question I keep coming back to is as AI completely reshapes how we work, um, who gets left behind and are we building systems that create new barriers while also uh, claiming to remove old ones?

Tim Flagg: Yeah, well, fascinating work. I'm looking forward to finding out a little bit more about all of those areas. Um, you mentioned that um, it kind of comes from the user, um, and thinking about the user's experience, I wonder whether we could start there. And um, it sounds like that's an area that many people will be familiar with. That concept of user design, uh, user experience. How does that then evolve into both thinking about, um, people with disabilities and what their needs are specifically? Um, and how are those two things related? Is that a relatively new field within ux?

Rachel Mole: Uh, new field? Certainly not. I think probably a field that has been taken seriously in just digital UX for many years. But I Think one that unless you know about it, it's not really something that people consider. And I think especially as digital skills have become like that, you no longer need an IT degree or, you know, to be in. Involved in the tech industry to now kind of own, um, whether it's like through product management or project management, um, part of the process of delivering a piece of technology. M. That's where we're seeing people really having no idea that even access should be something that they consider. Um, and usually it comes in too late. So it's tacked on at the end as, oh, no, we've got legislation and legal requirements we have to meet and it's then a very expensive retrofit at the end. Um, so I have lived experience disability. I'm disabled myself and, um, one quarter of the UK population is disabled. So when we're talking about user experience, uh, of disabled people, I think what we have to reframe it as is the user experience of tech users. Um, this isn't a small niche group of people. Um, and as we expand the disability kind of label into something called the social model, which describes disability rather than it being the person who's disabled, it's society that disables people. Uh, we can also start describing people as disabled or, um, them describing themselves as disabled people, uh, who are elderly, who are temporarily disabled with an injury, or even people who, um. If you're pushing a pram down a street or you have children screaming in the background of zoom calls, that is still how people can be disabled by the technology and the infrastructure that surrounds them. Um, so, yeah, I'm really keen to kind of move it out of a niche topic and into just user experience in general.

Tim Flagg: Yeah. And thank you. Because I think, um, that was a really good place to start because I admit I came in with a little bit of a sort of conception there that it was this sort of new thing, but I think frankly that shows my ignorance, uh, in terms of understanding where that fits in. So thank you for explaining that. Um, I think what. What was really interesting there was that you referred to that as being something that is in society more broadly. Um, and I think that's probably true that there is a lack of awareness, there is a lack of understanding of this. But then from a commercial perspective, you also link that to the fact that if companies aren't thinking about this at the beginning, then it actually has a bigger, uh, impact, um, and they need to fix things at the end, which has a commercial, A bigger commercial impact than had they actually thought about it at the beginning, not to mention the impact it has on their brand. Um, but one of the things I wanted to ask you on that though was around how these different disabilities um, vary. Because you mentioned one in four people have some form of disability. Now I'd love to sort of understand the variety because I'd imagine some are um, less and some are more on whatever dimension. And so when you're thinking about how to accommodate those different disabilities, there's not really one size fits all. So how should companies be thinking about that?

Rachel Mole: I think the concept of designing something that is universally accessible is a bit of a fallacy. Um, no one um, tool or product that is set um, will ever be able to achieve universal design purely because even for different disabilities the requirements and the needs overlap and will change um, based on what that individual needs. So when we're talking about disabilities it's everything from mobility, um, and bringing this back to tech. So this could be for example mobility. Somebody might be um, unable to use their hands or their fingers in a dexterous way. Uh, so typing on a keyboard or swiping on a touchscreen might be really difficult. Um, then we have um, cognitive. So that might be everything from dementia to something uh, like autism or ADHD or dyslexia, dyspraxia. So when we're thinking about tech, is the design intuitive? Is it very obvious of which buttons to press in which order to get the outcome that's desired? Um, then we have um, you know, uh, energy disabilities. So for that that's things like chronic fatigue and where um, you know, for me my eyes are. One of my conditions is uh, an energy disability. And my biggest bugbear when uh, doing something digital that requires me to sit down for half an hour and like a form and it doesn't save the progress. And yeah, I can't always guarantee that I'm going to be able to get to the end of that form if it's taking a lot like cognitively out of me and to then you know, finish it to the best of my ability. Um, and then we have visual disabilities, hearing disabilities, you know, both of those I think I'd hope would be pretty self evident of some of the barriers that people could face when um, using uh, technology. But uh, visual isn't just you know, uh, being uh, blind, visually impaired. It's also you know, color and color, uh, contrast which um, you know, I'm glad to say see appears to be getting a bit more attention. But you'd be really surprised as to how many websites have terrible color contrast and you just immediately something that is a very easy fix becomes um, impossible for somebody to navigate. Um, so you know, it also varies. I present my disability some days I um, get, I'm very slow cognitively, my energy is really poor. I'm in a lot of pain physically. So um, technology is often a lifeline on those days as well. Um, but if I'm facing off against uh, some software or an app or you know, even my um, inbox sometimes it's not an intuitive um, system and it just adds barriers to me being able to get work done. Um, so it's important to recognize that it's not fixed, that disability can fluctuate. No one condition presents in the same way in different people. So my condition affects me how it affects me and somebody else could have exactly the same diagnosis and they could have a very different experience themselves. Um, so it's a really kind of big conversation. Yes. But I think there's tools that are there to help us. So legislation like wcag, um, that kind of is a really great starting block that not many people um, follow.

Tim Flagg: Um, sorry, what was it called?

Rachel Mole: Wit wcag. So w a g. Um, and then there's been a recent one. So in the uh, EU Accessibility act as well, um, I can send over links to share along with the notes of this, um, for people to read up on that. And it's. So there are tools and parameters set out that kind of show a level of best practice. Um, and it's always helpful for people to follow that. But then also an understanding of that's the baseline, that's what we have to get. Right. Um, but there's so much we can do on top of that by listening to disabled people and considering us and our experiences in the development process.

Tim Flagg: Yeah, that's really interesting. You mentioned some of the uh, legislation and some of the codes of conduct which are out there. I know you were saying that's just the baseline, but could you help us understand what currently exists? Um, so if someone was creating ah, a website or will come onto AI tools in a minute. But um, what currently exists in terms of that legislation, Is it EU wide legislation? Is it UK legislation?

Rachel Mole: Yes, so it is. Um, interesting. So the, there is uh, the web Content Accessibility Guidelines. So that's wcag. Um, they are internationally recognized um, standards, uh, but they're not laws themselves. Ah, they serve as essentially a um, uh, compliance benchmark for global accessibility laws. So like the UK Equality act, um, and public sector body regulations. Um, so generally, um, they are updated and we recently had um, one, uh, had it updated so companies now are expected to meet WCAG 2.2. Um, and then there's different levels within that. So the highest level to achieve is now testing my memory here off the cuff. I think companies are expected to meet level aa um to avoid essentially like legal penalties. So a disabled person can reasonably expect a government website for example to uh, have that level because it's an indication of um, being able to access necessary needed information and that if they did not have access to that, that is then um, they're being discriminated against. Um, so it essentially ensures that websites are perceivable, they can be operated and they can be understood and they're robust. They are robust enough to um, for any user to be able to access the content. Um, so um, it's helpful to have that as a benchmark. Um and so the levels are AA and aaa. So there is a gold standard but um, that's rarely mandated as needing to be met.

Tim Flagg: Yeah, yeah, no, I've come across AAA before, um, had a great designer who whenever we're building websites and apps would always insist um, almost well literally by design from the beginning on ensuring everything was ah, aaa, uh in terms of how it was designed, which was great. Um, so I want to come on to AI a little bit now as well because um, that uh, the AI world, there are many different areas of it um from the foundation models um to the apps to the platforms, to companies who are just using it to transform uh their internal processes. But I imagine there are particular areas here which, which you're working on with some of your clients. Um, could you talk us through how AI is making the situation uh better or maybe making situation worse or maybe both.

Rachel Mole: So most of my work at the moment is on um, in that employment sector for AI rollouts. So thinking about um, you know we've given all of our team across our ah, across our company licenses to um, something like Copilot or um, we've. We've written policies that you know, have, have it all where we're ready to go but nobody's using it or not enough people are using it or we're getting requests through for people um, to use it as a reasonable adjustment or even the, the other side of that is we would like people to be able to use it as a reasonable adjustment because we see the potential of that. Um, but we don't know where to start. Um, so again this is the user experience of, of this technology is. I'm a really big believer that um as a reasonable adjustment. So in the UK that's a legal requirement under the Equality act, um for people who require um, an adjustment to do their work to the best of their ability but are hindered to do so because of a barrier in the workplace that is um, that their disability um, compounds. So um, it's, it could be anything from you know, something physical like an ergonomic chair to a piece of technology that you dictate to and it types up notes for you. Um, or it's a particular you know, project program that you put in information and um, finds um, you know, what takes a big chunk of text and it turns it into bullet points so it's easier to comprehend information. And where AI is really coming into its own I feel in this particular area is that it has the ability to potentially open the door to that universal design piece because it's not a static piece of technology in that it can learn from um, an individual and it can develop and um, train itself on um, the prompts and the. So in this respect we're talking about generative AIs and LLMs. Um, it can learn from an individual what that, what their need is. So um, it then becomes a very hyper personalized tool for somebody to use. Um, you know we are um, opportunity wise. You know, it's, it takes a lot right now. So reasonable adjustments, the process for that involves a lot of like repetitive asks, documentation, heavy work. Um, you know, even the simple act of like requesting a reasonable adjustment can be really tricky. It also requires somebody to disclose their disability. And having just uh, an LLM rolled out company wide kind of removes that initial stage of requiring that disclosure and trust because you're just giving that tool to everybody anyway. Um, so you know, done well I think anyway, AI could reduce emotional labor on disabled employees. It could reduce inconsistency across managers in terms of being given a reasonable adjustments often to the um, discretion of individual managers. Um, and it also creates you know, that process then becomes more um, more dignified and faster which then supports people in doing their work quicker and more effectively. Um, you know there's huge potential.

Tim Flagg: Yeah, I had never thought about that um, opportunity for a tool like an LLM which is being rolled out across a company to have that um, sort of democratizing empowering effect um for people with disabilities. But actually the way you've explained it makes sense that uh, you know, people wouldn't want to have to expose uh, themselves. But actually if there's a tool there which they can use, which can help that actually becomes really empowering. Um, and I love what you were saying about the sort of personalization because a number of the conversations we're having now with companies working in the agentix space um, is around how agents can be even more powerful. One of the things, one of the examples I often give is that you know, a lot of companies will have dashboards, um, and they try and break down huge amounts of data into spreadsheets and dashboards that summarize that data into charts or uh, diagrams or whatever it is so you can measure targets and KPIs and that kind of stuff. And that, that was kind of, that's been sort of like you know, something that lots of businesses have built in those dashboards. Actually not everyone's able to process that information and actually it's not a very human um, uh, it's not very natural process to read it like that. Actually the ability of agents now is that the agents can um, well one gather all that information much more efficiently but then they can bring it into a much more human um, ah, a human ah, digestible format that's personalized to you. So rather than having a dashboard, everyone has their own um, agent that can tell them the information they need. And not only can it tell the information they need in the way that they're able to process, you can then talk back to it or communicate back to it and it will then go and do the things that you want it to do. So it's, it's basically like acting like a human, isn't it? That's a human interface. And, and I wonder whether that is also starting, people starting to think about the opportunities for how you could use agents to empower people with disabilities as well. Yes.

Rachel Mole: However, I think that for many people practically, or rather many organizations practically on the ground, that is the dream of being able to get there. But frankly people just aren't using it in the workplace well enough. There is a massive training gap of ah, um, jumping from typing in your question into an LLM and getting a response back to where we are now kind of mainstream of building out prompts and knowing how to prompt correctly. The next jump of kind of training your own agent and all of the work that goes into that. We can't expect the layperson to be able to um, kind of match the speed of this progress while also doing their job and also um, you know, upskilling themselves on just digital skills in general. So there's definitely um, a m. Responsibility of employers to have um, AI training, you know, as an integral part of learning and development anyway. But then I think there's also a responsibility of the developers to consider that human side of interacting with this technology. And actually you know, as, as fun as it is to uh, be able to kind of build your own agent and um, and do that kind of side of that hyper personalization. Like I think there's, there's a gap there, there's something that's missing of you know, how do we bridge this technology so it's not just the tech, um minded people who are able to effectively use it and deploy it and um, have a continuous training and know what to look for and know uh, what to do. Um, if we're talking about um, this being an effective tool for an organization, there's an identified problem that I work at. Reasonable adjustments. Who's the audience for that? Disabled people. People who need even maybe don't class themselves as disabled but recognize they could do with a bit of support in a specific area and then actually what, what's their general skills level and um, and how do we either get them trained or get the technology to a point that it's understandable and it's you know, simpler and easier and more intuitive to use? Because if we don't do that, what we're going to find is that skills gap is just going to keep on getting bigger and bigger and we are going to leave people behind.

Tim Flagg: Interesting. And um, it's again you're pointing out some really interesting dimensions to some of the challenges that we've seen, but maybe we haven't looked at the particular dimensions that you're sharing now. You know we often talk around inclusion but actually from m. Looking at this from uh, the dimension of disability opens um, up some areas which might haven't, maybe haven't had that light shine shone upon them. Um, so training, I think in particular there's a lot to talk about training. Right. You know, so there's, there's many training courses out there. I'd love to understand how you think um, training can best support, uh, raising awareness of disability and reasonable adjustments for disability. Is it that we need to train the leaders of the organisations, the managers in the organisations or the everyday users? Those are the three levels that we typically look at when we're looking at skills. But is there one area that you think has more responsibility for understanding this or that? Does it need to be integrated into the entire company?

Rachel Mole: Gosh, I think there's even another dimension to that of training the developers on the impact of their technology. And um, you know, the AI trained On existing data, you know, we know it replicates existing bias and you know, that needs to be explicitly addressed at the design stage. Um, you know, if somebody with a disability is using a piece of technology and as part of their want and the desire to hyper personalise it, are sharing about their disability, what we can't then have is that technology to then make assumptions based on data about the capabilities of that person. Which, you know, there's a risk in that, um, we hold bias, it's a human thing to do and um, um, we have to recognize that the AI that we're developing, um, is going to hold that too, unless we are very aware of that and we have guardrails in place to um, identify that and um, train the AI to either recognize that in themselves and learn from that or, um, potentially even be more inquisitive. And when somebody does raise that they have a disability to ask more questions, to learn about their capabilities, not just make an assumption. Um, I think there's also a really big risk at um, that management level of an organization. The training for them I think needs to mainly focus on that as AI could potentially handle more of that admin side of inclusion. Um, that employers feel like they've then ticked the box for being inclusive without actually changing anything about the culture or understanding what that tool is doing, like truly doing for inclusion and why it matters or where it's potentially failing or where it could fail in the next certain amount of time. And what have they got in place to either flag that and stop it from happening or reduce any harm that could come out of that? Um, you know, I think it, I call it a bit like it's a compliance theater of, you know, it's just the appearance of inclusion, um, not actually any substance to it. Um, so for that management level there needs to be training on what do we mean by reasonable adjustments? How can this be rolled out? How can you still support people that this isn't just handing over the responsibility of support, it's a tool. It's nothing to replace anything, it's simply a tool to enable better support. Um, accessibility in AI isn't just about screen readers or alt text being written or text being condensed. It's also about whether the system, from design through to deployment, was built with disabled people's experiences in mind. Um, rather than that retrofit of um, where we then panic because harm has come out of it that nobody expected, apart from disabled people who are there shouting about it already. Um, and then, and then there's that training level of of the user and actually are they, do they feel empowered that when they log on in a morning, start work that they, they know they have a tool right there that is ready to help them and they know how to use it effectively? And I'd argue actually that for the majority that no, that isn't something that they feel empowered or equipped to use.

Tim Flagg: Yeah, um, well that's been really interesting there and I think you're right. There's many different ways that we should look at um, what it means to be inclusive here, um, in the training. So right from you mentioned the foundational models and the data which they're trained on. So that's a slightly different use of training but equally important and all of the data that goes into that um, at the moment is quite um, uh, there's a number of biases, many different levels on that. I think we might not be able to get into it in this discussion but we're looking at world models as well. Um, which is the next generation of model which is being built now. They have even different, uh, even more complex uh, um, analysis of the world. But maybe they're looking at that through a very biased lens of able bodied people. And so there's a whole, like I say in a whole other area we could look at there. But I think then we go on to looking within organizations at the skills that they need to have as managers. And whilst there are some tick boxes that you mentioned that people have to tick to show that they're being inclusive, it can't be a one and done thing. It has to be an ongoing process. Um, it has to be the start of being open and being inquisitive and wanting to understand this rather than just that uh, tick box exercise. Um, we got on to talk about inclusion. Now I wanted to understand from that, you know, what's your perspective of how we can change things? How can we make this more inclusive? Is it about getting the right people into the rooms and sort of, what sort of, what have you seen that's maybe um, been positive in that area?

Rachel Mole: Uh, so disabled people being in the room when AI tools are uh, designed. And um, by that I don't just mean you know this, it's a stereotype but you know the coder with autism, that's not representative of disability. So looking around the room and thinking ah, who's been involved in this process and is it representative? Um, co design is something that is getting more popular which is really encouraging to see. Um, but making sure that um, it's part of the Project process just as a standard to, um, have that level of input from the people who will ultimately be using this tool. Um, and also I think something that would really help and I think also empower employers themselves into making this part of their process. Is any employer who is adopting a new AI tool as part of their procurement process to be asking, you know, what was your, uh, training data? Uh, who tested it? What failure models for users with disabilities were used? Um, were they even used? Um, my, my research, both with my Churchill Fellowship and the White paper, showed that there is a massive gap in, um, legislation. Um, and developers and employers, I think, have a really big opportunity to um, really show what good looks like here. Ah, um, policy is always going to fall behind and lag behind. The rate of technology, uh, is exponential in terms of policy. I don't think we'll ever be able to quite catch up with, um, how it's being developed. There's a responsibility there, both for developers and for end users to, to use it responsibly and defining what that means and coming together, um, to talk about that and to um, you know, build, um, agreements for use so it is fair and it is equitable. Um, I think that would be my dream of being able to see, See something like that happen.

Tim Flagg: Yeah, I think that's a really good way of, uh. I was going to ask you, actually, what's the one thing you'd like to see happen? I think you've answered it there, which is people coming together. But it's also really, I think industry has a role here because we can hold the foundational models to account, um, because we are effectively their customers. So a lot of the companies, m, who are building the apps or the platforms or the services, they're paying the money to those foundational models. And as you were outlining, we should be saying to them, we expect, uh, to know that the data that you're providing to us has been checked for, um, biases or has been checked for these other, uh, considerations. And that you have done it in a way which is going to benefit everyone. And that's, I think, something which we're going to see more of as, uh, the industry becomes more ethical. Um, so I know you're quite involved in helping with that as well. Um, so just as we start to kind of wrap things up then I wanted to understand how can we, uh, stay in touch with you, follow your work? What things have you got coming, um, that you'd like us to know about?

Rachel Mole: Uh, I'm active on LinkedIn. Uh, please do, uh, connect, follow on there. I, um, post quite often, um, industry updates, news thoughts, things that are going on and uh, that's probably the best place to find me. Otherwise you can check, uh, out rachel molecule.com. uh, that's my AI ethics home. Um, or there's More Work Solutions which is my organization, my ops organization as well. Um, or you can get in contact with me, uh, directly at uh. Ah, hello, uh, @morework solutions.com and yeah, we can, yeah, I'm always up to, to chat and um, to explore this topic because, you know, I'm also very aware I only represent one. But my lived experience. I only have the views of what I, um, experience. And this is a continuous learning journey for myself and for all of us as well. So I'm always keen to connect with other, particularly like disabled people who are working in this space.

Tim Flagg: Great, thank you. And actually, just on that, I'm just curious, are there communities of people who are so coming together to kind of look at this within the AI space, how we can uh, raise awareness of disability, uh, and disability needs?

Rachel Mole: Yes. And, um, I'd really recommend checking out Zero Project. They have an AI, um, database of really good resources to use. And I'll send the link over to include with that as well. Um, it's a fantastic resource, one that I use very often. Um, and within that there's uh, their AI Advisory board. I've um, kind of key people to be looking out for who are working in the AI and access space.

Tim Flagg: Amazing. Well Rachel, thank you so much. I found it really interesting today. I've learned a lot. Um, and you know, I think you've helped me to fill in a lot of the blanks there. Um, some of the lack of information that I had, frankly around what disability is and the need to include that and have it more prominently within user design. Specifically as the AI industry itself is developing. I think we all need to be aware of the impact which the technology can have if we're not using it carefully is going to exclude people. But if we use it carefully, we can actually make sure we are including people. We are thinking about their needs, bringing them to that conversation. We've talked about the roles of employers. I think employers, uh, have a really important role to play here. Uh, and I know a lot of our members will be listening and taking on board what you're saying. And hopefully got. They've already got some programs in place. Uh, but then there's a broader industry. We really do need to be, um, pushing for greater, uh, consideration of this from the foundational models which we all rely upon and making sure that we have the right people in the room to help us understand this and to help to shape, um, the future technology. Um, so best of luck with everything you're working on and thank you so much for joining us today.

Narrator: Don't miss the next episode. Subscribe now. UKAI is the trade association for business businesses across the uk, tech and non tech, large and small. M AI is our business. Find out more at, uh, UKAI co UK.