Paul Murphy Demands Answers on AI and Disability Rights

Thanks a lot for the presentations and your work. Just to start with Disability Federation of Ireland. I was interested in what you say about the AI Act aligning with the UNCRPD. But my question is, is that actually happening in reality? Footnote 2 says, recital 96 of the EU AI Act stipulates deployers of high-risk AI systems, especially public bodies or private entities, providing public services, must conduct a fundamental rights impact assessment for using the AI system. But, like, AI is being used. Have those fundamental rights impact assessments taken place, or have they not taken place? Thank you, Deputy, for that. I think it's happening at such a fast pace, the development of AI, that for employers, for employees, for potential candidates to keep up with it, is hugely challenging. We've heard from colleagues who are appearing today about the regulation. We know from talking to colleagues within the disability community that they are experiencing negative impact around accessing employment because of AI. things like large data sets being used to exclude people who are more diverse in terms of their profile of education or employment and the intersectionality with gender as well and gaps in employment. So we ratified the UNCRPD in 2018. We are up in front of the UN committee shortly. We did our own alternative report around the implementation of the UNCRPD and there is gaps in implementation, most certainly. But Charlotte's going to talk to more of the technical aspect. Yeah, so, no, it's a good question. And as Ima said, I think things are moving so fast with AI. And although things are moving fast, Ireland, I think, also has an opportunity. It's kind of one of the first EU member states to have assigned its designated competent authorities. So, you know, we have an opportunity here in Ireland to kind of also get ahead on ensuring we're compliant in AI development with human rights. So I think in that sense, it really, if I'm not entirely sure in terms of your question on are they taking place, the fundamental rights impacts assessments, I know that it's the deployers who should be undertaking those and then after the fact then kind of assessing whether, you know, consulting with groups and so on to see if those impacts assessments are kind of robust enough. So I suppose if they aren't happening, it's a question of really raising awareness about these specific kind of protections within the AI Act that can kind of support the protection of people with disabilities and other minority groups. There are also the codes of conduct, which are, I think the commission is going to be bringing out a kind of template for these later on in the year. And these will be essential and it's Article 95 of the EUA Act and it really talks about the importance of consulting with marginalised groups. So I think it's a great question, is this happening? I think the fact that this today is happening as a hearing is fantastic because we need to make sure that the disabled voice is at the forefront. And, yeah, so I think it's kind of in terms of these, there are sort of systems in place that we can be looking to utilise. And then just going back to the DPOs, the Designated Competent Authorities, sorry, DCOs, DCAs, sorry. So it's important that they also are aware of the concerns with the disability community in terms of the risks, but also the opportunities and that they will consult closely with this community. Thanks a lot. Next, a question for the National Disability Authority. And it actually relates to some of what Eimear was saying there. I mean, they used to say about computers, kind of garbage in, garbage out. You can't expect, just because it's a computer, if you give it the wrong information, it's still going to come back. And with AI, I think it's like, you know, discrimination and oppression in, discrimination and oppression back out and actually probably amplified because that's kind of what it's doing. And you're referring to that about, you know, reflecting and perpetuating existing inequalities. Is that what Eimear was talking about there, about, you know, potential for people to be discriminated against in terms of job opportunities by AI? systems that just, you know, simplify things by saying, you don't fit our idealised model of what we want or whatever. Is this something that is happening? That people, disabled people are losing out on job opportunities because of AI effectively slashing through CVs and discriminating? Thank you, Deputy. Yes, I mean, we would have a concern that AI, because at the moment the algorithms are based on kind of the middle of the bell curve, that they are excluding anything that they would conceive of as not normal. So a story my colleagues are always telling me, which I find horrifying, is that in the examples of a self-driving car, it ran over a disabled road user every single time because it didn't see it as a road user. So there would be some of those concerns, particularly in the employment context. So I think some of the issues around this are improving data collection as a whole, so that the data sets that are fed into AI learning models are improved. And at the moment, I suppose disability data, or equality data, is not necessarily at the standards it should be. There is supposed to be an equality data strategy, a national data equality strategy coming out very soon. We hope we take some measures to address that, but it would be some of the, I suppose, measures that would be taken in that regard. And also for, I suppose, in the public sector side of things here, there are guidelines for good ethical use of AI in public service. And there's a useful kind of canvas board where people can assess the impact on rights holders of the deployment of AI in a particular context. I suppose our concern from our experience over the years would be that there's sometimes a hierarchy in terms of equality grounds, and often disability can be forgotten or come towards the end of that hierarchy. So you need the public sector to be aware of disability as an issue before they can even properly account for those assessments. So those would be just some of the concerns. Thanks very much. And just finally, sorry, a question for Tim from Voice of Vision Impairment. It's kind of concerning what you're saying about GDPR violations potentially happening, you know, people who are visually impaired using AI to, let's say, read a bank statement. Could you give an example of what that looks like? Like, what I can imagine is, let's say, someone maybe has a gambling problem, and that appears on their bank statement, you know what I mean? They're up there putting money onto a gambling site repeatedly. They could then, what, potentially be targeted with, like, ads for gambling sites, things like that, where it feeds over and, like, the information isn't separated. Obviously, if you're at home and you get a bank statement and you read it yourself, that doesn't go to Google or Facebook or an advertising company, whereas you're saying these, potentially this information is being used in that way. Yeah, I think it's just a potential risk here. I think the fact that you touched on there is that blind people and visually impaired people are, by definition, are more likely to be using AI to access sensitive material than if, for example, there was a data leak, you know, loads of people's records got released into the dark web or something like that, that, you know, a blind person would probably be proportionally more likely to be caught up in that, you know, their sense of information would be more likely to be caught up in it than someone who didn't need AI to access these types of sensitive material. Thanks a lot. Thank you. Okay. Thank you. Thank you. Thank you. Thank you.