WEBVTT
00:00:04.080 -->
00:01:09.480Welcome to the Technologies Impacting Society podcast, where we explore the profound ways that technology is shaping our world and transforming our lives. I'm your host Ina O' Murchu and in this episode, I got to explore the EU AI act with Yonah Welker through the lens of accessibility and assistive technologies, including high and unacceptable risk systems, emerging policies and regulation. Yonah's mission is to shape the future of algorithmic diversity, algorithms and policy, addressing human capacity ability, neurodiversity, and wellbeing such as social, emotional and human centered AI, emerging learning, work and city solutions. He has contributed to policy and ethic frameworks worldwide, and curated various exponential projects. Yonah has also spent over 60 World appearances to bring awareness to the neuro exclusion crisis that we have today. Okay, Yonah, you're very welcome to the Technologies Impacting Society podcast. Do you want to tell us a little bit about yourself before we kick off this podcast?
00:01:09.689 -->
00:03:06.719My name is Yonah, and my work is focused on AI robotics. In deep technology, addressing cognitive disability aspects of accessibility, I work on the three main levels on one hand as a technology. And at this level, on one hand, I serve as an evaluator for different programs, including projects funded by the European Commission, but also have my own startups and small portfolio of projects. Very soon, we plan to launch the acceleration program focused on technology addressing accessibility, I specifically interested in aspects of technology for autism, dyslexia, kind of cognitive differences in diversity. Second Level is policy activism. At some point, working with many, many projects and startups, I accumulated some kind of a personal demand to change something in safety, privacy. And I just felt that there is a gap between policymakers, technologists, and they exist in some kind of a silos. So I started to work on articles, policy suggestions, specifically addressing aspects of technology for disability or marginalized groups. And third level, is it bringing this work into the mainstream so I organized summits, creative projects create different exhibitions. And I as I've mentioned our last episode, I've created one of the world's biggest sermons for the good of humanity in Saudi Arabia, where we're in the concept of a museum was inspired by the topic of neurodiversity. So there are many things happen, but they're all connected to the idea and mission of a technology for humanity and cognitive disability in diversity.
00:03:07.289 -->
00:03:09.509Okay, well, it's a lot of a lot of contribution.
00:03:10.530 -->
00:03:14.789Yonah, takes up a lot of time.
00:03:10.530 -->
00:03:24.750While we're here today, we're going to we're going to chat on this episode about the EU AI Act. Can you fill the listeners in a little bit about this act, the history of this act?
00:03:26.340 -->
00:04:22.439Yes, so first of all, it's not the first big document and regulation proposed by the European Commission and European Union. So the tradition of such big documents was started in 2018 with a proposal of a GDPR General Data Protection Regulation. And since this moment, we had a lot of so called Brussels ethics. So Brussels ethic means that as Brussell or European Commission proposed some regulation or some policy, it push community to think more about ethics in other countries. So since proposal of GDPR, more companies technologies and ethics professionals started to think about data privacy, artificial intelligence Act is a next step.
00:04:22.620 -->
00:05:53.009It was initially proposed in 2020. And this document will be more focused not on just on data, but artificial intelligence and specifically aspects of a human centricity, transparency, safety, security, and how we identify different categories of risks and type of systems and how we assign different type of compliances and regulation dependent on the type of this systems. So it was initially proposed in April 2020 currently in just in the process of wrapping up. So it will be released very soon, policymakers still receive a lot of feedback from different communities, and is a very, very interesting area because there are a lot of good things because it brings more understanding in connection between different stakeholders, people who create technology, policymakers, different types of systems. So, EU act propose four levels of systems is unacceptable risk, high risk, limited risk or minimum risk. So it helps us to better understand it, but at the same time, there are some controversies. So because some groups of technologies or group of people or audiences are not described enough, and I think it will be one of the main topics of today's discussion.
00:05:53.879 -->
00:06:06.089And so how does this EU AI Act impact the development and the use of AI technologies for people with disabilities and neuro diverse individuals? Yonah?
00:06:06.810 -->
00:06:23.550Yes. So first of all, EU Act is a good example of high level policy documents and regulation, which propose categories and groups of as AI systems, and associated risks.
00:06:24.120 -->
00:07:25.470And essentially, the proposed four levels is I've mentioned first is an unacceptable type of risks. And it's typically relates to public scoring systems, or biometrics. It's specifically important for people with disabilities, because people with particular physical impairments or people without fingers, or without hands, who can't use these technologies, or these technologies can't work properly or identify such individuals properly. This category is actually covered this issue. The problem though they are not mentioned. So it cover the general population and this problem for general population but essentially also people with the disabilities. The second group has a high risk, high risk related to public services, transportation, infrastructure, hiring or educational platforms.
00:07:20.699 -->
00:08:09.930In answer again, though, disabled people are not directly mentioned in the document. This, all of these areas relate to this group. Why? Because until today, people with disabilities were discriminated at workplaces from 50 to 80% of people are not employed, and we have a really perpetuating bias. And if we're not able to fix it, it will not work properly with the same with the law enforcement. The same with the public and private services. For instance, bank system sometimes can properly identify your documents due to some neurological conditions or facial differences, different jester posture, gesticulation.
00:08:04.350 -->
00:09:08.009And this group basically cover this category of risks. The third group is more controversial specifically for disabled people. The third group called limited risk, and typically relate to the system where the main request from policymakers is just to provide the transparency. And this is the category relates to the emotion recognition, chatbots, and messaging platforms. So why is it more controversial? It's a more controversial because though it's not so dangerous or harmful for general population, this system sometimes can properly work for disabled people, and they bring more risk because people on the spectrum of autism spectrum or issues with the emotional expression or facial facial differences, they can be not correctly recognized by the pilot search systems.
00:09:04.230 -->
00:10:38.370Also, it's important to introduce the concept is on missions, or silos. So sometimes systems used as assistive technologies like is it sometimes a replacement of real life communication can create isolated ecosystem for people with disabilities who spent most of their time with chatbots or, or communications systems or social networks. So this risks are not covered enough. They're mostly focused on data input itself. And finally, is a minimal or no risk system. They relate to the platforms which not really use some complex AI and machine learning. So it's more about like us video games, using some personalisation mechanism and some internet signal alleges using AI. Here, there is no specific relation to disabilities. The key is a first three groups. And recently I've wrote a few articles, they will be published very soon with the updates and suggestions how this risk should be described differently for people with disabilities with physical ones, and cognitive disabilities, but also specifically how it should be addressed for law enforcement, for assistive technologies for police technologies, this system present very significant risks.
00:10:38.820 -->
00:10:50.490And it's not covered enough in it's not covered in the disability specific vocabulary scenarios cases currently is presented in a very general way.
00:10:50.789 -->
00:11:18.720When we were talking there just before we started, you know, there's been a good big contribution by the communities trying to bring these technologies to, you know, with a modicum of transparency and to assist individuals how can the developers and manufacturers of artificial intelligence technologies Yonah, ensure compliance with the EU AI Act, while still addressing the needs of the disability, and the neuro diversity communities?
00:11:19.919 -->
00:12:29.759So first of all, if we talk about the community involvement that we're at, we're not technologies that were mostly NGOs and nonprofits. So two years ago, the European Disability Forum, and after that the consortium of nonprofits and NGOs focus on accessibility and disability proposed the Open Letter addressing potential issues related to the Convention of Human Rights, both disabilities and how artificial intelligence act could embrace this specifics. So until today, when I say community, I must mean NGOs, in terms of technologists, researchers, I think they're still try to understand, like, what AI or AI Act, it really is, and how it should be implemented. Just right now is in the process of wrapping up, and the rest till controversial feedback from technology community, some people say, you just try to regulate everything, you try to limit our work. And in this way, you just stop the progress.
00:12:26.100 -->
00:12:48.690Other people say we need this regulation. Problem is people do not forget about the objective of the AI Act, because the objective of AI Act is not only to regulate, but facilitate and empower technology. So there are two actually elements behind it.
00:12:43.679 -->
00:13:46.320So when on one hand, it should actually support and create the dialogue between policymakers and technologies. And after that regulate, unfortunately, due to the some disproportions, we had too much public discussion on the about the problem of regulation limits, how policy can limit technology, it's not correct, actually, potentially with the proper additions, it can bring more clarity. For instance, we can add clarity, but legal statues, stakeholders, caregivers, different type of elements of correct vocabulary, terminology, knowledge frameworks, how we can recommend to adopt this technology, how we can introduce correct curriculum, how we can make sure we have a feedback loop. And it's critical both for physical disabilities or cognitive disabilities, or neurodiversity.
00:13:41.669 -->
00:14:28.110For instance, two years ago, I worked on this framework, focus on specific of technology for neurodiversity we had over around 20 criteria, like sensibility, color, visual differences, and basically our criteria. And we can bring this vision to this document. So policymakers, technologists will be better aligned. I think we just in the beginning, so there's still only grow in discussion in connection and I think as the document will be completely wrapped up and published and accessible for everyone who will be able to establish better connection.
00:14:28.350 -->
00:14:43.230How can the EU AI Act you want to help combat discrimination and bias against disabled and neurodiverse individuals in AI applications?
00:14:38.070 -->
00:14:43.230Is there a framework?
00:14:43.710 -->
00:15:37.350Unfortunately, naturally, so as I've mentioned, AI Act has a very high level vision of the risks of the systems. If you really like try to find it, you can find how it can be positive for people with disabilities. But actually actual descriptions do not reflect this. So that's why just recently, and it will be published very soon. I provide suggestions from our community about the bias, about how people are excluded about the exact cases about how it happens in terms of work technology and hiring technology in terms of a bank systems. In terms of police systems, it's not reflected now, currently is a very high level.
00:15:32.370 -->
00:17:05.849Okay, we have a public systems, we have a biometrics, but it's for a general population, people with the disabilities are not reflected currently. And this is just in the process of adding this layers. I believe, initially, unfortunately, people with disabilities were not included. Just a few years ago, there were two good documents, one from the International Red Cross, and another one from a United Nation experts focused on disability, which basically highlighted the fact that unfortunately, for many such documents, people with disabilities are not involved in the into the discussion, I mean, they are not involved. And unfortunately, just now we're trying to add this specific vocabulary cases, examples, differences of risks for people with cognitive disabilities. So for instance, so right now I work on another paper, where I reflect danger side of law enforcement cities, and how, for instance, autonomous system may be dangerous for people with a cognitive or mental impairments than for instance, I can recognize some particular visual or US audio or sound clues from the system, where I can react in particular way or my behavior pattern is different. So all of this is not reflected.
00:17:03.240 -->
00:17:08.430Unfortunately, it's a very specific things. And they're not reflected.
00:17:08.000 -->
00:17:19.220I just wanted to let you know that this episode was brought to you by Oriel's Magnesium Ionic Drops called magnesium, mineral balance.
00:17:15.289 -->
00:17:29.089Oriel's ionic magnesium drops help you to get that deep sleep and boost your energy and immune system. I've got a 10% off discount code called Ina10, that you can use at their checkout.
00:17:26.150 -->
00:17:29.930Go to the link in the podcast description below.
00:17:30.960 -->
00:18:23.069No, so I mean, the app basically doesn't go anywhere near promoting and protecting the rights of disabled and neurodiverse individuals at all. So there's a lot of discussion at the moment about bias in AI and regulation and the tech industry going yeah, will regulate it. The policy makers don't understand it will not the policymakers are saying they couldn't have the same depth of knowledge as the people themselves that develops the algorithms and the AI. But what you're saying this is a core, if we go this route, it's extremely dangerous. So what is kind of a solution? Because we need a solution, we need a positive solution. Because sure, we can focus on the dystopian, but that's easy. It's the challenge is looking for the positive outcomes. What is the positive outcome, say for the EU AI Act if we take take action now? In your opinion?
00:18:23.730 -->
00:18:53.099Yes. So first of all, as I've mentioned two years ago, 150 NGOs, I remember signed the document, focus on disability specific criteria, and EU AI Act So the good news that community organization, organizations, NGOs are very proactive, to improve this policy forever. And as I remember, there's already these updates related to biometrics.
00:18:53.279 -->
00:19:15.450So I think it will, it will happen gradually, because this document tried to reflect different things. For instance, there are problems related not only to people with disabilities, but immigrants, or people who have issues with a particular language, or they speak different language or they look differently, racial bias.
00:19:11.460 -->
00:20:21.180So there are a lot of parallel discussions. So policymakers, I think try to reflect them, it will take some time. It really requires some courage and united action from all of us, all of our stakeholders technology. So we're trying to I'm personally part of a few communities and they all try to prepare some documents, letters, send it to the policymakers. It relates to audit, it relates to risks it relates to relates to accessibility protocols. It also relates to, let's say, positions of different governments because some time ago, I've mentioned the Brussels ethics so just after the EU AI Act was released, the few other governments and other countries also released their updates for AI policy, including China, Japan, United States, also one of our communities work on the UK policy. So they try to connect to the vision of the different governments and how they could work together.
00:20:17.519 -->
00:20:42.900Because then you have more common ground and understanding of human rights or the safety of the dangerous side of technology about positive side of technology, some kind of impact matrix approaches to measurement, it's easier to come up with a kind of a common framework. So I think it will take time, but I believe we have a positive steps forward.
00:20:43.230 -->
00:20:49.200Baby steps, but because the data knows no borders, it's not contained to one government these effects.
00:20:49.619 -->
00:21:30.779This is essentially data, every different type of individual needs to be catered for. And that's, that's a challenge in its own right. So do we have time that's the thing, we have to make the time to make sure that this you know, everybody's covered in the act. You've spoken about the NGOs, but you seem to be kind of a unique individual. They're doing stuff on your own on your own. And there many other individuals like yourself out there, Yonah, you know, similar evangelical work, because it's a whole. It's a commitment. It's a purpose, it's purpose driven, you know, have you come across many other individuals like yourself, or do you find yourself in a unique space?
00:21:31.380 -->
00:22:29.849It's an interesting question. So first of all, there are a lot of amazing evangelists for neurodiversity, for disability, for ethics. So for instance, recently, we had a joint session, with a person with a mod steer net, she's focused on AI for children and children rights, which is amazing expert, there are such type of experts around the world, there are so many because it's a very challenging, it's really require a lot of commitment, and kind of intersectional knowledge because it's about kind of a law, human rights, social studies, but also technology design, or product development, sometimes even knowledge of educational systems or children's support. Also disability and accessibility, and also sometimes some kind of real real life experiences, which you're able to bring to the table. So yes, there are such people in we are cooperate.
00:22:25.380 -->
00:23:13.650So as I've mentioned, recently, we started working on the startup accelerator in the United States, and another person who will be my partner, also such type of evangelist, and she's actively working on technologies. Also she write books, and she's focused on pitch impairments. So I would say there are such people, it's still challenging, because one of the issues we face first is a data silos. So big companies like a Deep Mind, or open AI, they have a lot of computational power, they create this kind of a monopolies and they have better access for better research, work creation, more scalable products or solutions.
00:23:13.920 -->
00:24:32.910Also, they take a lot of attention of media, it's very difficult actually to promote some new documents, frameworks or policies, because literally the same type of people from technology world, just like take the majority of coverage. So advocates or patients or people who present our communities is just not heard enough or do not take enough attention. But it's really questioned to media because unfortunately, people are not really would love to hear about like real life stories or work. They interested kind of hype type of stories like AI will kill future humanity, or how it's destroyed or workplaces that are not interested in this highly specific conversation about approaches to AI and educational segment. Building curriculum for flip room or flipped classroom approach, like more specific real life cases is boring. It's not like a it's not so clear for the audience. They're interested in this hype situation, as hype cases like how we create placement for a human being AGI supercomputers and so on.
00:24:32.940 -->
00:24:43.049Sensation really, you know, that it's the sensational stuff that gets the press and it gets on the front cover. Yes, yeah. Not not the day to day use cases of real life.
00:24:43.380 -->
00:26:26.250Actually, it creates a lot of trouble because it's really, for instance, in our area, where we constantly looking for investors or partners, people just are not informed correctly, because there's a too much buzz in this bar. is very far from actual work. So for instance, you can find policymakers who never deal with actual technologies and never actually interview any startups in the field of work of this policymakers, or you have actors or writers in the field of AI who never talked to actual startup. So, I mean, they prefer to be focused on very high level, sky fight of conversation the same time, our problems are very specific, very concrete, as the real lives of real people who struggle who are excluded by workplaces, by educational system, due to the social issues, and we're just trying to improve social standards, social consciousness, and it helps us to improve technologies because algorithms do not create biases, society create them, people create them, we have issues with the lack of diversity, lack of access, lack of empathy, lack of social studies behind the technology team. So it's a more like, organizational, social, and cultural problems, which we should discuss in which in which we should implement is a part of educational curriculums, or preparation of teams in the field of our work
00:26:26.279 -->
00:27:31.529Slightly, I think, you know, you hit the nail on the head there, you know, it is it's a societal problem, it's a societal issue, but it's come so fast. Yonah it's like, you know, almost overnight AI is here, it's we've all known it's been coming. And people, there's no long term strategy, all of a sudden, it's like, all this stuff has to be regulated. And there are new jobs coming in, they will be in this fit in these fields. This is the area where the jobs are going to grow out of, but there aren't enough people there in place, I see it, I see myself with the with the industry. But what happens when you have the shift, everybody's an AI expert overnight, and the you know, the chumps from one thing to another, and so on and so forth, and doesn't actually help communities that are working towards goals in the area, you know, also area of neurodiversity and education itself, they would have had, they would always have had a marketing issue. So I'm doing this podcast we can get, we can try to get some of it out there a bit. And that helps. How would you see the next steps with the EU AI? What is going to happen?
00:27:27.390 -->
00:27:32.970With the whole area of neuro diverse? Where do we go from here?
00:27:33.180 -->
00:27:54.480So first of all, with all these limitations, or challenges, or a controversial discussion, I believe the EU AI Act have has a very positive impact in terms of starting this discussion and setting the standards, because initially, AI regulation was a part of GDPR.
00:27:50.250 -->
00:28:52.470And just partially mentioned, AI system is a part of a Data Universal. So now we have a separate document focused on AI is a good start, as I've mentioned, is automatically create the wave across other countries such as Japan, China, US, UK, we had the Italy who tried to bend, charge GDP. After that they changed their mind. It also pushed governments to think about digital serenity. So they started to think okay, maybe existing companies building some AI systems may violate this law, but maybe we could create our own systems aligned with our own standards. So I believe it also push the development of local systems, plus some local regulations specified towards particular country or a particular area or particular type of economy. Because when we talk about AI is not universal.
00:28:49.349 -->
00:29:21.809So for instance, when we're talking about neurodiversity, where accessibility many of our peers come in from Norway to countries like Sweden, like Denmark, or Norway, or Finland, or UK or Canada, at the same time, if we talk about Switzerland or the Middle East AI is more part of a smart cities are a financial or banking segment. So the utilisation of AI is a bit different. The charger challenges is a bit different.
00:29:21.990 -->
00:31:19.200So I believe EU AI Act is a good start for setting the standards of understanding of the risks is the first thing. Second thing is very important that they underline this unacceptable type of risk, where they put the biometrics the scoring and the public system. So the policymakers sent the signal that we can't automate public spaces. So public spaces will always will be kind of with some kind of a human involvement. We can't make everything automatic with a fingerprint. Use of a facial recognition, we can't use it. Because if we have a bias, it can be dangerous. It can be exclusive, and specifically in terms of a law enforcement in terms of police systems, and specifically in terms of the military systems can be very dangerous. And third, we have this further categories for a hiring for educational systems, for public and private services, or for it for a chatbot for kind of interactive and Creative Technologies. So we create this heirarchy of understanding of what we have, what type of risks this technology put towards broader society, the next step is to better understand different type of categories of people and how these people are different. For instance, people with disabilities, people on the spectrum neurodivergent immigrants, people presenting my ethnical minorities, so we try to understand so some specific subsets of this categories, associated risks, impact assessments, and more specific scenarios towards them. And that's what I'm trying to bring in my work for cognitive disabilities and disabilities in general.
00:31:19.470 -->
00:31:42.150Yes. And your it's great that that it's happening, and that it's been taken to task. And it's been worked on, because it's absolutely, it's, it's incredibly important that this, you know, segment of society does not get left behind, especially as we move more and more into the digital age and into a digital economy.
00:31:37.950 -->
00:31:52.349So, you know, you've mentioned during that what you mentioned, kind of the plus sides where the downsides of the technology and the misinformation side of it, we need to talk a little bit about the bad side of it also.
00:31:52.680 -->
00:31:53.160Yes.
00:31:53.250 -->
00:31:56.490What in your opinion, is, are the dangers, I'd say,
00:31:56.549 -->
00:34:49.860Yes, if we talk about disabled people, is above the problem that they're constantly live in some kind of isolated ecosystems, they work remotely, they may be cooperate with colleagues, with a family with a caregiver. And if we put them into the scenario of a constant interaction with a robot or a platform, can be very dangerous. If this ecosystem is driven by not genuine or not correct. Scenario, just imagine that your main teacher, it's something like ChatGPT. And it's based on not genuine information about society in reality, so this system teaches you run, maybe about some historical facts, or some even angles of understanding. For instance, now we have a lot of discussions, who train this systems, what their political view, and how to make it natural. Because every time we create the system, we put ourselves into the system, our views, our visions, our angles, because even if we talk about the history, go to a different countries, and you will see the same historical fact, is seen differently in different countries dependent, what's your site for? For instance, if you were affected by colonisation, in Africa, opinion of Africans about colonisation is a very clear negative few comments to some people in UK they could say, you know, economically, it was actually quite beneficial, you know, so and every type of information can be driven in very different angles. And what's why it's so systems if we create our content based platforms or educational solutions, they should be as natural as possible. Avoiding this type of biases, influences discriminative elements and type of political influence. And as a by the way, also, a big question. Behind the the European Commission AI Act is a question of a propaganda of a disinformation of the misinformation and using this powerful AGI systems to create massive amount of content or data or informational campaigns in order to influence particular group of people. And statistically individuals with disabilities. They're not only more frequently with terms of a violent crimes, harassment, abuse, social attacks, but also misinformation disinformation.
00:34:50.159 -->
00:35:12.269manipulations is very important to track not only actions, but also how these ecosystems work for People with disabilities creating really unnatural, accessible, emotionally physically, mentally human centered approach to them.
00:35:12.300 -->
00:35:50.969Yeah, no, absolutely. And then all of the things that we consider this the danger of centralised technologies, and as you said, Who, who, who manages? And how do we keep that neutral? That's, you know, will AI do that? In the end, you know, it's it is a but it is, it is a danger with centralized technologies. So we do that's another another thing we have to as a as humanity needs to take control take charge of like how we keep these, how we keep everything neutral. You know, it's quite the challenge. Especially right now, it's quite the challenge everywhere,
00:35:51.389 -->
00:36:29.250Would love to wish everyone who will listen to this podcast, keep in touch with me with our communities, we continue to publish a lot of papers, works, articles, which will bring more specific information, statistics, references, approaches to development research, our data, our safety, security, because even though such documents is EU AI Act is a very good examples of a policies we still need, should be very proactive as citizens, creators, technologies, and it's our job.
00:36:25.289 -->
00:36:36.239So kindly keep in touch with us and for sure. Be a part of this movement, create your own projects, and let's stay connected.
00:36:37.019 -->
00:36:43.710Great, great stuff. I'll include your links as well, in the episode below.
00:36:39.750 -->
00:36:53.639Okay, you want to thank you very much for your time. Thanks for tuning in. And don't forget to subscribe to the podcast and Spotify, Apple podcasts or your favorite podcast platform.
00:36:50.460 -->
00:36:58.409Gather let's uncover the transformative power of technology and its profound impact on society.
