HCAI 12 - Democratizing AI with Mark Coeckelbergh

Speaker A: Foreign.

Speaker B: Okay. Thank you, Mark Kickelberg, for agreeing to join us today. So you are professor of Philosophy of Technology and Media, the University of Vienna. And we had the pleasure, both me and Alan, to see your keynote yesterday because you're here in Gothenburg for WASPS conference on AI for Humanity and Society. And you talked about AI and democracy. But why don't you give us a little more fuller introduction, uh, to who you are?

Speaker A: Yeah, so there was my academic position. I also am, um, a guest professor for, uh, WASP HS in Uppsala University here in Sweden. And my specialty is thinking about technology, in particular robotics and AI. And currently I'm mainly interested in the political aspects of the technology.

Speaker C: Thinking of the keynote yesterday, when you were talking about democracy and AI, there was one thing that struck me roughly halfway. The definition of AI or what we talk about when we talk about AI. Because myself, I'm a computer scientist, you're a human computer interaction designer, you're a philosopher. And I was trying to figure out whether we really refer to the same thing when we talk about AI. Would you be able to give us some form of definition of what AI is for you?

Speaker A: It's definitely a problem that people don't agree about what AI is. Um, AI is also a hype. Um, AI is quite vague often. For me, AI is two kind of things. On the one hand, it's a technology, it's not just one thing, but there's algorithms, uh, models, but there's also data, there's infrastructure, programming languages, a lot of different aspects of this technology. So one could say it's a range of technologies, a system of technologies. But apart from that, it's also narratives about the technology, and those narratives isolate critically. For example, in the AI Futures project at Uppsala University, we look at AI design futures. So, um, for me, AI is the technology, but also the stories about it and the ways we think about it, the futures we have in mind for humanity. So there's a lot of aspects to AI, I think, and I think for interdisciplinary research on AI, it's important to first talk about.

Speaker B: Terry, uh, clarify M. You released the book AI Ethics in 2020, and since 2020, I think, you know, a lot of things has happened in AI, at least in the hyping of what that is. So has kind of your view on what AI is changed over that since 2020?

Speaker A: I think the general view hasn't changed. But of course, uh, since then we have, for example, development of generative AI. Uh, 22, especially as of sort of November 22nd. ChatGPT, big moment for AI I think, um, so what I do is currently look especially at those forms of AI and see what their effects could be.

Speaker C: This is the Human Centered AI podcast, right? And we try and look at it from different perspectives. We've so far had people from computing, from law, from governance, from all kinds of different disciplines, uh, where I think this is the first time we have a philosopher. And given that it is the Human Centered AI Podcast, I would like to ask you what for you, when we talk about human centered AI, what does that mean?

Speaker A: Yeah, I think it means first of all that we have to develop and use AI in ways that bring in also ethical values. So responsible use of AI and in particular trying to get those values into already during the development phase rather than after, uh, some years when we see the effects is the usual thing. What happens with technology is think about the Internet, think about social media, um, they just get rolled out over people and then we have the effects. And it's kind of um, it's still good to think about them, um, and to be critical about them, but in a way it's already too late. And so for me, human centered AI is kind of project developing that responsible as the second remark I think I definitely want to make about human centered is that we have to watch out that we don't delegate responsibility to the machine, to AI. So human centered for me also means that humans are responsible. That we recognize as AI is not an alien thing, it's not this kind of complete other Nord AI is made by humans, it's designed by humans. And so it means we are also responsible for it. Third, for me, human centered AI, ah, the reason why I don't use the term personally is that I'm critical of the human centeredness in the sense that I think not only humans matter ethically and politically. I think also non human entities such as non human animals matter. The environment can also have intrinsic value and not only instrumental for humans. For that reason I think it's good to think about uh, what could be a more green AI, what could be more sustainable AI, but also apart from its value for humans and sustaining humanity. We also have to think, I think about what does AI do to these non human beings. And so that's why I don't use the term uh, myself.

Speaker C: This makes me think of another term used roughly as much as human centered AI and that's the fairness, accountability, transparency and ethics. Sort of this movement around fate or fact and that it sounds like you would more be in line with the values of that than the human centered Aspects of AI?

Speaker A: Yeah, I mean, I think human centered is. If it's a term that can help to make the technology more ethical, more fair and so on, that's fine. But I think it's good to specify then what we mean by human centered. And then I think you get to this values and you get to do all these things. You just mentioned

Speaker B: you're not the first guest on the podcast who brings this up. Maybe we should actually change the name of the podcast to something broader, more than human AI or something like that. But going back to the sort of technology side, I'm wondering how much you could layer kind of AI in so many different ways. And we've talked about that a little bit before. But one way is to sort of, when it comes to this kind of responsible use of AI or how it can impact things, is it the use of AI that needs to be done responsibly, or is it the development of AI? Because you mentioned we have a chance to kind of affect already the development. But I think there's development in the applications of this technology and there's development of the technology. Do you think both are equally important or is there a difference?

Speaker A: Yeah, of course there's a difference in terms of, uh, what we have to look at, but I think both are equally important. And there. I would say that when it comes to the user, we shouldn't forget about the user, because often in discussions, the developers and the companies who developed are the only ones who are made responsible or even blamed for whatever goes wrong. But I think users are also responsible and should, uh, be responsible for what they do. So if a user, uh, tweaks a system to do all kind of ethically bad things, that's very clear. But even when it comes to things like misinformation and so on, one could say that it would be great if we had users that are more critical and know the limitations of the system and also then use the system in a way that is aware of that. That being said, uh, there are always users that are less educated. Users are more vulnerable also. Uh, so the responsibility of the designers and the companies who, um, sell these products or who bring them on the market is also very important. So I would say both are important. And to stress the one shouldn't be excused to forget about your work.

Speaker C: So this makes me think of your keynote yesterday. You were talking of AI, sort of like a medium where there's so much that could be done by targeting people, targeting users. So when talking about users also being responsible, would it then make Sense to sort of target users in different ways depending on how responsible they feel they are or the impact they can have on the environment, on this, on society and so on. Would that be an ethical use of AI then?

Speaker A: No, I think we have to there uh, look for the lowest common denominator. We have to I think make AI develop AI in a way that takes into account the people who are least educated, who are weakest, uh, in terms of capacities for autonomous reasoning, for critical distance from these technologies. At the same time we have to not take the situation for granted. So I think we should educate people, raise awareness about the technologies and in that way raise the threshold. Um, but we should I think aim for the lower threshold and protect people who might be easily misled, easily manipulated. Especially since this power of AI to do that, to help reach that manipulation. Because in the antics user it's uh, those who employ the algorithms employ the AI. I think who can be said to manipulate AI doesn't have intentions but AI is definitely used for this purpose. So I think we need to protect people against that and regulate in such a way that there is protection against.

Speaker B: How do you differentiate between like when we talk about the risks and uh, how it can influence this stuff. How would you kind of differentiate between AI as a technology and just social media as a technology? Because you could easily talk about kind of how social media can be used to influence at a big scale. But of course they're not the same things.

Speaker A: Um, that's right and that's a good question and I think it's for me I'm not necessarily interested in only talking about AI. I think there we have to, when it comes to targeting of users and manipulation, we definitely also have to talk about social media. So also in terms of responsibility there I think it's good to, to always look in what context is the technology used. And there we have the so called problem of many hands, that it's not just like one person, not even the CEO for example of a big tech company who is responsible. There are many people um, involved in the whole AI process. And especially if we then also take into account people who create these social media environments with all the algorithms and all the setup behind that. The people who have trained AI, of course, people who allow, enable the collection of data, um, all kind of things are done with this data. Ah, so there's this whole ecosystem of AI and I think for a full analysis um, of uh, ethical AI, human centered AI. Ah, so we need to take into account this whole ecosystem.

Speaker C: But I guess this would be similar to treating AI as any other technology. I mean, thinking of, say, the development of combustion engines. Right? You can develop them ethically, responsibly, or you can make sure they have high carbon emissions, et cetera. Right. So it's a. And there's not one single person involved in that. Right. So this would be an analogy to AI being like any other technology.

Speaker A: Then when it comes to responsibility, AI is very much like any other technology. There are this many hands. So the many hands burn is quite universal. Of course, it doesn't mean that these technologies are the same, but I think the combustion engine is also a good example to show people that it's not just this combustion engine. The combustion engine is part of car, for example. And cars have completely changed our societies and cultures. Uh, we live in certain ways, we work in certain ways. For example, there's this commuting, there's the, uh, entire geographies have changed as a result of the car. And similarly with AI, our world continues to change. So it's important to analyze and evaluate technology in a holistic way, taking into account this whole system and the influences on the whole system.

Speaker C: So, speaking of technologies. So AI being a, uh, technology, sort of like the car that changes society, and we've had a few of those technologies. I mean we had the industrial revolution, transportation, the postal system, these things that have, that have changed our society. How would you, if you were to speculate, how do you think we think about AI in 10, 15, 20, or even 50 years as a positive thing or just something that happened, or what are the values that we as humans will have with us?

Speaker A: Yeah, I think what tends to happen with every technologies and also with all the digital technologies that we developed is that first we have the hive and we have some wild scenarios. Then we have a period of adaptation. And once people are adapted to the technology, they see it as more normal part of life. Like now we don't speak about the Internet anymore because it is just part of our lives. Social media also have become like that more and more. And I think AI will also take this role. That doesn't mean that there are no problems anymore. So I think it's important to keep assessing these technologies, keep evaluating these technologies. And also we should. You, uh, know, the development of this technology is going so rapid. That might be that AI we talk about now, uh, won't be the AI in 10 years, 20 years. Right. If we already see that in 10 years, we now have a different AI than before. I think we should be ready to look at these changes and ready for another big change. I don't believe in things like singularity and that kind of scenarios, but nevertheless, it's true that the technology changes rapidly.

Speaker C: This makes me think of the early days of the Internet and how today it still is sort of the same technology. It still is a network of computers, but it's just so much different from what it was back in the 90s or even the early 2000s. Uh, it's evolved to how we want to use it. And I, I guess that's what AI will be as well.

Speaker A: Yep. For example, the Internet through the smartphone, I think also a lot of things changed. Um, like now we are always connected to the Internet and, uh, in similar ways. AI will be more integrated in our lives for sure, because these big companies are going to try to make more money with it and find new ways of, uh, selling this technology to us.

Speaker B: I think it's interesting, these sort of analogies. Internet is like electricity, right? And the internal combustion engine is something you put in a car, and it's the car we use. It's not the internal combustion engine. So here it's interesting, I think, uh, the vagueness in which we so far talk about AI makes it difficult to know whether you should talk about it as electricity or Internet or as, uh, an engine that you put in a car. Because you could ask, like, okay, so what is the car here? Maybe we can point to different things. Like, I think ChatGPT perhaps is a car, but we talk about more than chatgpt when we talk about AI.

Speaker A: We have to talk about the technology as related to these other things. And we can, in a more refined analysis, we can, uh, look at these different things separately. For example, with the car, we can look at the sustainability of the materials of the car, but we can also talk about traffic as a system and what, for example, autonomous cars will do in it. Right. But in this case, you have the kind of autopilot system, but that's related to sensors. So already technically, there's all kind of other things. But for example, the kind of analysis that I like is phenomenological. Uh, how we experience being in such a car, how we experience traffic as our kind of social situation and not just a technical situation. So, uh, I think it's interesting to do this analysis at different levels and in a way that's always aware that there are these connections.

Speaker B: I really want to understand your view on AI because you don't believe in the super intelligence. So what is it? What is the development AI? Where is it going? To get better or like, where are we going? Where is the development of AI in coin?

Speaker A: It's, uh, so hard to make predictions because we don't have an idea how it can be in five or 10 years. So I don't want to be put in a position where I have to make predictions. I can only in general say that it's likely that AI itself changes like it did with generative AI, or that, uh, it gets connected with another technology and then has again so many effects we're not used to. What this technology is going to be. I have no idea. But I think it's important as society to be ready for these changes instead of just afterwards commenting on them. Uh, I think the way to go is to have interdisciplinary research together with technical people, in my case, and for society to have sort of make connections between technical experts and people from other disciplines, but also policymakers, uh, and so on, but do so in a way that's more, that's permanent, uh, and institutional rather than now. Now we have this ad, uh, hoc discussions everywhere, but there's not really an institutional way of dealing with new technologies. And if we do that, I think maybe we cannot predict the next 10, 20 years, but we can predict the next years and we can try to anticipate the changes already, uh, before they happen, because we can directly then influence the development as a society. So I think if we do that, we can make a more refined analysis. Okay, what's going to happen the next years? And how can we as a society be ready for this and have a framework in terms of regulation, but also in terms of reflecting on the goals we want to promote with technology? I think what's now happening is often the technology is rolled out over us and the goals are defined by those who, uh, try to sell the technology, who will benefit from the technology. I think as a society we should think like, what are actually our goals? And then see how can AI, in this case the technology, contribute to those goals in order to take back control, take back some steering influence over the course of this technology. Uh, if we don't do that, then the developers and the people who have interest in these technologies, financial interests, they will decide for us what this technology is going to do. So I think for going back to my book why AI, uh, Undermines Democracy, I think it's important to make AI more democratic in that sense, to have different stakeholders of citizens also involved in discussions about where do we want to go. And I think that's a very different way than asking where is the technology? Going because some people presume that there is this deterministic path for technology that is going to some point and we humans cannot do something about it. So I think that Guess goes back to his point, like, no, it's designed by humans and we can take control, we can make decisions as a society, uh, not as individuals, but together we can do that.

Speaker C: This makes me again think of the early days of the Internet where there was a sort of a gold rush of the web. Right. People were just throwing money at ah, stuff and there were new services, new ideas, completely unregulated up until a point where it became regulated. And I see that we're sort of there with AI right now. It's the gold rush of generative AI. You do whatever you want more or less, because there's not a lot of regulation. I mean there is the AI act and the uh, Digital Service act, but they are still reasonably new and they're only sort of local for Europe. How do you think that regulations of these, so global regulations of these things will affect AI? Uh, once we have the global regulations?

Speaker A: Yeah, so it's a big if. Right. So if we were doing global regulation, I think we create a sort of level playing field for innovation in this area which ultimately first will have give lot of resistance, which ultimately stimulates innovation because then everyone knows the rules of the game and can do their thing with it. So I think yeah, we need global regulation. We don't need like maybe the kind of level of regulation that the EU does. Maybe we can just agree on a few rules. We can have like a sort of more minimal type of regulation globally and more regulation locally if that's what people long term support. But I think we need this global framework because I think the problems that they don't stop at borders, they go everywhere, technology goes everywhere. Uh, just by the nature of these digital technologies.

Speaker C: I'm thinking of uh, gdpr. So GDPR has been around for eight years now. And if I recall correctly, when the EU started with gdpr there were countries that weren't part of the EU that adopted it simply for the fact that so many of their users were behind the GDPR wall sort of. And I wonder if we'll start to see the same things about the AI act or other regulations.

Speaker A: Yeah, so GDPR has had a lot of influence in the rest of the world. People really looked at what Europe was doing with privacy regulation. I think that there will be a similar effect. Uh, not necessarily that people will copy the AI act, but they might copy the approach. For example, Risk based approach or they might be stimulated to also look at the ethical problems seriously. But of course there will always be the influence of the particular political culture or the particular political situation also in different countries. So I think we naturally will get some diversity in terms of what people do. For example, in the US it's more traditional to not have too much regulation and let companies do their thing. I expect that this will also happen, that there will be more regulation, but that's not what we have in Europe. So I think in the end every country has to make their decisions about this. But that being said, we need that global framework, some coordination with some minimal rules. So for to have the playing field.

Speaker B: So what level do you think regulation is most important? Your focus is on sort of the traffic. If you look at the car and not so much how to regulate engine. Perhaps then is that a fair analogy rather than find regulations for how traffic will unfold rather than like how much?

Speaker A: Uh, yeah, I think in general, in terms of general approach, I think what's. What the EU doesn't uh, do very well, I think is that it focuses too much on AI as a kind of thing, as a product and as technology. And I think it makes more sense to regulate the applications and the concrete form that it takes. In the context, of course it's much harder, but it makes more sense ethically because what a technology is in my view is what it does. And what it does depends on that relational context and holistic system. So I uh, think what's happening now is that certain kinds of AI are fixed as high risk, for example, or as low risk, some applications also. But there's very little flexibility in terms of context in terms of uh, also how these systems could evolve in relation to other things. In order to have that I think you need an approach that has different levels where you also regulate more. These larger systems. For example, where you think about AI in the context of healthcare, where you think about AI in military contexts and have revelations for these different areas. I think for me that makes more sense because then you can take into account the precise ways that AI, um, is connected to other things and will influence people.

Speaker C: So this will be sort of like regulating various industries. I'm thinking there's regulations for providers of healthcare, there's uh, regulations for providers of uh, protective gear. There's regulations for all kinds of different manufacturers. So this would be regulating AI from the different use cases that would be much uh, easier to navigate. Navigate what you can and cannot do.

Speaker A: Uh, yeah, yeah. I think when it comes to Minimal rules. You can have like very general rules about AI, but I think then, you know, you need. The more detailed stuff needs to happen in relation to those different sectors and different use cases.

Speaker B: It's refreshing to talk to a philosopher about this because you're coming from a different angle than I would normally do myself. So I'm like, when I look at AI technology, um, I'm thinking currently, if we talk about generative AI, in particular about large language models and ChatGPT and that sort of stuff, I think that we still don't have a vehicle. We're still kind of operating the internal combustion engine. It's like we take an engine and we put it here, let's see what we can do with that. And it can make a noise and it can cause emissions and all of that. And then. But what you're really interested in talking about, what are the risks of this when it's actually a vehicle around it and when that's moving around and causing society and we can experience the, the vehicle. So then I'm like, uh, do you have any ideas what those vehicles might be? Because I don't think they are designed yet. I think that we have a bunch of technology developers that still haven't really figured out what to design around. I'm kind of, from an hi perspective, um, I'm still looking for those vehicles. Where are those vehicles and where is all the work on those vehicles?

Speaker A: I agree with you. So what we do now is input AI in the old vehicles and see what they do. And of course they can improve some things and they can have some negative effects as well. But part of this quite unpredictable evolution is that we will find new vehicles, we will find new ways of using these new applications. And uh, those applications will be invented because we will make better use of the special thing that AI can do as compared to other technologies. But in order to find it out, I think it's important to have a close look at what are the technological developments now, where could they have to imagine, but not in a sense of just mere fantasy, but in the sense of making realistic scenarios for the next five, 10 years. What could be new use cases, what's on the table now already? What are people experimenting with? What are people thinking about? I think if we do that, we can sort of ethically bridge to that future rather than just letting everything go.

Speaker C: So with AI, we're still sort of fairly early in the process, right? We've come out of, uh, the last winter of AI now we've had the deep learning hype. We're in the generative AI hype. And now people are starting to think of the environmental effects of it and thinking of the other big breakthroughs in humanity's history. When we had Internet or the combustion engine or the industrial revolution, we just use that technology. But we never, at this point at least, we never looked at the ecological aspects of it. Now with AI, we are actually decades ahead in terms of that. Ah, because we're looking at it right now. How do you think that affects how we work with AI?

Speaker A: Yeah, that's a good point, I think. So we have the opportunity here to do that, um, now to already look at all these aspects. And that means that we, I think as humanity, instead of sort of blindly following the arrows of technology, we can now give direction. We can really decide our futures if we want to. But for that to happen, I think we need, uh, broader involvement of different perspective and not only the people in those specific companies who develop AI, we need to also ask artists, for example, we need to ask all kind of citizens on what they think, what kind of futures they think might happen, what kind of futures they want. Also in order to make sure that this time we can do it right. And with the ecological side of this, I think we, uh, have a huge opportunity today to make AI that on the one hand can contribute to dealing with climate change, but also to make it more sustainable and make it less harmful to climate. So I, uh, think that there's a sort of emerging awareness about that among developers already, but there isn't a broad discussion about that. So I think we need to talk more about AI and climate, AI environments in order to really make that happen. Uh, so I agree with you, there is the opportunity. But if we want to make sure that we do it right this time, we have to have the broader conversation also to have democratic legitimacy for it.

Speaker C: So, um, should that discussion start with governance or should it start with just brainstorming about what we can do? What would a, uh, reasonable approach for this deep. So from computing, there's some talk about measuring how much electricity we're using or how much water something is using. But this is just sort of stating facts. Yes, we use electricity and then nothing really happens after that.

Speaker A: Yeah, I think there are different things we can do. So at the level of the design of the systems, we can try to make them more economical, for example. Right. And less harmful to the environment. At the level of the organization, we can also try to organize this. Uh, I, uh, co authored a paper where we propose to gamify this to stimulate, um, designers to be more environmental when they design AI. And then at the level of society we need to have those new kind of procedures and institutions in place. I think that can contribute to the more sustainable and ecologically friendly design of AI where we can, you know, brainstorming is one thing, but we also need to have deliberation processes because we need to decide things. We need to decide what kind of level of pollution, what kind of level of carbon emissions is acceptable. It's partly scientific matter what's needed to fight climate change, but partly it's also a political matter. And so we need democratic processes that deal with these questions in political ways. Um, but with input of experts, experts who can for example, say like, you know, whatever you decide, we think that a minimum of these measures is needed. We say that you can't have a, you better don't have 10% things more than this percentage, uh, more than so many degrees. Therefore we think that these measures are best. Then a democratic process needs to play and um, respond to that.

Speaker C: So AI and democracy, is it AI and democracy, AI for democracy or democracy for AI?

Speaker A: Right. I think we need um, definitely, uh, all of these. We need AI for democracy, that helps democracy. We need to protect democracy against some process that are happening in terms of power distribution. I think now there's much too much power on the side of a few companies. So we need to do something about that, even just economically speaking. But yeah, then in the end I think we need to always think together, technology and society. So instead of thinking about social and political problems on the one hand, and technological problems on the other hand, we need to from the beginning, both in the development of technologies and in democratic processes, to ting them together, to decide them together. I think people should realize that politics is not only about economy and about so called people, things, human things. It's always also about nature, environment. It's also always about technology and about the way we relate to nature. I think if we manage to institutionalize this kotlin, both at the level of development and in society, in our democratic institutions, then we can have much uh, more responsible development of technologies, including AI.

Speaker C: Okay. All right, Mark, with those wise words, we would like to thank you for joining the podcast. Been a pleasure to talk to you. It was great to see your keynote yesterday. Thank you for joining.

Speaker A: Thank you. Mhm. It. Mhm. Sa. Sam.