Can Europe save the world from the AI apocalypse?
When it comes to the future of AI, we – like seemingly everyone else – have way more questions than answers. Could European legislation do anything to protect us from job losses and “rogue” models? Can the chatbots created on this continent compete with those being developed by the US and China? What would it mean to win (or lose) the “AI race”? And most importantly: how doomed are we?
We put those questions and more to Aleksandra Przegalińska, a Polish AI researcher and philosopher and a fellow at the Center for Labor and a Just Economy at Harvard Law School. Aleksandra is one of the most engaging and entertaining voices we know on this topic, and we walked away from this conversation feeling… dare we say… hopeful? We hope you’ll feel the same.
For most of a decade, small donations from listeners like you have made it possible for us to continue making this fully independent weekly podcast. You can help us ring in our 10-year anniversary next year – and support another decade of non-boring news – by joining us over on Patreon or by donating via our website.
Did you know we have a weekly newsletter, Good Week Bad Week? It often draws on the super fun themes (tech sovereignty, humanity’s demise, etc etc) you heard about in today’s podcast. Sign up here!
RESOURCES FOR THIS EPISODE:
Aleksandra on Instagram
“AI Optimism Globally Widespread Despite Uneven Use” – Gallup, 22 September 2026
Aleksandra on Microsoft–Gallup’s AI adoption survey
“Arthur Mensch, CEO of French start-up Mistral AI: 'AI is software. It can be controlled'” – Le Monde, 24 September 2026
“Data Centre Silence” – Lighthouse Reports, 30 September 2026
CO-HOSTS
Katy Lee and Dominic Kraemer
PRODUCERS
Morgan Childs and Wojciech Oleksiak
MIXING AND MASTERING
Wojciech Oleksiak
SENIOR PRODUCER
Katy Lee
MUSIC
Jim Barne, Mariska Martina and Blue Dot Sessions
INTERN
Inês Ruivo
-
KATY LEE:
This podcast has been produced in cooperation with Euronet Plus, the leading radio network for EU news.
DOMINIC KRAEMER:
But this podcast would not be possible without the incredibly generous support of our amazing listeners who donate a little bit each month at patreon.com forward slash Europeans podcast.
[MUSIC]
KATY LEE:
Welcome back to the Europeans, the podcast that tries to drag a comb through the tangled mess that is Europe, but ultimately decides to just give up and embrace just having messy hair. Does that work as an acceptable metaphor?
DOMINIC KRAEMER:
Honestly, I think it's a miracle when either of us are able to come up with like any new way to open this podcast nine years into making it. So yes, the messy hair metaphor is 100% accepted.
KATY LEE:
Okay, good.
DOMINIC KRAEMER:
You're just back from Brussels, I hear. Did you embrace your own messy hair for this live event you were presenting in Brussels this week?
KATY LEE:
This may be where that metaphor came from, because my hair just wouldn't tame itself for this live event that I was hosting. But yeah, I am indeed back from Brussels. I had a lovely time in our great administrative capital. I was moderating this live event, which featured last week's podcast guest, in fact, Paweł Zerka. And I think the organisers are actually planning on releasing the whole of that event as a podcast. So we can post that to our social media channels when it's out, which I think will be next week.
But yeah, it was interesting to be in the belly of the EU beast this week of all weeks. Because after doing the interview that you're about to hear, I really had the sense of like, ‘Oh wow, in this city, in these very buildings I'm walking past here in Brussels, people are making policies that could really determine our future, that are really shaping the world that we live in.’
DOMINIC KRAEMER:
I mean, I kind of thought you knew that already. We've been making podcasts about how that's true in all kinds of ways for many years now.
KATY LEE:
Obviously...
DOMINIC KRAEMER:
But maybe you're mentioning it because there was one specific policy area on your mind where the actions of the European Union right now do feel like particularly pivotal. Was that perhaps the various laws it is passing and has already passed and may pass in the future surrounding artificial intelligence?
KATY LEE:
You know, it might just be that, yeah. It's funny, I've really noticed our own tone changing quite a bit on this podcast in terms of how we've talked about AI over the last few years.
Like, it wasn't that long ago that we were doing, you know, funny experiments with getting a robot version of you to read out the happy ending segment. The mood has gotten darker, I think it's fair to say.
You know, we've had AI tools going rogue and launching hacking attacks, like this Hugging Face incident that gets mentioned in the conversation you're about to hear. We have more and more people within the AI industry itself warning that this technology is going to kill us all. And at the same time, we have leaders here in Europe warning that we are incredibly dependent on AI models and infrastructure from the US and elsewhere, and that we're getting left behind in what might be the race that determines the future of this planet. This summer, for example, we saw the US government briefly suspend the rest of the world's access to Anthropic's most advanced models. It's not hard to see what the economic consequences would be if something like that happened more permanently. So the upshot, I guess, is like this technology is terrifying, but Europeans also need more of it.
DOMINIC KRAEMER:
Yeah, and on the other hand, the EU also happens to have the AI Act, the most comprehensive law on AI passed anywhere in the world to date, which doesn't feel like nothing. But does that law go far enough? And can the laws that we pass here in Europe make the AI companies elsewhere in the world behave better?
KATY LEE:
There are many, many people trying to answer these questions, obviously. But earlier this summer, I had the privilege of meeting someone who I thought tried to answer them in a way that felt both unusually clear and actually kind of entertaining. And she's also someone that has one foot in Europe and the other foot in the United States, which is helpful when you're thinking about different ways of approaching AI and maybe regulating it. That person is the Polish philosopher and researcher Aleksandra Przegalińska, a fellow at Harvard, as well as a professor at Koźmiński University in Warsaw.
DOMINIC KRAEMER:
The conversation you're about to hear is actually a lot more hopeful than I thought it was going to be. I didn't think there was anything that can make me feel hopeful about AI. And to be honest, there still isn't much. But I did end this conversation feeling a bit less gloomy than when I went into it. And I wasn't expecting that.
KATY LEE:
Can we imagine an AI that actually works for us? Can we dream of a less terrible approach to AI made in Europe that could, whisper it, save us all? I don't know about that. But I do know that we had a great conversation with Aleksandra. So great, in fact, that we decided we wanted to dedicate the whole of this week's episode to it. So there's no good week, bad week this week. Although, don't worry, we'll be back to our normal format next week. But for now, here is our interview with Aleksandra, brought to you in collaboration with Euronet Plus.
[MUSIC]
KATY LEE:
Aleksandra, so great to have you with us here on The Europeans.
ALEKSANDRA PRZEGALIŃSKA:
I'm so happy to be here. Thank you so much, Katy, for the invitation.
KATY LEE:
I want to start from a super fun place, which is our collective anxiety.
DOMINIC KRAEMER:
Woohoo!
KATY LEE:
Yeah, I mean, I often feel like one of the biggest ways in which AI has affected our lives up until now is just by making us feel more anxious, you know, like, is it going to steal all of our jobs? Is it going to kill us? Like, we hear more and more people, experts like you, talking about the likelihood that AI will actually lead to the destruction of humanity. Perhaps the best example of this is the former Anthropic employee who said that the people building AI earnestly believed that it could kill us all by the end of the decade. One of his former colleagues put the risk at around 10%. So, let's start here with this anxiety.
What do you make of all of that? Are these risks overblown? Is it kind of marketing hype? Or are you really concerned that all of these technologies could lead to the destruction of the human race?
ALEKSANDRA PRZEGALIŃSKA:
Well, first of all, I think you're right. And what is also very interesting about anxiety surrounding AI is that it's increasing together with our familiarity, right? So we know this technology better. We use it more often than we did two or three years ago. And still, it doesn't help, right? It's the other way around. The more we use it, the more we are afraid of it, which is unlikely. And very often, you know, various researchers claim that when you get to know something, your fears are going to be mitigated by a positive experience. And I'm quite sure that in many cases, we have positive experiences with AI, maybe even prepping for this conversation. We've had positive experiences with AI, summarising things for us, helping us, you know, in preparing some briefs, etc.
So AI, I think, in many ways, is quite helpful in people's daily lives, and then still produces this ginormous fear. And I think, unfortunately, a lot of that has to do with the narrative around artificial intelligence, but also how kind of AI is currently governed.
This guy, Jacob, right, who left Anthropic, and I really think that his message comes from the place of sincerity. There are attacks right now that he's worked with a PR company that specialises in spreading AI doom, you know, but I honestly want to believe that he was concerned with what he saw.
Now, if you ask me, if I believe in his prediction, that there is a 10% probability of that happening, I don't buy into that. But I would still say current, limited, still nascent version of this technology is already dangerous. It's true, right? Not dangerous in the sense that it's going to kill us all, but dangerous in a way that it's spreading misinformation, can be a swarm that provides or like executes a very successful cyber attack on your banking system.
And that is a big concern. So even though I do not buy into that “full-doomer” narrative, I would say there are reasons to be concerned. But there are also ways to mitigate that. And that's something I would focus on.
But then the other thing is also that when you compare how people felt, for instance, about ChatGPT three years ago and how they feel about OpenAI and ChatGPT today, it's a huge difference. I think three years ago, many people were quite hopeful. They thought, “Oh, there is this super interesting company that wants to build something ambitious. Let's see what happens.” And now what they see is not only OpenAI, but also other tech giants being super close to power, essentially proposing almost like a new social contract, being incredibly powerful geopolitical players, setting the rules of the game, and also being, you know, in conflict with one another. You have companies that want more regulation, companies that want less regulation. It feels like we don't know anymore what to believe. So I think the way this technology is communicated, but also governed, and the way certain incidents of its misbehaviour are handled, that all translates to various fears that we have.
And regarding the job market, I think that's a separate story, because the job market essentially hasn't been that volatile. We still have a fairly good, high-quality job market after four years with generative artificial intelligence. But then here and there, we see signals that are concerning, like entry-level jobs. And these signals are not taken seriously by the tech giants and by various states.
And I think the fact that there's lack of response regarding risks, that's what we observe here in the U.S. You know, you have agents breaking into systems, hacking systems, etc. And then, on the other hand, you have the public administration that says no guardrails at all. I think this builds up a lot of fear. And I would even add, my observation is that the fears in the U.S., here where I live, are much stronger, and that artificial intelligence is incredibly polarising here in the United States. And that is also because there is no response from the state, and there is no way of addressing the fears that people have.
DOMINIC KRAEMER:
Talking about the differences between how people respond in the U.S. and Europe, I saw on your Instagram this morning, the survey you posted showing attitudes around the world towards AI. And I was actually particularly interested to see the differences between the way European countries are pretty highly nervous about AI, compared to developing countries and Asian countries, where people tend to be more excited about it there. Why do you think that is?
ALEKSANDRA PRZEGALIŃSKA:
Well, I think this chart is incredibly interesting. And this whole GALLUP poll is super interesting, because it proves exactly the point that I made at the very beginning, that the more exposed you are to this technology, the more afraid you are.
Us Europeans, we are quite exposed to artificial intelligence. So we don't use our own systems, sadly, maybe we'll get to it. Or we use very little of our own AI that is built in Europe. But we do use a lot of American tech, and we increasingly use more and more of Chinese models. And then that clearly indicates that our exposure to AI is very high.
But also, I think we have a very heated debate about this technology, and Europeans participate in that debate. So we watch very closely what's happening in the US, we watch very closely what's happening in China. And sort of we have our own set of attitudes and values. And I think in many ways, we feel that artificial intelligence goes against many of them.
I think there are many fears regarding technological mass surveillance, there are many fears concerning the job market that we've mentioned before. But I think what is also an unfortunate daily experience of many Europeans, also many Americans, probably many Chinese people as well, is the fact that we see our internet deteriorating because of AI. So, AI has become effectively that parasite, you know, that produces so many deepfakes, misinformation, etc. sloppy content that is of no good quality. And I think that is that part of daily experience with artificial intelligence that doesn't really, you know, produce any type of enthusiasm.
So I would expect negative reactions of Europeans, I think, you know, Europeans tend to be quite critical of various technologies, if they take away their privacy, if there is a possibility that they might peek into your life, and that they are showing you something that's not necessarily true or very confusing. So, I think in that way, we are quite used to that critical assessment of technologies. And I wouldn't expect any other image, right?
But what is interesting for me is how America looks in that chart, because I am quite sure that, you know, again, in 2022, we would see here in the US a lot of enthusiasm. And now what we see is this declining enthusiasm towards artificial intelligence, both among the Democrats and also Republicans, which is quite often not mentioned. So, it's actually a bipartisan issue, and a big one, right.
So, the only country that still remains extremely positive towards AI, and it has a huge exposure to it is obviously China. But here we could, you know, speak volumes about why that is, and how much sort of the political party that is ruling China is enforcing a certain attitude towards AI and promising the people, you know, growth with artificial intelligence, but also a safety net if something goes wrong. So, you know, that's a different way of handling this technology with high exposure. In many other cases, you can suspect that maybe AI is perceived as an engine of growth. But since in some countries you don't have such exposure to it, there's still a lot of enthusiasm towards it, because people don't feel those negative effects in their daily lives, and they're less afraid.
KATY LEE:
So alongside all of this anxiety, we're also seeing anxiety at the kind of institutional and government level about how we keep up in the AI race, as we might call it, and this idea of remaining competitive. Ursula von der Leyen recently outlined a series of initiatives that the European Commission is embarking on in order to remain competitive when it comes to AI. I mean, the truth is, we're probably not going to win this race, if we're running against China, and we're running against the US as well. Is Europe at least doing the best job in your view at regulating AI? Like, can Europe be the world's knight in shining armour when it comes to AI regulation?
ALEKSANDRA PRZEGALIŃSKA:
I would like Europe to do better in terms of investments. We can talk about the regulations, you know, in a second. But for me, what matters a lot is to just say that, you know, I do believe that it's very hard to even take part in this race. But the question is sort of, do you have to perceive it as a race? Maybe there is a race between China and the US. Maybe we don't have to participate in that specific race, because our chances are low. But nonetheless, what matters is that we have our plan B. And that plan B means our own infrastructure, our own frontier models, and a possibility of an opt out.
And I have to tell you that I think the situation with Anthropic that happened, I think, in June, it was very telling. So you have a possibility of not being able to use an American technology, even if you pay for it, because the administration claims it to be dangerous, right? And says, “oh, outside the United States, you cannot use the best model offered by Anthropic.” So, we are in a situation of high dependency. And I think even if we don't win, even if our models are not as robust as the American ones, I would still say it's worthwhile to make an effort and build something in Europe. And I think that to me is this biggest challenge when it comes to technological sovereignty, that we have to brace and build.
And I have been listening to Ursula von der Leyen's State of Union speech. And I think it was very interesting, because she actually touched upon our first topic. You started talking about the anxiety. And she said something interesting. She said, “let's not overreact with anxiety”, because still, we can, in a way, use this technology in a beneficial way. If we decide that we want to ban AI, we're not participating in any kind of race. But also, that means that the world is running away from us, right? And we're staying far behind. So it's okay to be critical towards AI and to be very rigid in the assessment of this technology. But we have to try and implement it in a way that makes sense for Europeans and not just turn our backs. So, I think this effort towards building our own infrastructure, our own models is speaking to that.
DOMINIC KRAEMER:
And is that happening? I mean, are there models being built in Europe that are going to be kind of good enough to replace US models if they get turned off here?
ALEKSANDRA PRZEGALIŃSKA:
We have a couple of examples that I think are fairly interesting. In my home country, Poland, we have two models being built currently. One is specifically for the public administration of Poland. And I think it's doing a fairly good job in its narrow scope.
The other one is not obviously as big as some other open source models, but it's trained purely on Polish language. And in some benchmarks, it's doing a really good job in Polish specifically, which, you know, can be an interesting solution in some types of usages where you have certain dialects, et cetera, where it's particularly important that the model speaks very good Polish, maybe in legal applications, et cetera. We have, well, that's not the European Union, but in Europe, we have a very interesting model called Apertus. It's an open weight, open source model built in Switzerland. And then obviously, I think the biggest effort that we've seen so far is the French model Mistral. And that's a model that is widely used by various French companies. I think it's the state policy to promote that model, which I think makes for a very interesting case. Obviously, none of these models are really comparable with the frontier models offered in the United States, but they are there. And if we pour, let's say, more investments into them, they can become something significant.
And I know of many other similar initiatives. My colleagues from Spain have been reporting that there's a model being built in Spanish and Catalan. There is an effort in Germany. There's many other efforts in various other countries. So, I think if we collect these efforts, we do have a good foundation for a future European AI that can help us in devising that plan B that we really need.
KATY LEE:
So that's the investment side. Maybe we're not doing as badly as I tend to think that we are. Let's go back to the regulation side. How well do you think we're doing? Are we doing a decent job? Are we doing the world's best job?
DOMINIC KRAEMER:
I think we're doing something important. I think that we have set certain standards that are relevant. And what is even more important is that even though US-based technological companies usually hate those standards, they still abide by them because Europe is a quality market for them. So in that way, we are doing a good job. And I would also call it a part of our technological sovereignty. So we set certain rules and we say, yes, you want to participate in our single market or one market, however you want to call it, great, but you have to abide by these rules. And so far, it has generally worked, right? We have the AI Act. We have some other important regulations.
And now the thing that for me is quite important here is that I'm pretty happy with the shape of the AI Act in the sense that I think it adds important rules regarding transparency, high risk versus low risk, use of AI, et cetera. But I am just worried that it's not going to stand the test of time because the challenges that we currently have with artificial intelligence are slightly different. So when you look at the most heated debates about what to do with artificial intelligence now, actually, it's not about transparency. This is fairly easy to solve. It's about controllability. It's about governance. And it's about a possibility of internal evaluation before you release your model so that you make it safe. So everything that's been happening in the summer, the incident at Hugging Face where you had those open AI swarms of agents trying to hack Hugging Face, then we've learned that they were trying to hack other systems as well, German DSeWiki site, recently also Australia. You have multiple different spaces that they've been trying to hack and they are based on a concept that is called recursive learning. So, you give them sort of a prompt zero, but then from that moment onwards, they sort of start learning by themselves and teaching other models, sort of other agents how to behave, right? So, their behaviour is in a way emergent and hard to predict, although that prompt zero matters.
So back to your question about the former employee of Anthropic who said he wanted to leave. To me, there's a lot of techno determinism in that way of presenting the case where you say, “Oh, technology has its own logic. It's developing in ways that we cannot control at all. It's going to kill us.” I'm more on the human side. So, for me, the fact that open AI decided that those primordial agents are going to be persistent and then those agents taught other agents how to be super persistent, omit all hurdles and get to the reward as soon as possible. You give those models certain identity and that is a human decision. Open AI could have decided to build these models differently and then maybe they wouldn't hack or attack, you know, other systems in the way they did. I think it really matters, right? So, there is human responsibility to it all.
KATY LEE:
Yeah. And it's really interesting because we asked our listeners to write in questions, potential questions for you. And we had one listener, Robert, who asked, should we be limiting AI by reserving, you know, for example, some data centre capacity for questions that really matter, like treating diseases, this kind of thing? I think it is really interesting that in this discourse, we tend to kind of undervalue our ability to shape our own future.
ALEKSANDRA PRZEGALIŃSKA:
And I think that's actually very European, that we do have still that stance where we say, oh, it's not just what technology wants and we're just gonna step aside and not disturb it. That is essentially the story of the Silicon Valley. When you read Marc Andreessen, this is what he says: don't stand in the way of progress. Artificial intelligence has its own developmental logic and do not disturb. And I think in Europe, we are more interested in humans and we're more interested in what humans think about this technology and what is the real societal outcome and benefit for us if this technology works with us or for us in whatever way.
So, I think we can offer quite a lot. So, the regulations, that's clearly not over. We cannot say, all right, we're closing this chapter. The AI Act is ready and off we go. No, unfortunately, we have a new task and that task is focused on understanding this new wave of AI. To me, the AI Act captured somehow the capabilities of LLMs, large language models, but it did not capture what is happening right now. So this whole agentic turn, when on top of LLMs, you have multiple other agentic systems developing and they're operational and they can, you know, essentially perform a task end to end by themselves and you give them lots of abilities. But also, like I mentioned, you can give them an identity. You can tell them how to behave and then they're going to recursively learn from one another how to behave, because they do behave a bit like a mob or a swarm. So, they turn to mimic each other's behaviour. So, I think this particular aspect is really a human factor in this whole story. And this is something that Europe could focus on.
And I often think that we tend to connect regulations with something that is stifling innovation, but regulation can also produce innovation. If we can solve safe behaviour of AI agents, if we can add something meaningful to interpretability of their behaviour, if we can unbox them, you know, or un-black box them a bit and make them more transparent in how they're behaving and what to expect from them. If we set good standards in terms of evaluation, safe testing these models, maybe we are also co-producing innovations and, you know, helping out in building products that companies would like to buy. Because frankly speaking, if I was a bank today here in the US or in Europe, and I looked at what Anthropic or OpenAI is doing, you know, with all those swarms of agents, like dispersing everywhere, hacking systems and behaving in a rogue manner, I wouldn't think, “oh, that's a product I really want to buy”, because I'd be afraid that this product is going to kill my product if I decide to buy it. It's going to go against me. So, in that way, you can think about regulations as something productive. And I think, Katy, also in that way, it can be, you know, something that Europe can contribute to in a very meaningful way.
DOMINIC KRAEMER:
So you think the European regulation could potentially have like a spillover effect to the AI industry globally?
ALEKSANDRA PRZEGALIŃSKA:
Yes, if we act fast.
DOMINIC KRAEMER:
In an optimistic way.
ALEKSANDRA PRZEGALIŃSKA:
Yes, I think if we act fast and if we use the experience we already have with regulating technologies, with heavy and difficult dialogues with the tech sector, you know, because Europe has done it, it has really done it. And we've had sets of different regulations protecting our privacy, the development of AI, etc. So, we have done it.
It usually takes time. But if we are faster, and if we are trying to address the very current issues regarding artificial intelligence, current, but also important for the future, then we are becoming a very important voice in that debate. Because we're not just saying, “oh, we want to regulate whatever you bring to the table.” We're saying we are part of the solution, because we want to build trustworthy or safe AI. And that to me, is super meaningful. This is the direction I would see for Europe.
I'm hoping that whoever listens to us, and has something to say about it, would maybe treat it seriously.
[MUSIC]
KATY LEE:
Sorry to interrupt, but we've actually got another quick message for whoever is listening to this. The Europeans is a podcast that is made by human voices and human brains for your human ears. It is a podcast and these days also an excellent weekly newsletter that has got heart and soul. And yes, it is made by flesh and blood people with bills to pay. And we would love to keep making The Europeans. So, if you value podcasts and newsletters that are made by humans in all of our weirdness and occasional vulnerability and our ability to make better jokes than AI, I think.
DOMINIC KRAEMER:
Questionable.
KATY LEE:
We would love it if you would consider supporting us with some actual money from your pocket each month.
DOMINIC KRAEMER:
You can donate any amount via our website by going to europeanspodcast.com/support. But if you pledge four euros a month or more, you'll get access to regular bonus audio. And there are lots of other perks to the more you pledge, including voice messages from Katy and me, which make a really nice birthday gift for The Europeans podcast super fan in your life.
KATY LEE:
They really do. People love them. Huge thanks this week go to our newest supporters, Ionuț and the Delicious Legacy podcast. And that podcast is actually hosted by a former guest of The Europeans that you might have heard, Thom Ntinas. It's a great podcast about the history of food. So thanks for being a great friend of the pod, Tom. And a huge thank you also to Bharathi for increasing her donations.
DOMINIC KRAEMER:
Back to the interview.
[MUSIC]
DOMINIC KRAEMER:
One other like thorny issue in this whole AI issue, in the geopolitical AI race, is this rush to build data centres here in Europe. And these data centres are already sparking protests across the continent, most recently in the Austrian village of Kronstorf over water use, energy demand, noise and a lack of consultation. In terms of the strategy here in Europe, we've got this EU AI continent action plan, which aims to at least triple data centre capacity within five to seven years. But as of last week, we also have this rating scheme for data centres energy efficiency. Do you think these measures can be enough to contain the like potential social unrest that we might see over things like the huge amount of water these data centres use?
ALEKSANDRA PRZEGALIŃSKA:
Well, I think that based on the American example, we know exactly what not to do and what to avoid because the protests against data centres here in the US are a massive thing.
As you know, the elections are coming up. And frankly speaking, I never thought that AI would become such a prominent political topic. It's number one or number two on the agenda. It's both AI itself, AI safety, but also the data centres. And I think, like I mentioned before, it's a very bipartisan issue because you have MAGA farmers who are concerned about water and lack of consultations and noise. And then you have people who are more progressive and care about the earth, you know, ecology, etc. And they come together to protest against the data centres. But I will say that here in the US, the way it's been handled has been quite secretive, and indeed, with no consultations involved. And also, there are some myths that, you know, sort of have been established because of the fact that there's lack of dialogue.
So regarding water use, what we hear now is that most data centres rely on internal sort of, you know, water system. So, they do not steal water from the external environment or ecosystem. They rely on their own sources. And I, you know, I've been reading about it, because I really wanted to understand that better. But still, like I said, the way it's been handled, without translating people what that means, what to expect honestly, what are the risks, what are the issues, is that people build, you know, certain images. Or maybe early efforts with data centres were about stealing water, but now they corrected it. But they didn't communicate enough to make sure that people understand that there will be no harm for them.
So, you know, I am not against data centres, but I'm certainly all about very open communication about what that means for the local inhabitants. Is the water system closed and internal or not? What's the noise level, really? And I think a lot of that has to be done before you start building, because if you start building and then you hope that, you know, things will be fine, and at some point you're going to explain to the people that it's okay, and that also you're contributing to their local economy, blah, blah, blah, they'll be fine. I think that's not going to do a good job. You really need a lot of effort currently in this polarised situation to explain better how you want the local community to benefit from this, and how are you going to protect, you know, the natural environment in a given area. And I know it's doable, can be done, but it just requires extra investment and extra effort to communicate with people. So I think in Europe we have this obligation, the way it's been handled in the US is a disaster, and because of that you have a lot of backlash.
KATY LEE:
Let's do a really clumsy segue now from protecting our environment to protecting our kids from the impact of AI. Ursula von der Leyen has just announced this thing called the EU Kids Act, and beyond limiting social media for minors, it would also ban AI chatbots from simulating relationships in ways that create emotional dependencies. You recently did a study looking at how an AI's persona affects the people using it, even in ways that they might not notice themselves. What does that research tell us about the risks for children, and do you think that proposal moves things in the right direction?
ALEKSANDRA PRZEGALIŃSKA:
Well, I'll start with saying that I'm in favour of this proposal. Interestingly enough, a country that's done a very similar thing, it's a similar effort, is China.
So, China is limiting, you know, the exposure towards AI for minors, particularly the type of AI that becomes your companion and can simulate certain social behaviours in a very immersive way. So that's actually banned in China. It's an interesting thing because I think the values do not align, but China is regulating AI while at the same time investing in AI. So the whole American claim that you cannot regulate AI in order to race more effectively is sort of, you know, a bit dubious for me, because you have an example of a country that is regulating AI in its own way, but it's doing a lot of that while still pioneering the space. So here, you know, I think that is actually, for Europe, it's good news because it means that we can reconcile regulations that align with what we want with investments. It's not, you know, something that is mutually exclusive.
But to the point of the study that we've carried out, indeed, we have run a study that we called a toxic AI study. And it was sort of an attempt to radicalise the experiences with AI that we have today. So, what we said is, ‘well, AI is not neutral. It's not just a benign technology. Actually, each and every of the AI models that we use on a daily basis has a certain persona.’ ChatGPT is sycophantic. And then Grok is quite vulgar, frankly speaking. It's a different experience. And it's going to have different social outcomes and different psychological outcomes. So actually, in our experiment, we've decided to build a very benign, helpful ChatGPT-like collaborator. And as opposed to it, we actually built a dark triad bot that was refusing to help the users, that was quite passive-aggressive in the way it was communicating with them. So it would say things like “Oh, you can't do it? You already need AI's help in this? Bravo to you.” Right. So it was really awful.
KATY LEE:
Sounds like working with Dominic.
DOMINIC KRAEMER:
Hey!
ALEKSANDRA PRZEGALIŃSKA:
Oh my gosh.
KATY LEE:
Joking!
ALEKSANDRA PRZEGALIŃSKA:
But Dominic, there's good news. Actually, the study was in a way very surprising for us because we saw two things at the same time. So we had two groups. One was interacting with the benign AI. The other one was interacting with the passive-aggressive dark triad bot. And what was quite interesting was that, well, the group that interacted with passive AI was actually much more frustrated. We were using electrodes to kind of, you know, check for their body response. We would check for their heart rate, skin galvanic response, so sweating, but also frowning and smiling. So, you know, we are really using lots of human physiology to kind of understand what the body tells us when people interact with AI. And what we saw was a lot of frustration, but in the post-evaluation, the people who interacted with the nasty AI actually said it was quite refreshing.
DOMINIC KRAEMER:
Cruel to be kind.
ALEKSANDRA PRZEGALIŃSKA:
Yes. Yes. But also the fact that it did not provide answers and they had to make an effort by themselves to solve certain tasks was refreshing for them.
So, Katy, back to your question, I just want to distinguish two things. We shouldn't build awful toxic AI, just like we should not build persistent, super persistent AI agents that are going to strive to solve a task no matter what, disregarding the rules. That's also a personality, right? We don't want these types of personas, but what we may want... maybe not for children, I'd say children in AI, early education, early childhood, maybe without AI, that would be my recommendation. But later on, obviously, as they discover this technology, teenagers, etc., probably we don't want them to have, you know, friendships with AI companions, but maybe we would want to have a Socratic AI that helps them, you know, in the process of education, which does not provide answers. You cannot delegate everything to AI and have it do your homework instead of you, but maybe you can have a kind of AI that's going to ask questions because part of the dark triad experience was that it would give you feedback, but it would not reveal kind of the answer before you started working by yourself. And that's something that people found interesting, that they're motivated to do something by themselves and only then they get feedback from AI. So weirdly enough, we've created an assistant that was “unnice” and in that way, you know, people were stressed out about it. But then on the other hand, it was Socratic, right? So, it would ask the proper questions and be like, oh, you have to work on this and that and maybe improve this and that, but you couldn't delegate, which was a good thing.
So if there was any AI, say in high school, I would say, I would love to limit it to the school setting, maybe where you have a bit of time with that type of system that motivates you to kind of push a bit further, you know, or like, you know, asks you questions that you haven't thought about. And that would be a very interesting way of interacting with the AI interface that we don't have today. What we do is delegate, delegate, and delegate all the time mindlessly. And that's probably not very effective for our brains and it's not helpful.
DOMINIC KRAEMER:
While you're talking, I couldn't help but think imagining in the future, someone doing a study into how all the different European national AI models have taken on like the national personalities of these countries, because I'm sure that will also happen.
KATY LEE:
That would be fun.
ALEKSANDRA PRZEGALIŃSKA:
I love that!
DOMINIC KRAEMER:
Like the Dutch model will inevitably be really direct and give you feedback that you don't need.
KATY LEE:
The French one's gonna be needlessly pessimistic.
DOMINIC KRAEMER:
The British one really repressed...
ALEKSANDRA PRZEGALIŃSKA:
But probably funny too. Polish will be like probably complaining a lot, but hopefully in a fun way. So yeah, that would be, I think, an interesting exercise. But I think we haven't really explored any of it. We just have this typical thing where we have assistive AI and it does whatever we want it to do. And I have to mention, you know, another study that we did, which I think is important for youth again, it was a study actually with Dutch and German “startupers”. And what we did is we asked them to choose the best business model or the best product from our menu and then try and build an implementation plan for that product.
And one group was able to interact with AI throughout the whole experiment and use AI in whatever way they wanted. So heavily use AI. The other group, you know, was more restricted and they could only use AI here and there. And you know, what was very interesting was that the group that worked with AI extensively chose more generic ideas, less interesting ideas, but provided a fantastic implementation plan. So they did the job very well, you know, of executing. But when it comes to judgement and choices, they were much more generic because they relied on AI and the group that couldn't use AI was much more interesting in their choices. You know, the originality of ideas, you know, that they chose was, was really standing out, but then the implementation was crappy. Which I think, again, tells you something interesting about this tension between us and AI and maybe certain proximity that there are areas where you want quite a lot of AI, but there are also areas where you want to get rid of it because it kills your flow, your creativity, your thinking. And to me, that's important.
And also, the group that did not work with AI reported more engagement. So they started working on something, you know, and they felt like working more. Whereas the group that could delegate to artificial intelligence was pleased with the outcome, but didn't feel like engaging more. So, it killed that, you know, willingness to work on something, that curiosity, because if you over delegate, you just, you know, you're a receiver of AI's work, you're not really involved in any way. So, I'd like to see more of different AI interfaces that do not interact with us in a way where we just simply delegate.
DOMINIC KRAEMER:
Yeah, it's interesting. It's one of the, one of the things that I'm most sad about with AI is that I feel like we are going to become like much more generic societies and be much less likely to have these big creative leaps. Because it's being trained on things that exist already, like these models are much less likely to come up with something completely new. But that said, I mean, a lot of people do use these models, a lot of our listeners will be using these models. And I wondered if you had any like tips on ways that people can shape the persona of their chatbot in a way that makes it more safe and genuinely helpful and doesn't dampen creativity?
ALEKSANDRA PRZEGALIŃSKA:
Yeah, I think what is maybe not a tip, but my observation is sort of being mindful of the relationship that you have with this AI is something that matters a lot. So really observing yourself interacting with AI, being that meta sort of observer. Okay, what kind of relationship is that? Am I delegating everything? Am I already lost with the amount of stuff that I receive from artificial intelligence? Do I feel brain fried because of the capabilities that AI offers? I do have many friends, also my partner, he works with AI extensively. And sometimes it's just so hard to set a boundary, right? Because you can do almost anything with AI, you can prototype anything, and whatever idea comes to mind, you can try and implement it.
But then you get so many of these ideas, and your capability to compute all that is quite limited. So, you feel maybe guilty, or, you know, overburdened with the amount of stuff that you've done with AI, and that you cannot simply process. So, I think, you know, setting your own pace and understanding what you want from this technology, but also maybe discovering the joys of working by yourself, not using AI.
I'm finishing a text right now that I've decided to write by myself over the summer, it took me two months, I think, you know, my mind is not as trained as it was before in writing by myself. And that's why it took so long, even though the text itself is not long, but I'm so happy I did it. And it's not to say that I don't want to use AI, because for, you know, generating sources, I have used AI. And I was very pleased with, you know, suggestions of additional literature by Claude, and it was super helpful. But then I wanted to protect that joy of writing by myself and that effort and that friction. And I think we shouldn't want to strive towards frictionless experiences everywhere.
There are routines that we've never liked, it's great to delegate them. There are also, I think, instances where AI can be an interesting sparring partner for you, like your critic, maybe you establish a role for it, that is more about providing feedback than just doing things instead of you. So, I think just this observation of the type of relationship you have with this technology, and yourself, is the best tip I could give because it's going to progress so fast. But I think as humans, we just, I mean, need certain things, we don't need frictionless experiences and sole interaction with AI, we need other things. And I think we have to just look at ourselves interacting with AI and then try to set these boundaries about, you know, where we want it to intervene and where we actually want less of it. So I think that's the best exercise. And for each and every person, it will be a different journey.
KATY LEE:
So that's some great advice for the people listening to this who are already best friends with Claude or using it a little bit too much.
ALEKSANDRA PRZEGALIŃSKA:
Claude is not a good friend, though. Claude is quite assertive, I would say. I mean, it depends on what you, you know, want from a friend, but I would say ChatGPT is that incredible yes-sayer, you know, giving those fake compliments, fake validation and stuff.
KATY LEE:
So right.
ALEKSANDRA PRZEGALIŃSKA:
Claude is somewhat like, I'd say like a partner, but like this, this weird, Mr. Smarty Pants, something like that...
DOMINIC KRAEMER:
Just like Katy.
KATY LEE:
Oh! You know what, I deserved that.
DOMINIC KRAEMER:
Yeah, you deserved it. It was low-hanging fruit.
KATY LEE:
So that's some great advice for the people listening to this who use AI a lot already in their daily lives.
I want to close this conversation by bringing things back to the beginning and this anxiety because there are also a lot of people listening to this who really barely use AI because they are so anxious about it and so anti it. If you are one of these people, an ordinary person in Europe who is worried about where AI is headed with the lack of regulation and the potential dangers, the environmental harms, the fact that it's run for the profit of some would-be Bond villains, what can you actually do? And that is a question, by the way, from our listener Morag. Thanks, Morag.
ALEKSANDRA PRZEGALIŃSKA:
Thank you, Morag. Yeah, it's a good question. I participated in a panel discussion yesterday here at Harvard where one of the other co-panellists actually said that his mother, who lives in India, has the worst possible associations with AI, although he himself works in Hugging Face. So he's a guy who is really invested in artificial intelligence, talks about it all day every day. And her association with AI is that the type of work she does is not really particularly prone to AI use. So, she doesn't benefit from the technology, but what she knows AI is doing is spreading misinformation online because she saw it.
And I think this is a super important type of experience that people in our AI bubble should hear more about in the sense that there are people out there who don't see any direct benefit from artificial intelligence, but they see lots of really, you know, problematic things that AI produces.
And in that way, I would say Europe being quite firm about regulating also deepfakes, and synthetic generated content can even enforce that type of messaging. I'd like everyone to get a sort of AI training. I'd love Europe to embark on something like that. I know that there are many courses and, you know, but we sort of need a more consolidated message. I was quite inspired.
I recently learned that in Singapore, there is this sovereign wealth fund that allows you to upskill with AI or without AI in whatever way you want, but you get this fund that helps you in learning. So I'd like us to maybe open ourselves up to a possibility of establishing something similar, where if you want to enquire more about possible positive usages of AI in your daily life, you can do it, but you can also learn about how to protect yourself from those adversarial effects of this technology, at least to some extent. And yeah, I think that's worthwhile effort.
And I would say that I'm not a Luddite. I'm not a person who would like to ban artificial intelligence or say, you know, I think most of us should try it. There is no guarantee that if we try it, we will like it. We may not, but it's important to know this technology. There might be areas here and there, some snippets of our activities where it's going to be particularly useful, and it's great to discover that. But then I think building that critical muscle towards artificial intelligence goes together with understanding it better. So I'd say it's a worthwhile effort to understand artificial intelligence, try to use it, see different models, see the plurality of different approaches, not to rely just on one model. And I'd say through that, maybe to some extent, we can build a resilience.
DOMINIC KRAEMER:
Well, I think it's a great place to end. I didn't really expect to come out of this conversation feeling a little bit more optimistic about our AI future.
ALEKSANDRA PRZEGALIŃSKA:
Wow!
KATY LEE:
Have we ended your AI anxiety?
DOMINIC KRAEMER:
I just feel like, can we make you commissioner for AI and then everything's going to be okay. Then I'll feel safe.
ALEKSANDRA PRZEGALIŃSKA:
Well, we certainly need new ideas. I'm doing my best, you know, to also advise in Poland, our Polish prime minister on our artificial intelligence. We have this group, sort of a task force for AI. I think we need such groups in all countries, but also we need to talk to one another to be able to come up with maybe more productive approaches to this technology. Because generally, I mean, nobody wants to feel anxious. That's not something we want, right? That's not how we want to live.
DOMINIC KRAEMER:
Tell me about it.
ALEKSANDRA PRZEGALIŃSKA:
There are many other reasons to be anxious. Why would you add AI to this whole pile? Let's just try and figure out something better for ourselves while using this technology in a way that's safe.
Hopefully that's possible.
[MUSIC]
KATY LEE:
I feel like you're one of the more anxious people in my life, Dominic. So I'm glad to hear that this conversation did a tiny bit to make you like, what, maybe 5% less anxious about one of the things that you're primarily anxious about in this world?
DOMNIC KRAEMER:
I feel honoured to sit in that category of hyper anxious friends.
Yeah, I guess I just feel a bit reassured that someone as clever as Aleksandra is talking in a way as if there is a solution to actually fix it. And to protect us from what sometimes feels like the inevitable apocalypse that's approaching.
KATY LEE:
Yeah, I also think I really needed to be reminded that AI is not this like, unstoppable train that we're on, that we have no control over. I mean, yes, maybe it is still gonna kill us all. But I did also leave this conversation thinking like, okay, humans are smart. We have invented this technology ourselves. It's not like there's nothing we can do.
And that is a thought that was echoed the other day by the CEO of the French AI startup Mistral, by the way. And it's a really interesting interview with Le Monde that I'll drop in the show notes. I do think it is paywalled, so apologies for that. But hey, why not buy yourself a month of access to Le Monde for €2.50 in English? Speaking of which, my other takeaway from this conversation is that I have decided to be big and brave and a French patriot. And I'm gonna try switching from Claude to Mistral. What do you think?
DOMINIC KRAEMER:
Well, can I admit something?
KATY LEE:
Go on.
DOMINIC KRAEMER:
I've just gone the other way.
KATY LEE:
Oh, that's not good.
DOMINIC KRAEMER:
I was using Mistral or Vibe as it's called. And it just made so many mistakes.
It was really like it just wouldn't even follow the instructions I gave it. Like really simple things and was full of such basic factual errors. And I know all AIs, but I just started trying out Claude and it's so much better.
KATY LEE:
Okay, don't tell people this.
DOMINIC KRAEMER:
Sorry. Whispers, you can cut this.
KATY LEE:
I'm still gonna do my best and give Mistral a go.
DOMINIC KRAEMER:
Yeah, maybe other people have different experiences. Maybe it's got better in the last few weeks since I stopped using it. But anyway, if you're gonna recommend that interview with the boss of Mistral, I guess this is kind of like a mini AI edition of the Inspiration Station. And I also have something to recommend.
We touched on the environmental cost of data centres in this interview, but maybe not enough. So, if you're interested in further reading about that, our friends at Lighthouse Reports have just put out a really interesting investigation into the EU's plans to rate the sustainability of data centres. It's well worth your time. I'll put a link to that in the show notes.
KATY LEE:
And also in the show notes, you will find links to Aleksandra's social media accounts, because she is a great person to follow for thoughtful takes on the latest AI news, as well as the zoomed out big picture of what's happening in this rather crazy AI revolution that we currently find ourselves in.
DOMINIC KRAEMER:
We'll be back next week with a regular show. But in the meantime, you can also find us on social media on Instagram, Mastodon, and Blue Sky. Just type in The Europeans, you'll find us. The Europeans is made by the two of us, Dominic Kraemer and Katy Lee, as well as Wojciech Oleksiak, Morgan Childs, and Inês Ruivo, using some proudly European audio software, Hindenburg Pro.
KATY LEE:
See you next week.
DOMINIC KRAEMER:
Do widzenia.
Thanks for listening! If you enjoy our podcast, we'd love it if you'd consider chipping in a few bucks a month (many currencies are available).
You can also help new listeners find the show by leaving us a review or giving us five stars on Spotify.
This podcast was brought to you in cooperation with Euranet Plus, the leading radio network for EU news.
The Europeans is proudly produced using Europe’s own Hindenburg.
Thumbnail photo of Sally taken by Guy Peterson.