When Harari Talks About AI After a Booed Macinka
Yesterday I went to a surreal event. Harari in Prague. Yes, Yuval Noah Harari, a great favorite of mine. I never pass up a chance to see him in person, but he had never lectured in the Czech Republic before; the closest he came was Berlin, the year before last.
The event was organized by the Solvo institute, a strange institution founded by the wife of the billionaire Tykač. Why a very conservative institute would invite a liberal socio-historian is not entirely clear to me. My guess is that the first name that occurred to them was one at the intersection with the sexy field of AI, and they didn’t think any further; Acemoglu was, after all, zu viel. Just a reminder that Harari is not exactly a cheap speaker. He charges roughly as much for a commercial talk as I do, only in euros.
In the end Solvo turned it into a conference called The Future is Personal, on the domain Harari.cz. I stared at the program a bit, went to watch two panel discussions, and found them shallow, uninteresting, with nothing new to offer. The kind of nodding-along debates where nobody touches anything uncomfortable; both the dramaturg (the program curator) who picked the guests and the moderators simply let it down. But fine, a focus on preaching to the converted would work if there were an audience. The afternoon program had somewhat lower attendance than our Vibecoding Talks, and in Forum Karlín that was quite visible. Oh well, a few of us set up a discussion club in the foyer and I was happy.
I came back only for the evening Harari. I caught the debate between Anna Shavit and Aleš Rod on the topic “Will Europe survive without children?” Aleš Rod understands the subject and brought a number of concrete, politically uncomfortable proposals: pensions at subsistence-minimum level from the years 2065 to 2070, a ban on early retirement, higher taxes, support for multigenerational homes through an interest deduction, a million crowns for the first child before age 23, shorter university studies, and prevention for people in their thirties through fifties as the cheapest way to keep them in the workforce. The moderation, however, was very weak; it neither picked up his ideas nor developed them.
Macinka’s Whistled-Down Migration
And then Petr Macinka and the Era of Migration. I could have sworn that when I bought my ticket for 4,000 crowns, he was scheduled quite far from Harari’s appearance, and by some miracle his speech got moved right before it. He was supposed to speak for 20 minutes. His arrival was met with whistling from the audience, and people treated his appearance as an intermission, chatting and going to get refreshments, which visibly made the minister nervous. The speech may not have been completely bad; it had a small idea or two, and he referred to Goodhart and his distinction between people who are “anywhere” and people who are tied to a place. But in fact he offered only a mild attack on Merkel, without hinting at what to do about the problem Rod had developed just before him: namely, who will work here if children aren’t being born. Poor, unproductive stuff; ChatGPT, paid version, would have written it better. And since people were already talking among themselves quite loudly and time was running on, Macinka wrapped up by saying he would save the rest of his Castro-style speech for another time. That was the only moment of introspective insight. Then he dropped off the stage and the applause finally began, because Harari had come on.
And Harari? Brilliant, as always. His strength is the ability to make complex concepts accessible in simple language.
In his talk Harari described three political paths that Europe (and other countries) can take in the era of AI empires. And he reminded us that it is the job of political parties to present these options, explain them, and let people choose among them by voting for the party that advocates them. But developments are now moving so fast that we don’t even manage to realize the choices.
The AI Button in Washington
Harari offers the image of a “button in Washington” and reminds us that infrastructure built on foreign AI can be switched off remotely, as Ukraine’s dependence on Starlink showed. He also notes that, unlike growing wheat or extracting oil, the value from data and AI can be concentrated in one place. Which is what is happening now.
According to Harari, countries outside the US and China therefore have three options:
The default option is to do nothing and become a vassal of one of the AI empires, in the Czech case most likely the American one. Such a state is no longer an ally, and when the Americans say “jump,” it jumps, because the AI that takes care of prosperity is controlled by America.
The second option is full independence based on your own AI infrastructure. But here the price tag is around 500 billion euros, and the Czech Republic can’t afford that. Of individual countries, hardly even Great Britain or India could, and even they would have to sacrifice their budgets for the social system, health care, and education.
The third path is a joint European project. The EU could carry such an amount, and it has the scientific and technical base for it; after all, a large part of American AI is built by scientists of European origin. The price, however, would be much deeper integration without individual states’ right of veto in every election cycle, because such a project would be unfeasible if it went through political review every few moments.
Each of these paths is expensive, and the role of political parties is to offer and explain the choices. According to Harari, parties should clearly offer voters one of them. Not consciously choosing one means losing independence within a few years, and if that is going to happen, it is better to accept the role of vassal consciously than to fall into it blindly.
Personally, I missed a fourth path: open-weight models. They let you jump onto a moving development, follow it at a small distance, and run AI on your own infrastructure, so the button would be ours. The price is that we take over the values and biases of the base model and remain dependent on American and Chinese companies continuing to publish them. And we would still buy the chips from Nvidia.
AI as Migrants
The rhetorically strongest moment of the whole talk, in my view, was Harari’s comparison of AI to migrants. In the Czech context it also followed directly from Macinka’s speech on migration and contrasted with it.
Harari summarizes the three fears of opponents of migration: that they will take our jobs, change our culture, and are not politically loyal. And he points out that all of these fears fit AI above all. Immigrants in the form of AI agents arrive at the speed of light, from just two countries, and take jobs mainly from people whose work consists of processing information. In particular, today’s recommendation algorithms on social networks change the culture of the majority society, as we are witnessing here as well. And it certainly can’t be said that AI algorithms are politically loyal: certainly not to the Czech Republic; if to anyone, then to the USA or China, but even more so to the corporation that created them, if to anyone human at all.
Harari sees a major contradiction in what AI companies tell investors and what they say in debates about safety. To investors they promise an all-capable superintelligence, without which the investment won’t pay off. When the talk turns to risks, they label them philosophical speculation. The companies want empirical evidence of harm (after all, nobody died in the Hugging Face incident), while accepting and presenting the benefits without evidence. In my opinion, the strongest analytical argument of the talk.
One of Harari’s main questions is the dilemma of whether we can control something that is supposed to be more intelligent than we are. It is a very important question, and Harari answers that it’s not possible. A chicken cannot control the farmer. Here, though, I think the case isn’t fully made. Intelligence in itself doesn’t give goals or power, and the whole field of alignment rests on the assumption that this connection can be influenced. Harari treats it as self-evident, albeit rhetorically perfectly, and presents it as an important warning.
Harari points out that as organic beings we need time to adapt to change. But AI doesn’t sleep, it accelerates everything and exhausts us in the process. That, according to Harari, is exactly what is happening to people on the front line, that is, in AI, banks, media, and politics: they don’t even have time to think about what is going on. The heads of AI companies reportedly tell him they would like to slow down but can’t trust the competition, which would overtake them. So time requires trust, but the AI systems themselves are destroying it. They build trust in algorithms and erode trust between people. A vicious circle emerges. Yet for human society, trust is essential.
The world is dealing with two problems, the breakdown of trust between people and states, and the development of superintelligence. The sensible thing would be to restore trust first and then develop superintelligence together, more slowly and more safely. We are doing the opposite, and if people can’t cooperate, they won’t manage what emerges from the labs either. From that follows the conclusion that this is a task for politics. He calls the companies’ claim that development is inevitable and that it is too late for politics a lie. According to Harari, a political approach to AI is essential. Harari says that American companies fear regulation from Brussels and Washington and that social networks may spread Euroscepticism for that reason. He also proposes regulations of his own, such as a ban on AI companions for children or a ban on AI passing itself off as a human (in the EU the AI Act requires this as of this year).
Harari, however, offers no concrete mechanism for slowing down, no international agreement, and no way to restore trust between the USA and China, which he himself considers a necessary condition. The argument is thus more a diagnosis than a proposal.
And with that I’ll end my reminiscence of Harari’s talk. The talk presented the problems we face around AI very well and vividly. But it turns out that the solution is political rather than technical, because tech companies can’t reach agreement among themselves. And we, for our part, can’t allow the fate of the world to be decided by a very narrow group of people who have their own interests and represent almost no one except a few billionaires.
The Czech foreign minister spoke about the migration of yesterday, about a closed chapter, and overlooked what is already here. He did mention AI briefly at the end, but only as a backdrop from which he drew no conclusions. Just a few minutes later Harari showed that nobody is controlling the real migrants of today. And that we can probably bring them under control only together, within deeper European integration, which Macinka has long opposed. This contradiction between the speech of yesterday and the picture of today shows how naive Macinka’s stance is.
Inviting Harari so that the luster of his fame and wisdom would also fall on a failed foreign minister might have worked. It didn’t. Above all, it shone a light on the chasm between wisdom and a lack of imagination.