Multinational Collaboration and Advanced AI — Yoshua Bengio | CDS 2026

We generally underestimate the changes that are likely to come with AI. We generally underestimate the power that AI will have and will give to those who control it. And democracy is about sharing power. But if you have these systems eventually could be even smarter than us in many ways that concentrate huge amounts of power, right? Because we kind of dominate this planet because we are smarter than animals, right? But if we build machines that have much more intelligence than we do, at least in some areas, that's going to change the game completely. And it's a real threat, this concentrated power, the fact that it is very very expensive to build, that it takes particular resources that you know are not easy for anybody to get. The fact that it's currently driven by two countries and maybe you know 10 companies in the world. This I think is something we should be careful about. If we let things go with the current trend, there's a good chance we end in a world where we won't have much agency, like maybe we'll have democracy in our countries but all the power will be somewhere else. The decisions will be taken somewhere else. And you know I think the mistake would be to think that this is it, that you know we're done or let's hope for the best when we still have agency right now. It's not too late. As Tom was saying, we have time to build an alternative to the current leading models where the power over these AIs will be shared, will be multi-, will be really driven by the public good and public governance.

I'm not worried about it. By definition, you know, ethics is about dealing with values that are in tension. And yes, we want progress but we also don't want to lose human dignity. We don't want to lose our agency in choosing our future. We don't want incredible economic disparities to grow even more. And in order for us to be able to make the choices that we really want collectively, we need to have those tools in our hands. So we do need to build alternative infrastructure. So regulation is extremely important because that's a way to make sure companies will not go in places where we don't want. But we live in a world where for example if European companies were not allowed to implement AI in many different ways that say their competitors in the US are, then the European economies would not be competitive against the Americans or the Chinese. So, we're stuck in this global game. Ultimately, the only way that we'll be able to make sure AI goes in the direction that is good for humans is with global coordination. Now, I know it sounds like utopian, but it's like climate, right? So, if it's only the competition between countries, we can't solve these problems. The only way we can bring about a future that's good is if we have international coordination and that's true in particular for AI. I anticipate that because of the weaponization of AI that is now more and more possible, for example with models and the models that will follow, for national security reasons, we're going to start seeing governments, you know, forced to agree with each other about the sort of thing that needs to be done like evaluating how powerful they are and can they be used by terrorists, for example. So there will be coordination but I think it is also important that you know it's not just two countries deciding for the future of this technology and how it's going to drive the world.

Of course. Yes, there's way too much concentration of power.

I mean Elon Musk tell me how you can take away my power. So I think the most important ingredient to try to deal with these questions that seem intractable is public awareness. Once people understand that the future of their children can be taken away from them, they will demand that their governments do something. And individual governments say in Europe probably can't do much or my government in Canada can't do much. But together we can work with democrats around the world to make sure that the power of AI is going to be managed properly, the risks are going to be managed so that we think more of it as a global public good, right? Something that we have to manage collectively because let me give you an example. If it's just the market forces, we're going in very very bad places. AI is getting better and better at pressing on our psychological buttons and getting us in a way addicted to positive feedback that we're getting. And eventually this is a way to manipulate us as well. This is a way to remove some of our social life and it's something that is profitable and so just left to their own devices the companies will want to do it like they want to sell stuff and people will buy it even though it's bad for them just like cigarettes right. It is because governments got into dealing with the issue of tobacco that we ended up, you know, we're now in a better place and it's not just going to happen by the forces of competition and markets.