Antonio Torralba: "I have cameras too, and that doesn't make me a good photographer; the same will happen with AI."

BarcelonaAntonio Torralba (Madrid, 1971) was fascinated by artificial intelligence (AI) from a young age. "I'd always liked computers; not for playing games or programming, I was really interested in AI in particular," he explains in a conversation with ARA.within the framework of the Talent Arena organized by Mobile World Capital Barcelona. At just 15 years old, he began working with some language models, and this obsession has led him to become a guru in this field. He is a professor of electrical engineering and computer science at the Massachusetts Institute of Technology (MIT), directs the MIT School of AI, and was also co-director of the MIT-IBM Watson AI Lab. He also serves on Spain's International Advisory Board on AI. He studied in Barcelona and France, but in 2000 he completed postdoctoral training at MIT and then settled in Boston.

How would you define AI?

— There are two ways of looking at it: from an engineering perspective, it would be the construction of systems that can solve complex tasks, that interact with the world, that adapt... It's the construction of systems, from self-driving cars to language systems with which you can interact. But from a scientific perspective, AI is the study of human and animal intelligence, that is, the way to understand the mechanisms that allow us to learn, understand the world, and how we do it. This is the part that interests me most: my research is not focused on specific applications but on understanding what intelligence is. The word artificial, for me, it is something secondary.

So you're proposing to use machines to understand how our brains work?

— Of course. Machines live in the same universe as we do. They're constrained by the same laws, and the principles needed for a machine to function will be principles similar to those we have. One interesting thing that's happening is that we're seeing a certain convergence between the solutions that are working in AI and somesolutions that occur in biological systems. This convergence is still a long way off, but it's closer than it was 10, 15, or 20 years ago. Perhaps there are principles that govern intelligence, whether artificial or natural. And then, these emerging systems, like ChatGPT, are artificial systems, but they can also be studied as natural objects.

ChatGPT, DeepSeek... Should we be afraid of them?

— No. It's a technology that needs to be learned, but it's still a tool. Being afraid of these tools makes us distance ourselves. And I really think it's much more constructive not to be afraid, to approach them, to understand them, because in the end, it's a tool that will make many things easier for us. For example, when writing articles, you might think that maybe ChatGPT will suddenly be able to do it. I don't agree, because these systems can create images; there's a huge difference between when I ask them to create an image, or when an artist who understands style and aesthetics uses the same tool and asks for an image: they do it much better.

Because?

— Because it's not just about using the tool and that's it. Then you have to evaluate the result, know how to interact with the machine to steer it in the direction you think is best, and a normal person won't have that level of sophistication. This level of sophistication is what has always shaped the artist. I also have cameras, and that doesn't make me a good photographer. Why? Because a photographer isn't just a person who has a camera; they're a person who knows how to use it in a sophisticated way, and the same will happen with artificial intelligence tools.

What will be necessary, then, is to learn to master the language in order to speak with the tool?

— You'll likely need to master all of these areas, just as when using a camera, it's not enough to know the laws of light; you also need to know how to adjust the focus.

From Boston, do you see a difference in how AI is progressing in Europe compared to the US?

— There really is a big difference, and I think Europe is now also trying to invest more in artificial intelligence to contribute to what's happening and keep up, but I think there's still a lot of work to be done.

But the differenceIs it as big as it looks from here?

— Yes, there really is a quite noticeable difference. But there's a lot of investment, and now in Europe, too, there are a number of companies, especially in France and England, that are doing very interesting things.

Is there too much regulation?

— Every society makes every decision based on its values. The point is that if you regulate heavily and want to make significant progress, you must offset this with funding to introduce the techniques necessary to satisfy the regulation. Therefore, regulation, if not accompanied by funding, is a hindrance.

Where will we be in 5 or 10 years?

— Five years is a long time... In fact, if you think about it the other way around, where were we five years ago? But if I had to make a prediction, I'd say the world of robotics is the one that's going to change the most because all the advances we're seeing in AI today are in the software field. The machinery part isn't changing that much. We always thought the future would be a future of robots. That hasn't happened; now it's a future of systems you talk to over the internet, which have nothing to do with what was imagined in science fiction books 20 years ago. I think the world of robotics is the revolution that should have happened and still hasn't. Today, the most sophisticated robot we have is a vacuum cleaner that, from time to time, gets stuck under chairs.