Demis Hassabis, Nobel Prize winner in Chemistry: “We will need a handful of great advances before we reach general artificial intelligence” | Technology | EUROtoday

“I think we are at an important moment for science and artificial intelligence (AI). In the last two or three years, we have seen how AI tools are becoming powerful and mature enough to be able to apply them to really relevant real-world problems.” The one who says these phrases is Demis Hassabis, probably the most related and elusive characters within the self-discipline. This 48-year-old Briton is the founder and CEO of Google DeepMind, the AI ​​analysis arm of the know-how big, and the current winner of the Nobel Prize in Chemistry.

Small and timid in look, there may be silence wherever he passes. He is adopted by a small entourage, like rock stars. Dress in Silicon Valley trend: blue t-shirt and black blazer, denims, sneakers. His public appearances are as few because the interviews he has given to this point. He agrees to take a seat for a couple of minutes in London with round twenty worldwide media, together with EL PAÍS, throughout the framework of the AI ​​for Science discussion board, organized on Monday by his firm and The Royal Society. There he deploys his quick verb and avoids the extra political questions, equivalent to what AI regulation he expects when Donald Trump returns to the White House. This self-containment is helped by having James Manyika, vp of analysis at Google, at his aspect.

In that assembly with journalists, Hassabis left a mirrored image that clashes head-on with the proclamations of some rivals. “Large multimodal models of generative AI [los que son capaces de interpretar textos, imágenes y vídeos] They are going to be a fundamental part of the global solution to develop general AI [la que iguale o supere a las capacidades humanas]but I don’t think they are enough on their own. “I think we’re going to need a handful of other big breakthroughs before we get to what we call general AI,” says this former chess prodigy. A wake-up name to those that, like Sam Altman, CEO of OpenAI, or Elon Musk, know-how entrepreneur and robust man of the longer term Trump Administration, have been making certain that this superintelligence is simply across the nook.

Yours isn’t just any opinion. Hassabis is the character of the second within the AI ​​world. Google handed him the reins of AI analysis final 12 months, merging his firm, DeepMind, which beforehand functioned as a form of primary science laboratory considerably unbiased of Google’s construction, with the opposite divisions devoted to advancing that know-how. . The most seen end result has been a fast push for Gemini and different generative AI instruments.

But Hassabis’ consecration got here a month in the past, when he and two different DeepMind colleagues gained the Nobel Prize in Chemistry for the event of AlphaFold, an AI instrument that has been capable of describe the construction of the 200 million identified proteins. This is an advance that’s just about inconceivable to realize with out the increase of AI and confirms what Hassabis maintains: AI is destined to be one of many pillars of scientific advances within the coming years.

The son of a Greek Cypriot father and a Singaporean mom, Hassabis remembers that when he based DeepMind, in 2010, “there was still practically nothing you could do with AI.” Over the years, machine studying appeared (deep studying) and reinforcement studying, strategies that gave an actual increase to the self-discipline. And in 2017, Google scientists introduced a brand new algorithmic structure that enabled the delivery of generative AI. “It took several years to figure out how to use that type of algorithm and integrate it into hybrid systems, like Alpha Fold, which includes other elements.”

“During the first years we operated on a more theoretical level. We focus on games and video games, which were never an end in themselves. It gave us a controlled environment in which to operate and ask questions. But my passion has always been using AI to accelerate scientific knowledge. We managed to scale until we were able to solve a real-world problem, such as the structure of proteins,” remembers the engineer and neuroscientist.

Is there any space of ​​science that AI can not have an effect on? “I think we are well prepared to help solve those problems that can be described as a big search through a combinatorial space. You have to be able to build a model and describe the goal in terms of a metric. For example, in protein folding, minimizing the error between the actual positions of atoms and the predicted ones.”

Another concern that arose through the dialog with Hassabis was the environmental footprint of AI. Are you apprehensive concerning the quantity of vitality and water you devour? “My opinion is that the benefits of these systems will far outweigh the energy consumption they have. Tools for weather forecasting, electrical grid optimization or materials design will contribute to solving climate change. I mean, we also have to try geopolitically, but that doesn’t seem to be working very well. So I would try technical solutions: new battery designs, new superconductors, nuclear fusion. That is what AI can contribute.”

The government scientist

Hassabis has earned a popularity as a scientist. The Nobel is a very powerful of an extended record of awards gained earlier than turning fifty, together with the Princess of Asturias for Scientific and Technical Research in 2022 for “his contributions to the advancement of Artificial Intelligence and its full integration into society or the 2023 BBVA Foundation Frontiers of Knowledge in Biomedicine Award for AlphaFold.

A chess master at the age of 13, Hassabis was designing video games at the age of 16. He decided to focus his studies on neuroscience, a field that he has always been passionate about, and in 2010 he founded DeepMind with two colleagues, a start-up dedicated to AI research. which Google bought in 2014. Hassabis and DeepMind gained international recognition thanks to AlphaGo, a program that managed to beat Lee Sedol, world Go champion, 4-1. The feat was notable because intuition is essential in the ancient Asian game, much more complex than chess and with more possibilities on the board than atoms in the universe. And AI is supposed to do a lot of things, but it doesn’t handle intuition. The algorithm learned to play by itself and developed strategies never seen before.

But DeepMind’s flagship project to date is AlphaFold. Predicting the structure of 200 million proteins is considered a key milestone in curing diseases such as cancer or Alzheimer’s. The third version of the program, Alpha Fold 3, which predicts the interaction between proteins and the rest of the essential molecules of life: DNA and RNA, small molecules and antibodies. His team is also working on a magnetic field optimization system that may be essential to achieving the dream of nuclear fusion. “It’s something that could come in the next decade,” he says.

Hassabis’ new place in Google’s organizational chart, as head of the AI ​​analysis department, might take time away from his scientific profile, which has confirmed to be glorious, to the detriment of a extra managerial position. Company sources say that he continues together with his typical initiatives and concepts. Time will inform whether or not or not AlphaFold is his final nice scientific milestone.