Blaise Agüera y Arcas (Providence, Rhode Island, 50 years old) is one of the most influential figures in the development of artificial intelligence. In June 2022, he published a reflection in The Economist about LaMDA, the Google chatbot he was working on, describing how interacting with this AI made him feel like "the ground was shifting beneath his feet." His colleague Blake Lemoine even claimed to have conversed with a machine that had "soul," which cost him his job after leaking confidential information. Agüera, then head of Research at Google Research, admitted to EL PAÍS that LaMDA forced him to rethink many ideas about intelligence. Five months later, OpenAI launched ChatGPT, generating a global impact. Google had been developing similar technologies for years but felt they were not yet ready for the public. However, the competition accelerated the process. Agüera, a physicist and computer engineer of Catalan and American descent, has been reflecting on the nature of intelligence for years. What defines it? Can a machine be intelligent? His answer is affirmative, and he argues it in his book What Is Intelligence? (not yet translated into Spanish), where he explores the concept in depth. Currently, Agüera is vice president of Google and leads the Paradigms of Intelligence (Pi) group, dedicated to fundamental research in AI. He is recognized for developing federated learning, a technique that allows training AI models using data distributed across multiple devices. From Chicago, he spoke to EL PAÍS via video call. In his book, Agüera anticipated that the arrival of large language models would spark a debate about intelligence. He observes two positions: denial and acceptance. "Most deny that AI is truly intelligent, arguing that it only simulates. But if a student passes an exam, you can't say they are just pretending to know. Intelligence is measured by the ability to solve complex problems, not by occasional mistakes," he explains. Regarding the definition of intelligence, Agüera argues that there is no universal consensus. For him, it is the ability to understand and act in a complex environment, modeling both oneself and the environment and anticipating the future in a useful way. This implies that there are multiple forms of intelligence: mathematical, social, emotional, musical, among others. Does AI fit into this definition? "Without a doubt. Every time you interact with an AI, you are its environment. Its usefulness depends on its ability to model you and understand the context of the conversation," he states. Agüera identifies as a functionalist: he cares about the outcome, not the internal process. "If language models only memorized, they wouldn't be able to solve tasks for which they haven't seen examples. Generalization is key: if you can solve problems outside of what is memorized, that is intelligence," he argues. On the difference between humans and machines, he points out that the distinction lies in the "how," not the "what." "Nature and AI solve similar problems by different means, just as birds and airplanes do with flight. We do not fully understand the brain, but AI has driven advances in neuroscience," he comments. Agüera argues that life is computational, not in the literal sense, but as a science of causality. "Causality arises when we analyze causes and effects, and that requires decision-making ability, which is computation. This is how living beings and the brain function," he explains. In his book, he proposes an expanded view of evolution, beyond classical Darwinism, incorporating symbiogenesis: cooperation to create new entities, as occurred with multicellular organisms. This process, he believes, drives creativity and evolutionary novelty. Will AI equal or surpass human intelligence? "Some pioneers believe it has already happened. Current models outperform humans in many individual tasks. Human intelligence is collective: together we achieve feats like traveling to the Moon or transplanting organs. LaMDA marked a leap with unsupervised learning, modeling language without specific tasks," he notes. Agüera maintains that prediction, like that performed by Gemini or ChatGPT, is a form of intelligence. "At first, I thought predicting the next word was just pattern coincidence. But as we scaled the models and trained them with large volumes of data, they began to solve complex problems. There is no magic, just larger models," he concludes. Recently, Google announced the Suncatcher project, which aims to bring data centers to space to process AI. Agüera, an enthusiast of the project, reveals: "The idea arose thinking about energy efficiency. AI consumes more and more energy. The long-term solution is space solar energy, much more abundant than on Earth. We want to take solar cells and computing off the planet." Is it feasible? "It's a project for decades ahead, but we are already taking the first steps. Space computing will be common in the future," he assures. In the meantime, he advocates for improving AI efficiency on Earth, revitalizing nuclear energy, and boosting renewables, especially solar. Source: elpais.com