Blaise Agüera y Arcas, a vice president at Google, argues that recent advances in artificial intelligence show that intelligence does not require any mystical ingredient, but rather larger and more capable models. In his new book What Is Intelligence?, Agüera contends that the boundary between human and machine intelligence is becoming increasingly blurred at an intellectual level.
Agüera, a physicist and computer scientist who leads Google’s Paradigms of Intelligence research group, first drew widespread attention in 2022 when he wrote that an internal Google chatbot, LaMDA, made him feel “the ground shift beneath his feet.” At the time, a Google engineer claimed the system was sentient, a view Agüera rejected, though he acknowledged that the technology was forcing him to rethink long-held assumptions about intelligence.
Five months later, the public release of OpenAI’s ChatGPT triggered a similar sense of shock worldwide. Google had been working on comparable systems for years but had opted not to release them, a decision that left it racing to catch up once competitors moved first.
In his book, Agüera argues that intelligence should be defined by outcomes rather than internal mechanisms. If a system can repeatedly solve complex problems and generalise beyond memorised data, he believes it meets a meaningful definition of intelligence, even if it sometimes fails. He describes himself as a functionalist, saying intelligence is demonstrated by what an entity can do, not by whether it thinks in a human-like way.
Agüera also rejects the idea that large language models are merely “stochastic parrots” that repeat memorised text. He notes that tasks such as multi-digit arithmetic or reasoning over novel problems cannot be solved through memorisation alone, implying that models must discover underlying rules during training.
He views artificial intelligence as a natural extension of evolution, arguing that life itself is computational at its core. In his view, biological intelligence and machine intelligence solve similar problems through different materials and processes, much as birds and aircraft both achieve flight using different mechanisms but the same physical principles.
Asked whether AI could equal or surpass human intelligence, Agüera said that in many domains, large models already outperform individuals. Human intelligence, he argues, is largely collective rather than individual, emerging from cooperation and accumulated knowledge. He credits breakthroughs such as unsupervised learning in language models, pioneered in systems like LaMDA, as a decisive turning point.
Agüera has also spoken publicly about Google’s long-term ambitions to address the rising energy demands of AI. He recently expressed enthusiasm for “Suncatcher,” a research initiative exploring the possibility of processing AI workloads in space using solar energy. While describing the idea as a multi-decade project, he said the growing power needs of AI require thinking beyond Earth-based infrastructure.
In the meantime, Agüera said progress will depend on making AI systems more efficient, expanding renewable energy and reconsidering nuclear power as a cleaner alternative to fossil fuels. He believes these steps are essential as AI becomes a permanent and increasingly powerful part of modern society.
