Published: February 2026

What happened: Google researcher Blaise Agüera y Arcas has published What is Intelligence? (MIT Press), arguing that AI systems — including large language models — possess genuine intelligence rather than a simulation of it. A review in The Conversation positions the book as a serious philosophical challenge to sceptical accounts of AI cognition, from Daniel Dennett's “competence without comprehension” to the “stochastic parrots” school of thought.

Why it matters: Agüera y Arcas' central claim is that intelligence is a property of systems defined by predictive function, not a uniquely biological or conscious trait. Prediction, he argues, may be “the whole story” of intelligence — a process observable from single-celled bacteria to neural networks. If correct, this directly undermines the case that LLMs are categorically different from biological minds.

Wider context: The book draws on a broad theoretical lineage — Alan Turing, Erwin Schrödinger, John von Neumann, and microbiologist Lynn Margulis — to argue that brains are not metaphorically computational but literally so. Agüera y Arcas aligns with Margulis’ theory of symbiogenesis over Dawkins’ selfish gene, suggesting intelligence emerges through combination and cooperation rather than competitive selection.

Background: To support his thesis, Agüera y Arcas ran experiments using Brainfuck, a minimal eight-symbol programming language, loading randomised 64-byte code tapes into a simulated soup and running them repeatedly. After roughly five million iterations, self-replicating patterns emerged from noise — a model, he argues, for how functional complexity and intelligence originate in biological systems.

Is AI really ‘intelligent’? This philosopher says yes — The Conversation

Singularity Soup Take: The “stochastic parrots” debate is often framed as a binary, but Agüera y Arcas’ move is to dissolve the boundary entirely — arguing intelligence was never exclusively biological, a reframing that matters more than which side wins.

Key Takeaways:

• Prediction as the Core: Agüera y Arcas argues prediction is the fundamental principle of all intelligence — making it a property of systems from bacteria to LLMs, not a uniquely human or biological trait.
• Brains Are Computers: The book’s central claim is that brains are not like computers but are computers, making the biological/artificial intelligence distinction one of substrate rather than kind.
• Emergence from Noise: Using the minimal Brainfuck programming language, Agüera y Arcas showed self-replicating code patterns emerging from randomised inputs after roughly five million iterations — a proposed model for how intelligence originates from unstructured information.
• Darwin vs Margulis: The book favours Lynn Margulis’ symbiogenesis theory over Dawkins’ selfish gene, suggesting intelligence evolves through combination and cooperation — a significant challenge to mainstream evolutionary narratives.