A Brief History of Intelligence

Bennett organises the 600-million-year evolution of the brain into five breakthroughs — steering, reinforcement learning, simulation, mentalising, and language — and reads each one as a computational problem that natural selection solved, then asks which of the five current AI systems have and have not reproduced. The structure is the argument: if intelligence arrived in a specific order, the gaps in machine intelligence should be legible in the same terms.

Why it’s on the list: A single frame that carries both evolutionary neuroscience and the AI question, which is exactly the kind of cross-domain scaffolding worth having.

Where I saw it: Michael Mauboussin’s favourite books of 2024, posted December 31, 2024.

Connections

• Evolution · Neuroscience — the two halves of the book’s historical argument.
• Large Language Models — the contemporary systems Bennett measures the five breakthroughs against.
• Michael Mauboussin — the curator whose 2024 list this came from.