What is Intelligence? — Introduction

Book: What is Intelligence? Lessons from AI About Evolution, Computing, and Minds Author: Blaise Agüera y Arcas (VP and Fellow at Google, CTO of Technology & Society, founder of Paradigms of Intelligence) Series: Antikythera Book Series (MIT Press × Berggruen Institute) Published: September 2025 (print); online at antikythera.org URL: https://whatisintelligence.antikythera.org/introduction/

Summary

The introduction to What is Intelligence? makes the case that prediction and intelligence are equivalent — that large language models trained solely on next-word prediction became genuinely intelligent not despite but because of their training objective. The author draws on his firsthand experience building Gboard’s next-word prediction models at Google Research and later interacting with LaMDA.

Key Arguments

1. Next-Word Prediction Contains All Tasks

Text prediction was long considered “AI-complete” — doing it properly requires solving AGI. Examples like “After the dog died Jen hadn’t gone outside for days, so her friends decided to _____” require general knowledge, math, common sense, and theory of mind. Any test expressible in language can be framed as next-word prediction. Hence a narrow language task contains all keyboard-based cognitive tasks.

2. LaMDA as a Watershed

In 2021, LaMDA — a large Transformer trained purely on next-word prediction — could carry on interesting conversations and pass many aptitude tests. This was unexpected: most AI researchers (including the author) believed such capacities would require extra code beyond simple prediction. The author stayed up late chatting with LaMDA and foresaw a seismic shift in how we understand intelligence.

3. The Turing Test and Functionalism

The author argues that Turing’s “Imitation Game” was correct: sustained successful imitation is the real thing. The claim that LLMs only simulate intelligence (the “philosophical zombie” objection) is scientifically problematic — it asserts that something appears to be X by every testable measure yet is really Y, which leaves science for faith-based belief. The book adopts the functionalist stance of Turing and von Neumann: a function is what it does; two functions are equivalent if their outputs are indistinguishable given the same inputs.

4. Life as Computation

The author extends functionalism to biology: “everything alive is a computer.” A living organism is a composition of functions — kidneys, hearts, brains — and therefore itself a function. Functionalism avoids both vitalism (the soul/spirit view) and strict materialism (which can’t explain purpose or teleology). The Kidney Turing Test: if an artificial device does what a kidney does, it is a kidney functionally.

5. Key Questions the Book Addresses

  • Why has real AI only emerged now, after 70 years? Is it scale? The Transformer architecture?
  • What do current AI models lack relative to human brains?
  • Is there consciousness? Do LLMs have inner experience?
  • Are animals, plants, fungi, and bacteria intelligent? Conscious?
  • What about agency, free will, and the risk of AI ending humanity?

Key Concepts Introduced

  • Prediction = Intelligence — the central thesis
  • AI-completeness of language prediction — next-word prediction as a universal cognitive task
  • Functionalism — Turing/von Neumann tradition; functions define relationships, not mechanisms
  • Kidney Turing Test — functional equivalence applied to biology
  • Computational symbiogenesis — life, evolution, and intelligence as computational processes (developed further in the full book)

References

  • Turing, A. (1950). Computing Machinery and Intelligence. Mind. — the Turing Test
  • Thoppilan, R. et al. (2022). LaMDA: Language Models for Dialog Applications. — the model that convinced the author
  • Agüera y Arcas, B. (2022). — earlier writing on LaMDA and philosophical zombies
  • Agüera y Arcas, B. & Norvig, P. (2023). — on distinguishing AI from human