Peter Denning: Turing's Core Assumption About AI Has Been Wrong Since 1950

Peter Denning, a professor of computer science, argues in a new book published in July 2026 that modern AI rests on a false foundation: an assumption Alan Turing set out in his famous 1950 paper "Computing Machinery and Intelligence." According to Denning, this makes true AGI fundamentally unattainable, regardless of how large language models become.

What was Turing's mistake?

In 1950, Turing put forward the assumption that human thinking is fully described by computable rules and can therefore be simulated by a computer. Everything that followed in AI research builds on this assumption, from the expert systems of the 1970s to the transformer-based LLMs of the 2020s.

Denning considers the assumption wrong. In his view, four key components of intelligence fundamentally resist formalization:

• Common sense — a contextual understanding of situations that is neither described by explicit rules nor derivable from them
• Intuition — the ability to reach correct conclusions without formal logical inference; it develops over years of experience
• Culture — a system of values and norms absorbed by living in a community, not by studying texts about it
• Practical knowledge (know-how) — skills that pass on only through practice alongside a master, not through instructions

"The most important aspects of human intelligence, including common sense, intuition, culture and practical knowledge, cannot be encoded in computers," —

Peter Denning.

Why scaling LLMs doesn't solve the problem

Denning takes direct issue with the view that dominates the industry: make a model large enough and it will "understand" the world. In his opinion, this is an illusion. Large language models operate on statistical patterns in symbols, but they lack the lived experience — cultural, bodily, social — that shapes human understanding.

Denning's argument echoes philosopher John Searle's classic 1980 critique, the "Chinese room": a computer manipulates symbols according to rules without understanding their meaning. Denning updates this argument for modern LLMs, claiming that the transformer architecture does not close this gap. According to Science Daily on July 13, 2026, the book is positioned as one of the most systematic academic critiques of the foundations of AI research in recent years.

What this means

If Denning is right, the current strategy for reaching AGI — scaling data and parameters — does not bring us any closer to human-level intelligence. The practical usefulness of LLMs remains intact: text generation, task automation and data analysis are still valuable tools. But calling today's systems "intelligence" in the full sense of the word is, according to Denning, incorrect.

Frequently asked questions

What did Turing claim in his 1950 paper?

In "Computing Machinery and Intelligence," Alan Turing proposed testing machine intelligence through an imitation game: if a judge cannot tell a computer from a human in a text conversation, the machine can be considered to think. Denning argues that this very criterion set the wrong course, reducing intelligence to linguistic behavior and ignoring the need for lived experience.

Does Denning deny that modern AI is useful?

No. Denning distinguishes between a "smart tool" and "real intelligence." LLMs are useful for automating tasks and working with text, but in his view they cannot become AGI, because they lack the cultural and practical experience that human understanding requires.