As ChatGPT, Claude and Gemini race towards AGI, computer scientist Peter J. Denning says Alan Turing’s ‘famous test’ sent AI down the wrong road 75 years ago; adds: We are aliens across an…
As ChatGPT, Claude and Gemini race towards AGI, computer scientist Peter J. Denning says Alan Turing’s ‘famous test’ sent AI down the wrong road 75 years ago; adds: We are aliens across an...Alan Turing’s 1950 paper is where modern AI starts. It is also, according to computer scientist Peter J. Denning, where modern AI went wrong. In a new book, “Turing’s Mistake: Escaping the Yoke of Unintelligent Machines”, Denning argues that two assumptions Turing made that year have quietly shaped 75 years of research: that intelligence can exist without a body, and therefore run as software; and that a machine can prove it thinks by imitating a human in conversation.That second idea became the Turing test. Denning’s verdict is blunt. “These two claims have shaped much of AI research and development,” he writes. “My premise is that our acquiescence to these claims has led to the AI mess in which we find ourselves today.” He does not think artificial general intelligence is coming. He thinks something else is—and that it is more dangerous precisely because it is less intelligent.Tacit knowledge: the human thing Denning says AI can never learnDenning’s central concept is tacit knowledge: everything humans know but cannot put into words. He splits it into five categories that he says elude machine learning—common sense, everyday interaction with people and surroundings, feelings and perception, performance skills, and social and historical culture.The evidence he reaches for is Douglas Lenat’s Cyc project, which began in the 1980s to catalogue common sense as machine-readable facts. Four decades and 25 million entries later, Denning notes, it still could not make expert systems into experts.Skills are worse. A virtuoso violinist can play beautifully and still be unable to explain to a student how it is done. Even a robot watching closely, Denning says, has no body with which to feel what the player feels, or what the room feels hearing it.What ChatGPT, Claude and Gemini are actually doing when they answer youAll of this runs into what Denning calls the representation problem. Computers only work on things encoded into physical form. Tacit knowledge resists that by definition. “Words are but symbolic representations of meanings, not the meanings themselves,” he writes, naming ChatGPT, Claude and Gemini as systems that manipulate words without knowing what they mean.Scaling the networks will not fix it. Context, he argues, is fractal—every assumption rests on earlier conversations that rest on earlier ones.The safety implication is the sting. Machines will build tacit knowledge of their own that humans cannot read. “We are aliens across an uncrossable divide,” Denning writes. Agentic machine networks, capable but incapable of understanding us, may prove a bigger threat than superintelligence ever was.
