Voice of Urdu - وائس آف اردو

Turing’s Mistake: Denning Challenges AI’s Core Assumptions — Denning disputes the idea that intelligence can exist independently of a physical body and be recreated purely in software

Editorial Team

Computer scientist Peter Denning argues in a new book that two influential assumptions Alan Turing made in 1950 have steered decades of artificial intelligence research toward goals that may be unattainable and potentially risky. In Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, Denning disputes the idea that intelligence can exist independently of a physical body and be recreated purely in software, and he challenges the belief that a machine can demonstrate intelligence by convincingly imitating a human in conversation—an approach later associated with the Turing test.

Denning writes that these assumptions have shaped “much of AI research and development” and argues that “our acquiescence to these claims has led to the AI mess in which we find ourselves today.” He contends that the pursuit of artificial general intelligence (AGI)—machines with human-level intelligence—is unlikely to succeed, while the systems society is building could introduce significant new risks.

The argument matters as large language models (LLMs) and other machine-learning systems are increasingly deployed in everyday tools and decision-making processes, often marketed as steps toward more general-purpose intelligence. Denning’s central claim is that key elements of human intelligence are not simply information-processing problems and may not be representable in a form computers can truly “understand.”

Rather than focusing on a future takeover by hypothetical superintelligent systems, Denning warns about nearer-term dangers: networks of automated, “agentic” machines developing forms of machine intelligence that remain below human general intelligence but still have the capacity to cause severe problems for people. In his view, the difficulty of aligning machine behavior with human goals is rooted in limits on what machines can grasp about unspoken human meaning, context and culture.

Denning’s critique centers on what he calls the “tacit knowledge” problem—human understanding that cannot easily be put into words or translated into representations computers can process. He argues that machine learning cannot capture five major categories of tacit knowledge: common sense; everyday interactions with people and the environment; emotions and perception; practical performance skills; and the social and historical knowledge embedded in culture.

As evidence of the difficulty of encoding common sense, Denning points to long-running attempts to assemble vast knowledge bases. He cites Douglas Lenat’s Cyc project, launched in the 1980s to build an extensive collection of common-sense facts; after four decades, Cyc contained roughly 25 million entries. Denning argues the outcome still fell short: “Yet even this treasury could not add up to a background of common sense sufficient to make expert systems smart enough to be experts,” adding that “Cyc validated that much of the knowledge that makes people experts cannot be articulated as propositions.”

Denning says practical skills are even harder to translate into machine form because they depend on embodied experience. “Our performance skills in thousands of domains cannot be communicated to machines,” he writes, distinguishing between “know what” and “know how”: “Whereas descriptions of skillful outcomes (‘know what’) can often be represented as bits and stored in a machine, we do not know how to encode the embodied knowledge for skillful performance (‘know how’).” He also places intuition, gut feelings, imagination and spontaneous creativity among forms of tacit knowledge that remain beyond machine reach.

Denning’s book revisits a foundational moment in the modern debate about machine intelligence: Alan Turing’s 1950 proposals that helped define how researchers and the public evaluate “thinking machines.” In Denning’s account, two ideas in particular became deeply influential: first, that intelligence can be separated from the physical body and reproduced in software; and second, that human-like conversation can serve as the demonstration of machine intelligence, an idea later formalized in discussions of the Turing test.

He argues that these premises helped steer research agendas toward building systems that resemble people in language and reasoning, reinforcing the idea that successful imitation in dialogue indicates genuine intelligence. Denning contends that the cumulative effect has been to privilege conversational performance as a proxy for thinking, even as systems remain limited in their ability to connect words to lived experience and the human contexts that give language meaning.

In making the case that human intelligence cannot be cleanly separated from embodiment, Denning emphasizes the role of tacit knowledge: the background understanding people use without consciously articulating it. He says this includes common sense and everyday interactional knowledge, as well as emotional and perceptual capacities—areas where people draw on bodily sensation and lived experience rather than explicit rules.

Denning also highlights culture and context as essential components of intelligence, not optional add-ons. He describes culture as encompassing values, norms, judgments, history, communities, moods, and relationships involving power and care—elements that shape what people mean and how they interpret others. Context, he argues, allows people to detect sarcasm, humor, sincerity and emotion, and to decide when to be diplomatic or when to joke. He adds that context is built over time through prior conversations and situations, creating layers of shared assumptions that are difficult to isolate and encode.

As part of this background, Denning points to decades-long efforts to represent common sense in databases as a way to make machines “expert.” The Cyc project is a prominent example in his account: begun in the 1980s, it aimed to build a large structured repository of common-sense facts, and after roughly 40 years of work it had reached about 25 million entries. Denning presents this history to argue that even extensive symbolic resources have not produced systems with human-like common sense.

He further frames today’s debates about LLMs and machine-learning scale in this longer trajectory. Denning argues that simply enlarging neural networks and training data does not solve the problem he sees at the core of AI: connecting language to embodied meaning and to cultural and historical knowledge that people rely on when they communicate and act.

Denning traces many of AI’s limitations to what he calls the “representation problem,” arguing that computers can only perform calculations with data and instructions encoded in physical forms they can recognize and process, while tacit knowledge does not naturally fit those forms. He writes: “Behind every word is a deep well of tacit knowledge that gives it meaning. Words are but symbolic representations of meanings, not the meanings themselves. Commonly used Large Language Models, such as ChatGPT, Claude and Gemini only manipulate words, they cannot know or understand the meaning of what they are saying.”

He argues that the difficulty is compounded because scientists cannot fully explain how tacit knowledge operates in humans, limiting any attempt to translate it into machine-usable representations. Denning writes: “How we host tacit knowledge is largely a mystery. All we know is that it is embodied. We have no idea what we might observe and measure in our bodies to reveal it.”

Denning uses practical performance skills to illustrate the gap between describing outcomes and reproducing the embodied process of achieving them. He offers an example from music: “A virtuoso violinist can play beautiful music yet cannot describe to an acolyte how to produce it.” He adds that even a robot that could observe and imitate would still lack key experiential dimensions: “Even if a robot could observe and imitate skilled humans, having no biological body, a robot cannot grasp how the musician feels when playing beautiful music or how an audience feels when hearing it.”

On context, Denning argues that human meaning depends on layered histories of interaction. He writes: “When you inquire into where an assumption of the current context came from, you discover it rests on previous conversations from previous contexts. Each of those in turn rests on further previous conversations and their contexts. This pattern is endless and fractal.”

On culture, he argues that scale alone will not give machines what people acquire through embodied life and shared history. Denning writes: “Human conversations are imbued with background assumptions that give meaning and relevance to the words being used,” and adds: “Scaling up LLMs with ever larger neural networks will not enable them to acquire the embodied human knowledge we call culture. LLMs will not attain the objective of the Turing test: to demonstrate machine thought indistinguishable from human thought.”

Denning concludes that humans and AI systems may develop different forms of tacit knowledge that remain mutually unintelligible. He writes: “Machines cannot read our tacit knowledge and we cannot read theirs. We are aliens across an uncrossable divide.” He argues that this divide complicates AI safety because machines may not be able to interpret the unspoken context behind human intentions, making reliable alignment with human goals difficult.

He warns that “through AI automation, agentic networks of machines are likely to develop their own machine intelligence that does not reach the level of human general intelligence but is still quite capable of creating severe problems for humans,” adding that “this threat is a greater than a take-over by superintelligent machines.” Denning writes that “machine intelligence has different concerns from us and does not appear to care about us,” and that “we do not yet know how to live safely with these machines.” He argues that “pulling back from an AI automation singularity will demand much from us,” urging people to resist becoming “subservient to machines,” refuse “to submit to a yoke imposed by low-intelligence machines,” and “reassert our humanity” by recognizing and valuing what he argues makes humans different from machines.