In his book Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, prominent computer scientist Peter J. Denning contends that two core assumptions made by Alan Turing in 1950 continue to shape AI research and development today. The first assumption holds that intelligence can exist independently of physical embodiment and be replicated in software. The second posits that machines can demonstrate intelligence by successfully imitating human conversation—a principle that became known as the Turing test. Denning argues these foundational claims have steered AI development in problematic directions.
Denning’s central argument focuses on tacit knowledge, the vast reservoir of human understanding that cannot easily be translated into words or computer-processable formats. He identifies five categories that machines cannot capture: common sense, everyday interactions with people and environments, emotions and perception, practical performance skills, and the social and historical knowledge embedded in culture. Denning notes that even ambitious projects like Douglas Lenat’s Cyc initiative, which accumulated roughly 25 million entries of common sense facts over four decades, failed to produce sufficiently intelligent expert systems. He emphasizes that embodied knowledge—such as how a virtuoso violinist produces beautiful music—cannot be encoded and transmitted to machines lacking biological experience.
Denning identifies what he terms the “representation problem” as the fundamental barrier. Computers operate exclusively through encoded data and instructions in physically recognizable forms, while tacit knowledge does not naturally fit this framework. He argues that large language models like ChatGPT, Claude, and Gemini can only manipulate words as symbols; they cannot grasp the meanings those words represent. Since scientists remain unable to fully explain how human bodies host tacit knowledge, translating it into machine-usable form remains impossible.
Context and culture present additional obstacles that prevent machines from achieving human-level intelligence. Intelligence depends heavily on contextual awareness that allows recognition of sarcasm, humor, emotion, and appropriate social behavior. Culture encompasses values, norms, history, communities, and relationships involving power and care—all background assumptions that give meaning to human communication. Denning concludes that scaling up neural networks will not enable machines to acquire this embodied cultural knowledge.
Denning raises significant safety concerns about this knowledge gap. If advanced AI systems cannot interpret the unspoken context behind human intentions, aligning them reliably with human goals may prove impossible. He suggests that machine intelligence operating with different concerns and logic may develop capabilities to create severe problems for humans without reaching general intelligence levels. Denning advocates for humans to reassert their humanity and decline to become subservient to machine systems.
Originally reported by ScienceDaily. Read the full story →