News summary produced by Claude AI
Computer scientist Peter J. Denning contends that two key assumptions made by Alan Turing in 1950 have fundamentally shaped artificial intelligence research for the past several decades, potentially steering the field in an unproductive direction. In his book “Turing’s Mistake: Escaping the Yoke of Unintelligent Machines,” Denning challenges Turing’s beliefs that intelligence can exist independently of physical embodiment and that machines can demonstrate intelligence through successful human conversation imitation, which later became known as the Turing test. Denning argues these foundational premises have contributed to what he describes as the current state of AI development and its associated challenges.
Central to Denning’s critique is the concept of tacit knowledge—the extensive human understanding that cannot be easily articulated or converted into computer-processable formats. He identifies five major categories of tacit knowledge beyond machine learning’s reach: common sense reasoning, everyday physical and social interactions, emotional and perceptual understanding, practical performance skills, and culturally embedded knowledge. Denning cites Douglas Lenat’s Cyc project, which spent four decades compiling approximately 25 million common sense entries, as evidence that such knowledge cannot be adequately captured through proposition-based systems. He emphasizes that practical skills present an even greater obstacle, noting that expertise in thousands of domains—such as accomplished musical performance—involves embodied knowledge that cannot be encoded or transmitted to machines lacking biological physicality.
Denning frames these limitations through what he calls the “representation problem,” arguing that computers can only process data and instructions encoded in forms they can physically recognize. Tacit knowledge, by contrast, does not naturally conform to such digital frameworks. He notes that behind every word lies deep tacit knowledge that provides meaning, and that current large language models manipulate symbolic representations without understanding their actual significance. This creates a fundamental divide: since scientists cannot fully explain how tacit knowledge functions in humans, they cannot translate it into machine-compatible formats.
Intelligence also depends heavily on context and culture, factors Denning argues machines cannot adequately replicate. Context enables humans to interpret sarcasm, humor, emotion, and social cues, while culture encompasses values, norms, history, and relationships that give conversations meaning and relevance. Denning explains that assumptions underlying human communication rest on endless chains of previous conversations and cultural contexts in a fractal pattern that machines cannot access. He concludes that scaling up neural networks will not enable language models to acquire the embodied human knowledge constituting culture.
Denning warns that the gap between human and machine understanding raises significant safety concerns. Rather than fearing superintelligent machine takeovers, he suggests the greater threat comes from networks of less sophisticated machines developing their own forms of intelligence with different priorities that neither understand nor care about human interests. He argues that humans and machines may ultimately operate as “aliens across an uncrossable divide” and calls for society to reassert human values and declare what makes humanity distinct from machines.