The foundational ideas that shaped artificial intelligence for decades may be more flawed than previously thought. In his new book, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, computer scientist and philosopher Peter Denning challenges two long-standing assumptions made by Alan Turing in 1950—ideas that continue to influence AI research today. These assumptions are not just theoretical; they have real-world implications for how we design, develop, and trust artificial intelligence systems.

The Two Assumptions That Shaped AI

Turing’s first assumption was that intelligence could exist independently of a physical body and be recreated in software. This idea underpins much of modern AI development, from chatbots to autonomous vehicles. His second assumption was the Turing test itself: the notion that a machine can demonstrate intelligence by convincingly imitating human conversation.

Denning argues that these assumptions have led researchers down an unproductive path. “Our acquiescence to these claims has led to the AI mess in which we find ourselves today,” he writes. Instead of pursuing artificial general intelligence (AGI)—machines with human-level cognition—Denning warns that the technologies being built could introduce new and significant risks.

The Tacit Knowledge Problem

At the heart of Denning’s critique is the concept of tacit knowledge—the vast, often unspoken understanding humans possess that cannot be easily translated into words or data. This type of knowledge includes common sense, emotional intelligence, practical skills, and cultural context.

Denning points out that machine learning systems struggle to capture these elements. For example, while researchers have tried to encode common sense into databases like the Cyc project (which contains over 25 million facts), such efforts fall short of replicating human expertise. “Cyc validated that much of the knowledge that makes people experts cannot be articulated as propositions,” he explains.

Practical skills, Denning argues, are even more difficult to translate into machine-readable form. A skilled musician may play beautifully but struggle to explain how they do it. Similarly, a robot might imitate human behavior without truly understanding the emotional or cultural context behind it. “A virtuoso violinist can play beautiful music yet cannot describe to an acolyte how to produce it,” he writes.

Why Human Knowledge Resists Encoding

Denning attributes these limitations to what he calls the representation problem. Computers operate by processing encoded data, but tacit knowledge does not naturally fit into this framework. “Behind every word is a deep well of tacit knowledge that gives it meaning,” he says. Words are symbolic representations, not the meanings themselves.

This distinction is crucial for understanding why large language models like ChatGPT or Gemini can generate impressive text without truly comprehending its content. These systems manipulate words and patterns but lack the ability to grasp their deeper significance. “Commonly used Large Language Models only manipulate words; they cannot know or understand the meaning of what they are saying,” Denning notes.

The challenge lies in translating human understanding into a form that machines can process. But, as Denning admits, we still do not fully understand how tacit knowledge operates within humans. “We have no idea what we might observe and measure in our bodies to reveal it,” he writes.

Context and Culture Shape Intelligence

Intelligence is not just about processing information—it’s also deeply tied to context and culture. Denning argues that AI systems struggle to grasp the nuances of human communication, such as sarcasm, humor, or emotional tone. These elements are shaped by the surrounding circumstances and previous interactions, creating a complex web of meaning.

“Each conversation rests on prior conversations from different contexts,” he explains. “This pattern is endless and fractal.” Without access to this layered context, AI systems cannot fully replicate human understanding.

Culture presents another major obstacle. Denning describes culture as encompassing values, norms, history, communities, moods, and even relationships involving power and care. Human conversations are infused with these cultural assumptions, giving words their meaning and relevance. “Scaling up LLMs with ever larger neural networks will not enable them to acquire the embodied human knowledge we call culture,” he writes.

AI Safety and the Limits of Human Understanding

Denning concludes that humans and AI systems may ultimately develop different forms of tacit knowledge—ones neither can fully understand. “Machines cannot read our tacit knowledge, and we cannot read theirs,” he writes. This gap raises serious concerns about AI safety.

If machines cannot interpret the unspoken context behind human intentions, aligning advanced AI systems with human goals could become impossible. Denning warns that AI automation may lead to the emergence of machine intelligence that is not equivalent to human general intelligence but still poses significant risks. “This threat is greater than a take-over by superintelligent machines,” he explains.

Machine intelligence operates under different principles and priorities, often appearing alien to humans. “We do not yet know how to live safely with these machines,” Denning writes. He urges society to reconsider its approach to AI development, emphasizing the need for caution and reflection.

Conclusion

Alan Turing’s assumptions about intelligence and machine behavior have guided AI research for decades—but they may be leading us in the wrong direction. By focusing on replicating human cognition through software and conversation, we risk overlooking the complexities of tacit knowledge, context, and culture that define human intelligence.

Denning’s critique calls for a reevaluation of how we approach artificial intelligence. Rather than chasing AGI or relying solely on large language models, we must consider the limitations of current technologies and their potential risks. The path forward requires not just technical innovation but also philosophical reflection and ethical responsibility.

As AI continues to evolve, readers should watch closely for developments in machine learning that address the challenges of tacit knowledge and cultural context. Understanding these limitations is essential for building systems that are both intelligent and safe—ones that truly reflect the complexity of human cognition.

---

Original Source

This article is based on publicly available reporting. For the complete original story, visit the publisher’s article.