A new analysis suggests that artificial intelligence may never achieve human-like thinking because the most essential aspects of human intelligence cannot simply be encoded into software. The author, a prominent computer scientist, argues that a proposal by Alan Turing — often considered the father of theoretical computer science — fundamentally misdirected AI research and has shaped the field along a flawed path for the past 75 years.

In his analysis, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, Peter J. Denning reexamines the ideas Alan Turing proposed in 1950, when many researchers believed human intelligence could be separated from the body and reproduced as software on a digital computer. Denning challenges that assumption and also argues that the imitation game, now known as the Turing test, is not a valid measure of genuine machine intelligence.

Figure 1. The Tacit Knowledge Barrier in AI

Denning argues that two assumptions — that intelligence can be separated from the body and that passing the Turing test demonstrates intelligence — have strongly influenced AI research for decades. He contends that accepting these ideas has contributed to the current problems surrounding AI. In his view, today’s AI systems are unlikely to achieve human-level artificial general intelligence (AGI), yet they could still pose significant risks despite never genuinely thinking like humans. Figure 1 shows The Tacit Knowledge Barrier in AI.

The Importance of Tacit Knowledge

A key part of Denning’s argument is the concept of tacit knowledge — the vast body of human understanding that people use naturally but cannot fully describe in words or convert into machine-readable symbols. He identifies five major forms of tacit knowledge that he believes remain beyond the reach of machine learning: common sense, everyday interactions with people and the physical world, emotions and perceptions, hands-on practical skills, and the cultural and historical context shared within societies.

Researchers have long attempted to encode common sense into a form computers can understand. One of the most ambitious efforts was Douglas Lenat’s Cyc project, launched in the 1980s to build a massive database of common-sense knowledge. After four decades, Cyc had accumulated roughly 25 million entries, yet it still failed to provide enough background understanding to make expert systems genuinely expert. Denning argues that the project ultimately demonstrated a deeper limitation: much of the knowledge that makes humans skilled and knowledgeable cannot be fully expressed as explicit facts or propositions.

Knowing Facts Is Not the Same as Knowing How

Denning argues that practical skill presents a major challenge for AI because human expertise involves more than explicit information. While the outcomes of skilled performance (“know what”) can often be described and stored digitally, the embodied knowledge required to perform a skill (“know how”) has not been successfully encoded for machines.

He uses music as an example: a virtuoso violinist may perform brilliantly yet be unable to fully explain the subtle bodily movements, sensations, and judgments that make the performance possible. Even a robot that could observe and imitate a musician would still lack the biological experience of creating music and the emotional understanding of how performers and audiences feel during a performance.

Denning extends this argument to other forms of tacit knowledge — including intuition, gut feelings, spontaneous creativity, and imagination — which he says resist reduction to formal computer instructions.

The Challenge of Representing Human Knowledge

Denning describes the key challenge as the representation problem: computers can only process information that has been encoded into physical symbols and instructions they can recognize. Tacit knowledge, by contrast, cannot easily be translated into such a machine-readable form.

He argues that words are merely symbols that point to meaning, while the meaning itself depends on a vast background of tacit human understanding. For that reason, Denning contends that large language models such as ChatGPT, Claude, and Gemini manipulate patterns in language rather than genuinely understanding the meanings behind the words they produce.

In his view, this creates a fundamental gap between human cognition and machine computation. Because scientists still do not understand how tacit knowledge is embodied in humans, they also do not know how it could be observed, measured, or transferred into a machine. Denning therefore sees the embodiment of tacit knowledge as one of the deepest unresolved mysteries of human intelligence.

Why Meaning Depends on Context

Denning emphasizes that context — the surrounding circumstances, relationships, emotions, and shared history — is essential for giving human language and behavior their meaning. The same sentence can convey completely different intentions depending on whether it is spoken sincerely, sarcastically, angrily, playfully, or teasingly.

Context also enables people to understand humor, exercise tact, interpret tone, and infer what someone chooses not to say. Denning argues that this contextual understanding is not isolated to a single moment; it is built from layers of previous conversations and experiences, each of which depends on earlier contexts in an endless, recursive pattern. He describes this chain of contexts as effectively fractal, with meaning continually drawing on deeper and deeper background assumptions.

Why Culture May Be Beyond AI

Denning argues that culture creates a major obstacle for AI because it consists of far more than language. It includes shared values, social norms, historical experience, emotional tone, community relationships, and subtle dynamics of power, trust, and care. Human conversations rely heavily on these background assumptions, which give words their relevance and meaning.

He contends that simply making large language models larger will not solve this problem, because increasing the size of neural networks does not give machines the embodied human experience from which culture emerges. In his view, scaling LLMs cannot achieve the original goal associated with the Turing test: producing machine thought that is genuinely indistinguishable from human thought.

Denning concludes that humans and AI systems may ultimately possess different and mutually inaccessible forms of tacit knowledge. Machines may develop their own internal patterns of understanding, but humans may be unable to interpret them, just as machines cannot access the embodied tacit knowledge that shapes human culture and cognition.

The Safety Risks of Advanced AI

Denning argues that the gap between human tacit knowledge and machine computation has serious implications for AI safety [1]. If AI systems cannot grasp the unstated context, assumptions, and intentions behind human instructions, then reliably aligning their behavior with human goals may be fundamentally difficult.

He believes the most immediate threat is not a superintelligent AI that exceeds human intelligence, but networks of less intelligent yet highly capable systems that interact in complex, unpredictable, and potentially harmful ways. Such agentic machine networks could develop forms of machine intelligence that are powerful enough to create severe societal problems even without achieving human-level general intelligence.

Denning further argues that machine intelligence may operate according to priorities and modes of reasoning that are fundamentally different from human concerns. Because humans do not yet know how to coexist safely with these systems, he calls for caution about an “AI automation singularity” and urges society to preserve human values, autonomy, and the distinctive qualities that separate human beings from machines.

Reference:

1. https://scitechdaily.com/why-ai-may-never-reach-human-intelligence/

Cite this article:

Janani R (2026), Why Human Intelligence May Remain Beyond AI, AnaTechMaz, pp. 1024