The Turing Mirage: Has 75 Years of AI Research Been Built on a Fundamental Error?

For three-quarters of a century, the trajectory of artificial intelligence has been guided by the visionary, yet potentially flawed, blueprints of Alan Turing. Since his seminal 1950 paper, Computing Machinery and Intelligence, the quest to replicate human-level cognition in silicon has served as the North Star for computer science. However, according to Peter J. Denning, a distinguished computer scientist and author of the new book Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, this pursuit may have steered the entire field into a technological cul-de-sac.

Denning’s critique is not merely a technical disagreement; it is a fundamental challenge to the philosophical foundations of AI. He argues that our obsession with the "Turing Test" and the dream of disembodied machine intelligence has blinded us to the true nature of human cognition, leading us into an "AI mess" characterized by systemic risks, broken promises, and an unbridgeable ontological divide between human understanding and machine processing.

The Foundations of a False Promise

To understand Denning’s thesis, one must examine the two pillars upon which modern AI research rests. The first is the assumption of "disembodied intelligence"—the belief that cognitive processes are essentially software routines that can be extracted from a biological host and run on any sufficiently powerful hardware. The second is the Turing Test itself: the notion that if a machine can successfully mimic human conversation, it must, by definition, be intelligent.

"These two claims have shaped much of AI research and development," Denning writes. "My premise is that our acquiescence to these claims has led to the AI mess in which we find ourselves today."

By treating intelligence as a series of data-processing tasks rather than an embodied, contextual experience, researchers have prioritized the appearance of intelligence over its substance. This has resulted in the rapid proliferation of Large Language Models (LLMs) that excel at linguistic pattern matching but lack the underlying "knowing" that defines human expertise and awareness.

A Chronology of the "Representation Problem"

The history of AI is essentially a 75-year effort to solve what Denning terms the "Representation Problem."

• 1950: Alan Turing publishes his seminal paper, proposing the "Imitation Game." The field of AI is born under the assumption that simulation equals reality.
• 1960s–1970s: The era of Symbolic AI (GOFAI). Researchers attempt to encode human logic into formal rules. This approach fails to account for the nuance and ambiguity of natural language.
• 1980s: The rise of Expert Systems and the Cyc project. Led by Douglas Lenat, Cyc sought to manually input "common sense" into a database. After 40 years and 25 million entries, the project proved that common sense is not a list of facts, but an implicit, fluid framework.
• 2010s–Present: The Deep Learning revolution. Using massive neural networks and statistical probability, models like ChatGPT and Gemini achieve unprecedented linguistic fluidity, leading to a new wave of belief in the imminence of Artificial General Intelligence (AGI).

Denning posits that each era has merely shifted the goalposts, moving from logic-based rules to statistical probabilities, yet all have failed to bridge the gap between "knowing what" (information) and "knowing how" (embodied skill).

The Tacit Knowledge Barrier

At the heart of Denning’s argument lies the concept of tacit knowledge—the vast, unarticulated reservoir of understanding that guides human behavior. Denning categorizes this into five domains: common sense, everyday interactions, emotions and perception, practical performance skills, and cultural/historical context.

The Failure of Formalization

The Cyc project serves as the primary historical evidence for Denning’s critique. If common sense could be distilled into bits, Cyc would have succeeded. Instead, it demonstrated that human expertise is not a collection of propositions but an intuitive background.

"Cyc validated that much of the knowledge that makes people experts cannot be articulated as propositions," Denning notes. This is particularly evident in high-level physical or creative skills. A concert violinist, for instance, cannot fully document the precise neurological and muscular coordination required to evoke a specific emotional response from an audience. The knowledge is "embodied"—it exists within the interaction of the nervous system, the instrument, and the environment.

The Limits of Large Language Models

Denning’s critique of current AI technology is stark: LLMs do not "understand." They are sophisticated symbol manipulators. "Behind every word is a deep well of tacit knowledge that gives it meaning," he argues. "Words are but symbolic representations of meanings, not the meanings themselves."

Because LLMs lack a body, they lack the lived experience that generates meaning. They inhabit a mathematical hyperspace, not a physical world. Consequently, they can simulate the syntax of human thought while remaining fundamentally alien to its semantic depth.

Context, Culture, and the Fractal Nature of Meaning

Intelligence is not an isolated phenomenon; it is deeply embedded in context. Human communication relies on an endless, fractal web of previous conversations, shared history, and cultural norms.

Denning emphasizes that culture—the values, power dynamics, and historical narratives that define a community—cannot be "learned" by an LLM through increased parameter scaling. When a human interprets a sarcastic remark, they are not just parsing words; they are navigating a complex social landscape of intent, relationship, and mood.

"Scaling up LLMs with ever larger neural networks will not enable them to acquire the embodied human knowledge we call culture," Denning insists. This creates a ceiling for AI development: it can optimize for the "Turing Test" of surface-level imitation, but it can never achieve the contextual awareness that makes human thought uniquely adaptive and purposeful.

Implications for AI Safety and Human Agency

The most concerning implication of Denning’s work is not the threat of a malicious "superintelligence" taking over the world in a science-fiction scenario. Rather, it is the threat of an "alien" intelligence that is neither superintelligent nor human-aligned.

The "Uncrossable Divide"

Denning suggests that we are headed toward a future where human-made machines develop their own, inaccessible forms of tacit knowledge. Because these systems operate on logics derived from high-dimensional statistical probability rather than biological survival and social interaction, their goals and behaviors will remain opaque to us.

"We are aliens across an uncrossable divide," Denning writes. This creates a critical safety failure: if we cannot understand the "why" behind an AI’s decision, we cannot effectively align it with human values. We are effectively deploying systems that are fundamentally "unreadable" to their creators.

The Path Forward: Reclaiming Humanity

Denning concludes with a call to action that is as much philosophical as it is technical. He warns against the "AI automation singularity"—a trajectory where we voluntarily outsource our judgment to systems that do not share our concerns or our morality.

To avoid this, he advocates for a radical shift in perspective:

1. Stop the worship of AGI: Accept that human-level general intelligence for machines is likely a biological impossibility, and pivot research toward human-augmented tools rather than autonomous agents.
2. Recognize the "Yoke": Acknowledge that the drive to make machines "human-like" is the source of our current systemic instability.
3. Celebrate the Difference: Reassert the unique value of human embodied intelligence—our intuition, our gut feelings, and our capacity for spontaneous, context-aware creativity.

Conclusion: A New Era of Human-Machine Relations

The debate sparked by Peter J. Denning is a necessary corrective to the hype cycle of the 2020s. By challenging the foundational assumptions of Alan Turing, Denning invites us to consider that our greatest mistake may have been attempting to build machines in our own image.

As we stand at a crossroads in the development of artificial intelligence, the choice is not merely between more or less technology. It is a choice about whether we will continue to chase the mirage of a digital mirror or choose to invest in a future where technology remains a tool for humans, rather than a system that defines our humanity. Denning’s Turing’s Mistake serves as a stark reminder: we cannot automate our way to understanding, and in our rush to build intelligent machines, we risk losing sight of the very things that make us intelligent in the first place.