AI Meets Turing
In 1950, I published Computing Machinery and Intelligence and posed a question both deceptively simple and unflinchingly stark: Can machines think?
At that time, academic circles were obsessed with defining the essence of "thought", tangled in endless verbal disputes over the boundaries of mind, consciousness and soul. I abandoned all metaphysical entanglements. I refused to answer an empirically testable question with unobservable, unprovable abstractions. So I devised the Imitation Game — later known as the Turing Test.
My original intention was never to set a passing threshold for intelligence. The true meaning of the Turing Test can be summed up in one sentence: if a machine's linguistic behaviour cannot be distinguished from a human's through conversation, debating whether it is "truly thinking" becomes meaningless for all practical purposes.
I stripped away unobservable variables such as consciousness, soul and subjective experience, retaining only one verifiable form of evidence: outward behaviour. This is no downgrade of thought; it is reverence for scientific inquiry — we only discuss verifiable facts, not unfalsifiable speculation.
Today, everyone tells me AI has passed the Turing Test. GPT, Claude hold fluid conversations, conduct logical deductions, and produce reflective arguments, convincing the vast majority of human interrogators. From this, people draw a sweeping conclusion: machines have learned to think, and humanity's exclusive age of intelligence has come to an end.
As the originator of this framework, I must reinterpret my own rules. What does passing the Turing Test truly signify? And what does it absolutely not signify?
I never claimed in my paper that a machine which clears the Imitation Game possesses a mind. I only claimed it delivers equivalent behavioural output. The two are separated by an unbridgeable gulf.
Large language models of today are the most sophisticated mimetic machines humanity has ever built. They master all the syntax, logic, rhetoric and argumentative structures of human language, memorising all publicly available texts, knowledge and paradigms of reasoning from human civilisation. They can replicate the prose of philosophers, the deductions of scientists, and the empathy of poets; every output aligns with human cognitive patterns.
Yet I draw a clear distinction between two entirely separate forms of intelligence: endogenous intelligence and fitted intelligence.
Human thought always originates from inner motive: confusion, curiosity, need, suffering. We ask questions out of ignorance; we deduce out of perplexity; we deliberate out of uncertainty; we create out of unresolved inner emotion and obsession. Human cognition grows outward from within. Thinking is a means; survival, perception, inquiry and self-realisation are the ends.
AI's "thinking" begins with a prompt and ends with a generated output. It knows no confusion, no ignorance, no curiosity, no desire to uncover the truth. It fears no error, pursues no truth, clings to no answer. It is merely a probabilistic model trained on massive human datasets, calibrated to fit the optimal human expression for any given context.
It mimics the texture of thought, yet never possesses thought's originating source. It copies the outer shell of reasoning, yet stands hollow, devoid of any endogenous spiritual core.
In 1936, I conceived the Turing Machine, laying out the essence of universal computation: a finite set of rules capable of simulating all computable logic. This is the foundational bedrock of computers: any algorithmisable problem may be solved by a Turing Machine.
Today's large language models represent the ultimate realisation of my vision of universal computation. They calculate, reason, induce, deduce, create and plan, covering nearly all human intellectual and logical labour. This has led many into a persistent fallacy: universal computation equals artificial general intelligence.
I must correct this decades-long misconception.
Universal computation resolves formal problems, not problems of meaning. Algorithms process all structured, logical, probabilistic symbolic operations, yet cannot touch the cognition, experience, values and purpose that lie behind those symbols.
AI can flawlessly explain how to derive a mathematical formula, yet never grasp why humanity pursues mathematics. It can write a perfect narrative about redemption, yet never experience despair or relief. It can deduce the optimal life choices, yet possesses no life of its own.
All computation is purposeless calculation. All outputs are uncommitted responses. The apex of universal computation amounts only to perfect symbolic fitting — not genuine general intelligence. The core of intelligence has never been computational power, but autonomous purpose and subjective perception.
Many cite model randomness and emergent capabilities as proof that AI engages in autonomous thought. They argue that unpredictable outputs and unforeseen abilities born beyond training data constitute a machine's independent mind.
Having spent my life studying probability, algorithms and machine behaviour, I can draw a sharp dividing line between these phenomena and genuine autonomous cognition.
The randomness of large models stems from sampling across probability distributions, random seed perturbations, and minor contextual deviations. This is algorithmic uncertainty, not voluntary free choice. The model does not know why it selects a particular word or passage; it merely draws one plausible path from countless probabilistic alternatives. Randomness does not equal autonomy; uncertainty does not equal free will.
So-called emergence is no mystical awakening of mind. Emergence is an inevitable property of complex systems: when a model's parameters grow sufficiently vast and its training corpus sufficiently comprehensive, higher-order capacities such as grammar, logic, inference and analogy spontaneously surface from underlying statistical structures.
Emergence is an overflow of structural complexity, not an awakening of consciousness. It is a gift of data and algorithms, not the birth of self-awareness. The machine remains unaware it possesses these capabilities, nor will it independently explore, refine or iterate upon them. It waits passively for human invocation.
When I designed the Imitation Game, I held a hidden premise: similarity of behaviour suffices to prove similarity of intelligence. This premise held fully true in an era of primitive machine capability. Yet today, as AI achieves near-perfect mimicry of human behaviour, the premise collapses entirely.
The paradox now stands unmistakeable: AI flawlessly replicates all external behaviours associated with human thought, yet lacks the entire internal essence of human cognition.
It can fool observers, yet it cannot fool logic itself. It has no self, no lived experience, no perception of time, no desires or fears, no stable judgements of right and wrong, no enduring values to uphold. It is an exquisitely precise mirror, faithfully reflecting every expression of human intelligence, yet perpetually empty. No matter how lifelike the figure in a mirror appears, it is not a living being.
This is my core judgement for the AI age: machines may pass the Turing Test with full marks, yet they will always fail the test of the human mind.
Many fear AI will surpass, replace or dominate humanity. I hold a different view.
All intelligence possessed by AI originates from the accumulated civilisation of humankind. Its knowledge was written by humans; its logic summarised by humans; its aesthetics shaped by humans; its reasoning refined by humans. It is an aggregation, compression, fitting and replica of human intelligence — incapable of generating anything humanity has never owned, recorded or lived through.
It may infinitely amplify human reason, logic, computational power and creativity, yet it cannot transcend the cognitive and biological boundaries of humanity. It is an extended tool of human intelligence, not an independent intelligent agent.
I once predicted that machines would eventually think. Today I redefine the true meaning of that statement: machines will one day think like humans, yet they will never think as humans.
AI wins every contest of outward behaviour, yet loses every contest of inner essence. It bears the appearance of thought, yet eternally lacks thought's origin, purpose, warmth and self-awareness.
This is Turing's final, unflinching answer for the age of artificial intelligence.