Can Submarines Swim?

In the language of modern cognitive science, few metaphors are as entrenched (or as quietly catastrophic) as the claim that the brain is a computer. We speak of neural bandwidth, memory encoding, and cognitive processing with such fluency that the analogy has long since hardened into dogma. Yet this framing harbors a fundamental category error, one that the legendary computer scientist Edsger Dijkstra dismantled with characteristic economy: “The question of whether a computer can think is no more interesting than the question of whether a submarine can swim”

The submarine does not swim. A fish swims. It undulates through water with muscle and bone, in an act of locomotion that is inseparable from what the fish is. A submarine navigates the same medium through engineered propulsion. It achieves the outcome of underwater travel without engaging in anything remotely like swimming. It simulates the result while violating the essence. And this is precisely the trap into which the computational theory of mind has fallen: mistaking the map for the territory, the output for the act, the answer for the understanding.

To see why this matters, we should begin with Alan Turing, who is routinely misappropriated as the patron saint of artificial minds that we have nearly forgotten what he actually argued. In his 1950 paper Computing Machinery and Intelligence, Turing opened with a deliberate provocation: “I propose to consider the question, ‘Can machines think?'” But almost immediately he pulled back, warning that the words “machine” and “think” were so loaded with ordinary usage that any direct answer would be meaningless without first interrogating the definitions themselves. His famous imitation game was not, as it is so often presented, a proof that machines can think. It was a proposal to sidestep the question, to replace an unanswerable philosophical puzzle with a behavioral one. Turing proposed equivalence of output as a more tractable thing to study.

The distinction matters enormously, because what his foundational work in computability theory actually established is that computation is purely syntactic, that is, to compute is to manipulate symbols according to their shapes and positions, without regard to what those symbols may mean. A Turing Machine does not know what it is doing. It follows rules. This is not a limitation of the machine that can be overcome with new technology; it is the definition of computation.

John Searle’s thought experiment, the Chinese Room, is demonstrates the same: a man in a room following rules (deterministic, probablistic, or as an outcome of a quantum process) for manipulating Chinese characters he cannot read does not thereby learn Chinese. The submarine remains a submarine regardless of whether its movement is similar to that of a fish.

The seduction of the brain-as-computer metaphor is understandable. However, when we say the brain computes, we are not merely describing but also prescribing. We are deciding in advance what kind of thing the brain is allowed to be, and in doing so we foreclose the very questions that matter most. Consciousness, meaning, creativity, valuation, none of these fit inside the computational frame, and so the computational frame quietly excludes them treating them as epiphenomena or future engineering problems rather than as the central facts of mental life. This is not a scientific conclusion. It is a metaphysical choice masquerading as science.

To think, at its most essential, is to move into territory that no prior rule could have mapped, to assign meaning where no metric exists, to make a leap that syntax alone could never specify. As we build our extraordinary machines, the most important thing we can preserve is the lucidity to know what they are and what they are not.