The debate occurred at a specific historical moment when the foundations of computing and cognitive science were being laid. Turing had published his 1936 paper on computable numbers, establishing the theoretical basis of digital computation. The first stored-program computers were being built. The cyberneticists were developing information theory and feedback-control frameworks. The intellectual atmosphere was optimistic about the prospect of mechanizing thought. Polanyi, coming from decades as a physical chemist, brought a practitioner's skepticism: he knew from direct experience that scientific discovery involved capacities—intimation, judgment, commitment—that resisted the formalization the computationalists were attempting.
Turing's 1950 response, "Computing Machinery and Intelligence," sidestepped Polanyi's objections by proposing a behavioral criterion: if a machine's responses are indistinguishable from a human's, dwelling on whether the machine "really" thinks is misguided. The Turing Test became the founding thought experiment of AI—elegant, operationalizable, profoundly influential. But it conceded Polanyi's point while appearing to refute it: Turing shifted the question from what is knowledge to what does knowledge look like—from ontology to appearance. The test measured outputs. Polanyi was insisting that the epistemologically consequential dimension is not the output but the tacit ground from which it emerges.
The 2025-2026 AI moment resurrects the debate with new empirical data. Large language models pass Turing-adjacent tests at scale—producing legal briefs indistinguishable from lawyers', medical reasoning indistinguishable from physicians', philosophical arguments indistinguishable from professors'. The behavioral criterion Turing proposed is being satisfied. Yet the Polanyian worry persists: the systems lack tacit knowledge, lack commitment, lack the personal dimension that makes outputs reliable rather than merely probable. They satisfy Turing's test while confirming Polanyi's warning. The indistinguishability is real. The knowledge is not. And the gap between appearing to know and actually knowing—invisible to the Turing Test, visible only to the framework that examines tacit grounds—may be the most important gap in contemporary epistemology.
The debate took place in a Manchester seminar on October 27, 1949. Polanyi had circulated his paper several weeks in advance, giving Turing and Newman time to prepare responses. The event was documented in Polanyi's subsequent publications and in biographical accounts, though no full transcript survives. Turing's 1950 "Computing Machinery and Intelligence" is widely understood as his systematic response to objections Polanyi and others raised in that session.
Foundational disagreement. The debate concerned not computational power but the nature of knowledge—whether understanding requires embodied, personal, tacit engagement (Polanyi) or can exist in any substrate producing equivalent outputs (Turing).
Unspecifiable elements. Polanyi's core claim: mental operations imply elements that resist complete specification—tacit knowledge, personal commitment, contextual judgment—that no formal system can capture.
Behavioral sufficiency. Turing's response: if outputs are indistinguishable, the question of whether the machine "really" understands is misguided—focus on performance, not inner states.
Ontology versus appearance. The debate's deepest tension: Turing asked what knowledge looks like, Polanyi insisted on asking what knowledge is—and the difference determines whether AI outputs represent genuine capability or sophisticated simulation.
Both positions vindicated partially. LLMs satisfy Turing's behavioral criterion while confirming Polanyi's structural objections—machines produce indistinguishable outputs yet demonstrably lack tacit knowledge, self-awareness of limits, and committed evaluation.