Knowing vs Recognizing

In the AI Story

Every major discovery Pasteur made followed the same sequence: perceptual recognition first, propositional knowledge second. The tartaric acid crystals — he recognized the geometric asymmetry before he could name molecular chirality. The Lille fermentations — he recognized the organisms as causal agents before he could design the experiments proving it. The old cultures — he recognized that they conferred immunity before he understood attenuation. In every case, recognition was generative; propositional knowledge was the product.

Large language models reverse this ordering. They begin with the corpus of published knowledge and produce outputs that are combinatorial rearrangements of existing propositions. The outputs can be sophisticated; they can identify patterns no individual researcher would have linked. But the cognitive pathway is different, and the difference matters because pathways produce different downstream capacities. The prepared mind possesses what the system does not: a felt sense of where an observation leads, what questions it opens, what experiments it demands.

This directionality is not computed from data. It emerges from the prepared mind's landscape of accumulated experience, which provides the topographic context in which a new observation finds its position and vector. The observation sits in a specific place, and the landscape's contours suggest the direction of further investigation the way a valley suggests the direction of water flow.

Origin

The distinction is implicit throughout Pasteur's experimental writings but articulated most sharply in the book's fifth chapter. It draws on Michael Polanyi's tacit knowledge, Hubert Dreyfus's critique of symbolic AI, and Gilbert Ryle's distinction between knowing how and knowing that.

Contemporary cognitive science has given the distinction empirical grounding. Gary Klein's work on naturalistic decision-making identifies recognition-primed decision as the signature operation of expert practice — a perceptual event operating on thousands of accumulated patterns, not propositional retrieval.

Key Ideas

Propositional vs perceptual. Knowing takes the form of sentences; recognizing takes the form of events. Both are real; they are categorically different operations.

The generative sequence. In every major Pasteurian discovery, perceptual recognition came first and propositional knowledge followed. The recognition was the engine; the knowledge was the product.

AI systems reverse the sequence. They begin with propositions and recombine them. This produces sophisticated output but not recognition.

Directionality is felt, not computed. The prepared mind knows where an anomaly leads because it sits in a topographic landscape of experience. Systems operate in flat data-space without this context.

The failure mode. The substitution of comprehensive detection for genuine recognition produces practice that is efficient at identifying known patterns and impoverished at recognizing what falls outside them.

Debates & Critiques

The distinction is contested by strong-AI advocates who argue that sufficiently scaled pattern recognition becomes functionally indistinguishable from recognition. The book's position aligns with Harry Collins's framework: interactional expertise without contributory expertise, competent surface without generative depth.