Connoisseurship

In the AI Story

Polanyi introduced connoisseurship to explain how scientific peer review actually functions. The official account presents peer review as the application of explicit methodological standards: reviewers check whether experiments follow protocols, whether data support conclusions, whether reasoning is valid. But Polanyi observed that the most important evaluations scientists make cannot be reduced to these explicit checks. The reviewer exercises judgment about whether the work is significant—whether it advances understanding in a way that matters to the field. This judgment is connoisseurial: it draws on the reviewer's accumulated tacit sense of what the field knows, what it needs to know, what lines of inquiry are promising and what are exhausted. The judgment cannot be formalized. Two competent reviewers examining the same paper may reach opposite conclusions, both grounded in genuine expertise, because their tacit grounds differ.

The AI age creates an acute connoisseurship crisis by flooding every domain with outputs that meet explicit standards while varying wildly in their satisfaction of tacit ones. When Claude Code generates a software implementation, the code may compile cleanly, pass all tests, and execute correctly—meeting every explicit criterion of functionality. But whether the code is well-designed—whether its architecture will scale, whether its abstractions are the right ones, whether its structure will be maintainable five years hence—requires connoisseurial judgment that only deep experience provides. The senior engineer possesses this judgment. The junior does not. The market, evaluating only explicit functionality, cannot distinguish between code grounded in deep connoisseurship and code that happens to work despite shallow understanding. Both ship. Only one rests on foundations that hold.

Educational institutions are discovering that traditional assessment frameworks assumed connoisseurial evaluation that AI makes obsolete. The philosophy professor who reads a student essay evaluates not merely whether the argument is logically valid and evidentially supported (explicit criteria AI-generated essays easily meet) but whether the essay demonstrates genuine philosophical engagement—whether the student has wrestled with ideas, integrated readings into her own thinking, developed a position she could defend under Socratic questioning. This evaluation is tacit, drawing on the professor's decades of reading philosophy, teaching students, distinguishing superficial competence from deep understanding. When students submit AI-generated essays, the professor's connoisseurship is rendered useless—she is evaluating a product that gives no evidence of the personal engagement her tacit standards were developed to detect.

Origin

Polanyi first used the term in Personal Knowledge (1958) and developed it most fully in his later essays on scientific authority and the growth of knowledge. He drew the concept from the traditional use of "connoisseur" in art and wine appreciation, extending it to cover all domains of skilled judgment. The extension was deliberate: Polanyi wanted to show that scientific judgment operates through the same tacit, personal, irreducibly evaluative structure as aesthetic judgment—both require cultivated sensibility, both resist reduction to rules, both depend on the judge's biographical accumulation of engaged experience.

Key Ideas

Evaluation transcends rules. The capacity to distinguish excellent from merely adequate cannot be reduced to criteria—it is a sensibility built through years of engaged attention to a domain's exemplars.

Personal and inarticulate. Two experts' connoisseurial judgments may differ irresolvably because each draws on tacit ground shaped by unique biographical trajectories through the domain.

Built through exposure. Connoisseurship develops through sustained attention to the domain's range of quality—encountering not just good examples but mediocre and terrible ones, building discrimination through comparison.

AI lacks evaluative ground. Pattern-matching produces outputs consistent with training data's statistical regularities but cannot assess whether outputs meet the tacit standards that only embodied expertise applies.

Market cannot detect absence. Explicit metrics measure focal outputs; connoisseurship evaluates subsidiary quality invisible to measurement—the depth, coherence, and integrity that determine whether competent work is also excellent work.