Collective Tacit Knowledge

In the AI Story

Collins arrived at this concept through his long immersion in the gravitational wave physics community, where he observed that the knowledge required to operate a detector could not be fully transmitted through publications or even through technical training. It was transmitted through apprenticeship — through the specific social relationship in which an experienced physicist conveyed, often without being able to articulate, the community's shared sense of what counted as a real signal, what anomalies could be safely ignored, and what standards the community actually held itself to as opposed to what it claimed in its methodological statements.

The concept has direct consequences for AI evaluation. When a large language model is trained on the textual output of a community of practice, it absorbs the community's explicit knowledge in its entirety. What it cannot absorb — because it was never written down — is the collective tacit knowledge through which the explicit knowledge was produced and is maintained. The model has the text. It does not have the social life from which the text emerged and in which its meaning is sustained.

In Collins and Thorne's 2026 paper, the authors test this directly. They examined how gravitational wave physicists decided to ignore a specific fringe science paper — a decision that required the community's collective tacit knowledge about credibility, plausibility, and the implicit standards of what counts as a paper worth engaging. The physicists could reproduce the reasoning. The language model could not. It produced plausible-sounding arguments, but the arguments lacked the specific social grounding that gave the physicists' judgment its authority.

The relationship to the apprenticeship problem is direct. Collective tacit knowledge is transmitted through the social processes of apprenticeship — the slow, friction-rich relationship in which newcomers absorb community norms by participating in community practice. When the practices that sustain the community change faster than the collective tacit knowledge can adapt, or when the apprenticeship structures are bypassed by tools that produce competent-looking output without requiring participation, the knowledge thins. It does not disappear immediately. It fails to transmit to the next generation.

Origin

Collins developed the three-species taxonomy most systematically in Tacit and Explicit Knowledge (2010), building on Michael Polanyi's foundational insight that 'we know more than we can tell.' The key move was to subject Polanyi's concept to sociological analysis — to distinguish knowledge that is tacit for contingent reasons (relational), knowledge that is tacit because the body has its own intelligence (somatic), and knowledge that is tacit because it exists only in social practice (collective). The third species was the one Collins identified as most consequential for the AI question.

Key Ideas

Irreducibly social. Collective tacit knowledge is not stored in any individual mind or any document. It exists only in the ongoing practices of the community that maintains it.

Transmitted through participation, not instruction. The knowledge passes from experienced to inexperienced practitioners through the social relationship of apprenticeship, not through explicit teaching.

Unavailable to systems trained on text. Training data captures the community's explicit output, not the social practices through which that output was produced.

Lost when practices dissolve. When the community that sustains the knowledge dissolves or is bypassed, the knowledge does not persist in archives. It disappears.