The Vademecum Problem

In the AI Story

The specific conditions of the AI moment exacerbate the problem. The technology itself evolves so rapidly that any understanding encoded in institutional structures is likely to be outgrown by the technology before the structures can be revised. Regulatory frameworks enacted today govern tools that will be substantially different by the time the regulations take effect. Curricula designed this year prepare students for a landscape that will have shifted by graduation. Corporate strategies adopted this quarter presuppose capabilities that next quarter's models may render obsolete or dramatically expand.

The orange-pilled thought collective, because it includes many of the people making institutional decisions about AI, exercises disproportionate influence over which understandings are stabilized. The builders' perception — that the transition is qualitative, that the tools represent permanent shift, that productivity gains are real and transformation irreversible — is being encoded into institutional structures at a speed reflecting the collective's institutional power rather than the maturity of the underlying knowledge.

This is not to say the builders' perception is wrong. Fleck's framework does not adjudicate between competing perceptions. It describes the process through which perceptions become facts — and the current process exhibits the hallmarks of premature stabilization. The stabilization is occurring before competing collectives — critics, elegists, educators, workers who will bear the costs — have had adequate opportunity to contribute their perceptions to the negotiation.

Operating in a permanently provisional epistemic environment requires institutional forms designed for revisability rather than stability. This is easier to prescribe than practice. Institutions are not designed for provisionality. They are designed for reliable execution of established procedures. Provisionality undermines authority, creates stakeholder uncertainty, and requires monitoring resources that compete with execution resources. But the alternative — premature stabilization of understandings built on shifting ground — is worse.

Origin

The concept is developed implicitly through Fleck's 1935 analysis of how syphilis concepts migrated from experimental journals into clinical handbooks, and made explicit in his later essays on medical epistemology. Its application to AI is a natural extension Fleck himself could not have anticipated but whose framework handles cleanly.

Key Ideas

Institutional commitment outpacing epistemic maturation. The core mechanism — resources flowing faster than understanding can stabilize.

Investment as self-reinforcing argument. Institutional stakes shape subsequent evidence evaluation to favor confirmation.

AI's structural acceleration. The technology evolves faster than institutions can absorb any stable understanding, creating permanent journal conditions.

Disproportionate collective power. The builder collective's institutional representation shapes which understandings get encoded in structure.

Revisable institutions as response. The solution is not refusing commitment but designing commitments that preserve the apparatus for revision.

Debates & Critiques

A deep question is whether the pace of AI evolution makes all institutional encoding of AI understanding pathological, or whether some forms of encoding — those with built-in sunset clauses, monitoring mechanisms, and revision procedures — can be epistemically legitimate. The answer has direct governance implications: if the former, current regulatory efforts are structurally inadequate; if the latter, the question becomes how to design institutions for the permanent provisionality Fleck's framework implies.