Institutional Memory or Institutional Confabulation?

AA new hire asks why a particular approval step exists, and a colleague who's been there eight years gives a confident, detailed, entirely plausible answer involving a vendor dispute from before either of them joined. The story holds together, fits the shape of the process it explains, and gets repeated to the next new hire almost unchanged. It also turns out, when anyone actually checks the old records, to be wrong — the step was added for a different reason entirely, and the vendor-dispute story was assembled, plausibly but incorrectly, from a half-remembered fragment and the need to give the new hire a satisfying answer on the spot. Nobody lied. This is its own phenomenon, not institutional gossip.

What Confabulation Actually Is FoundationalKnowledge that endures for decades — core principles

WilliamWilliam Hirstein's account of confabulation in human cognition is precise about what's actually failing. A confabulated account isn't a lie, because the person producing it believes it; what's missing is a normal checking process that would otherwise catch a plausible-sounding answer before it's treated as true¹. Hirstein's key move is separating two capacities that are usually bundled together and can fail independently: the creative ability to construct a plausible answer to a question, and the separate ability to check that answer against what's actually known before asserting it. Confabulation is what happens when the first capacity runs and the second doesn't.

Institutional Folklore as Confabulation FoundationalKnowledge that endures for decades — core principles

KarlKarl Weick's account of organisational sensemaking supplies the institutional version of exactly this gap. Weick lists plausibility, not accuracy, as one of sensemaking's defining properties: people construct a story that's coherent, memorable, and fits the available cues well enough to act on, and that bar is met far more often than the bar of historical accuracy is even checked against². An organisation's explanation for its own rule — the vendor-dispute story above — is a textbook case: it's coherent, it fits the shape of the process, it resonates with what everyone already expects an approval step's origin story to sound like, and none of that bears on whether it's actually what happened. The story survives and spreads because it's useful for explaining the present, not because anyone checked it against the past.

The LLM Parallel, Precisely Stated FoundationalKnowledge that endures for decades — core principles

ThisThis site's page on understanding large language models makes the equivalent point about machine hallucination: nothing in next-token prediction distinguishes a plausible continuation from a true statement, so confident, fluent, false output is a structural property of the technology rather than an occasional bug. The parallel with institutional folklore is real but needs stating precisely, not loosely. The underlying mechanisms are different — Hirstein's confabulation is a failure of a specific neural checking process; a language model's hallucination is the absence of any ground-truth check in its generation process at all, not the failure of one that's otherwise present. What the two genuinely share is the surface signature: fluency and confidence carry no information about accuracy in either case, which is exactly why "it sounded right, and everyone already believed it" should never, on its own, count as evidence that an institutional explanation is historically correct.

Strongest Objection: Not All Folklore Is False, and Provenance Is Sometimes Contested FoundationalKnowledge that endures for decades — core principles

NoneNone of this means every unverified institutional story is wrong — plenty of folklore turns out to be accurate once checked, and treating all of it as suspect by default would be its own kind of overclaiming. The actual problem Hirstein's framework identifies isn't repetition, it's the absence of a check: a true story told without ever being verified against a record is, at the moment of telling, in exactly the same epistemic position as a false one, because nothing available at that moment distinguishes the two. There's a second complication here, left open rather than resolved: who gets to decide what counts as the "real," checked version of an organisation's history is not always a neutral question, and whoever currently holds the authority to declare one account official can shape what the record itself preserves — a later page in this part takes that problem on directly.

Provisional Conclusion FoundationalKnowledge that endures for decades — core principles

InstitutionalInstitutional memory and institutional confabulation are not two different things happening in different organisations — they're the same sensemaking process, with and without a working check on the output. This cluster's page on learning organisations already names the relevant check by a different name: double-loop learning, which questions the governing assumption rather than just patching the visible symptom, is exactly the kind of process that would catch a confabulated institutional story before it hardens into an unexamined, widely repeated account of "why we do it this way." An organisation without that check isn't remembering its own history. It's generating a plausible-sounding account of it, fresh, every time someone asks.

Questions for Further Thought

• How would an organisation actually audit its own folklore for accuracy without the audit itself becoming just another plausible, unchecked story about "what the records really show"?
• Is there a reliable behavioural signal — the way fluency and confidence turn out not to be reliable signals — that distinguishes a genuinely checked institutional account from a confabulated one?
• Does a written record protect against this, or can a document itself become the confabulated story, repeated and trusted because it's written down rather than because anyone checked it against anything earlier?

Related Topics

• The Half-Life of Organisational Knowledge — what happens to the checkable record itself once the person who could verify a story against it has left.
• Is a "Learning Organisation" a Real Claim or a Compliment? — double-loop learning as the checking mechanism this page's confabulated folklore is missing.
• What Part of the Organisation Does a Person Think For? — the Amanda-and-Bob attribution mechanism as a partial, informal check on whether a claim traces back to anyone accountable for it.
• Understanding Large Language Models — the structural account of hallucination this page's LLM parallel draws on precisely, not loosely.
• Who Is the "We" in "What We Know"? — who gets to declare an institutional account "the real history" in the first place.

Further Reading

• Hirstein, W. (2005). Brain Fiction: Self-Deception and the Riddle of Confabulation. MIT Press.
• Weick, K. E. (1995). Sensemaking in Organizations. Sage Publications.

References