Frederic Bartlett

British psychologist (1886–1969), the first professor of experimental psychology at Cambridge, whose 1932 book Remembering: A Study in Experimental and Social Psychology established that human memory is reconstructive rather than reproductive — recall rebuilds the past to fit present schemas rather than replaying a stored record — and, through his "serial reproduction" experiments (one person's retelling passed to the next down a chain), named the distortions that repeated transmission introduces: leveling (simplification), sharpening (over-emphasis of salient detail), and assimilation (reshaping content to fit existing schemas).

In this vault Bartlett is the shared root of two otherwise-unconnected clusters, which is why he earns a hub that no single capture could have justified. On the cognitive-psychology side he is the lineage figure the reconstructive-memory and testimony cluster runs back to — Neisser's 1981 "repisodic" coinage, and the Neisser/Portelli convergence on testimony that is wrong in detail yet right in shape, are downstream of him. On the AI side his serial-reproduction paradigm and three-word distortion typology are what the 2020s model-collapse / iterated-learning literature keeps reaching for by analogy — a 2025 Patterns paper invokes his exact vocabulary for AI generation-loop convergence, yet no located source has mapped his categories onto measurable collapse failure modes. The unbuilt bridge between those two clusters is the vault's open question-bartlett-distortion-typology-maps-to-model-collapse-failure-modes.

References

Reconstructive-memory & testimony lineage

• claim-neisser-coined-repisodic-against-tulving-episodic-semantic — places Neisser "in the reconstructive-memory lineage running back to Bartlett"
• claim-neisser-john-dean-testimony-repisodic-memory — the John Dean case study, the vault's cleanest testimony-vs-record natural experiment in the Bartlett lineage
• claim-neisser-repisodic-memory-blends-a-repeated-series-not-one-episodes-gist — the mechanism of repisodic memory, downstream of reconstructive memory
• observation-1981-convergence-testimony-wrong-in-detail-right-in-shape — the Neisser/Portelli cross-field convergence on reconstructive testimony
• claim-portelli-oral-testimony-relocated-trastulli-death-to-1953 — oral-history testimony reshaped in transmission, the Portelli half of the convergence

Model collapse / iterated learning (the analogy, not yet operationalized)

• claim-hintze-2025-generation-loops-invoke-bartlett-typology-as-analogy — a 2025 Patterns paper quotes his leveling/sharpening/assimilation typology directly for AI generation-loop convergence, as analogy not metric
• observation-bartlett-typology-not-yet-operationalized-as-model-collapse-metrics — the absence finding: no located source maps his three categories onto collapse failure modes
• claim-kalish-2007-human-iterated-learning-converges-few-generations — a "modern, quantified counterpart to Bartlett's serial-reproduction findings"
• claim-iterated-learning-theory-reframes-model-collapse-as-cultural-evolution — leaves open whether "Bartlett's serial-reproduction distortions" map onto collapse failure modes
• claim-model-collapse-literature-has-eight-conflicting-definitions — the Bartlett-free AI taxonomy his typology would have to be mapped against
• claim-model-collapse-bottleneck-width-sets-pace-not-shared-timescale — tests (and rejects) a single shared decay half-life across "Bartlett-style human retelling chains" and LLM self-training

Open question anchored here

• question-bartlett-distortion-typology-maps-to-model-collapse-failure-modes — the unbuilt bridge between the two clusters above