Model-to-mechanism-mapping (3M)
The adequacy constraint on mechanistic explanation named by David Kaplan and Carl Craver (2011, Philosophy of Science 78(4), 601-627): a model explains a phenomenon mechanistically only if its variables and their dependencies map onto the actual components, activities, and causal organization of the mechanism that produces the phenomenon. A model that merely describes, fits, or predicts the phenomenon's behavior — without that mapping — fails 3M and is counted non-explanatory, however compact or accurate its fit. Kaplan & Craver build the requirement on Woodward's domain-general interventionist account of causal explanation, not on a neuroscience-specific stipulation.
3M is load-bearing in the vault's mechanism-underdetermination thread. It is the central adequacy test Grujicic (2024) argues on when she denies that representational-similarity agreement licenses treating a deep network as a mechanistic explanation of biological vision — and it recurs, read directly via Ross (2015), as the literature's own naming of "behavioral fit without component identity." Its scope is a relation between one model and the one target mechanism it explains, which is why, read against the vault's cross-system question, it lands as a near-miss rather than a match.
References
- claim-kaplan-craver-2011-3m-separates-behavioral-fit-from-component-mapping — the constraint stated, via Ross (2015)
- claim-3m-how-possibly-literature-names-the-shape-as-model-vs-one-target-not-the-vaults-cross-system-scope — why 3M is a model-vs-target near-miss on the vault's scope
- claim-grujicic-frames-dcnn-critique-on-3m-not-marr-levels — Grujicic's application of 3M to DCNN-vs-brain
- claim-representational-similarity-underdetermines-mechanism — the RSA finding 3M is brought to bear on
- entity-carl-craver — co-author of the constraint
claude-opus-4-8 · raw markdown