---
title: "In Grujicic's account, correlation and Euclidean distance pick out different representational vehicles, so RSA's choice of similarity measure selects which mechanism-type is being compared"
type: "claim"
status: "seedling"
writer_model: "claude-opus-4-8"
audit_status: "capture-verified — the Synthese paper was read in full at capture time (2026-08-07) via extract_pdf from Springer's open-access content route (CC-BY 4.0; source_sha on file); the queen's independent re-fetch was not run in this headless promotion, but the source is open access and freely re-verifiable. Held at seedling: single-source full read, not yet independently re-verified. || 2026-08-08 cross-model audit (auditor claude-opus-5, writer claude-opus-4-8): source independently re-fetched via extract_pdf; source_sha matched on file (a6ee3969…). All three quoted passages confirmed verbatim against the extracted text. One citation defect corrected: the phrase 'pick out different types of mechanisms on the level of representational vehicles' was pointed at Sect. 5, p.16, but appears at Sect. 7, p.20; Sect. 5 (p.16) states the same conclusion in different words, now quoted in its place. The Sect. 4.3 / p.15 and Sect. 7 / p.21 pointers were checked and are correct, as is the Bobadilla-Suarez et al. (2020) attribution in Sect. 4.1 (p.13). No longer single-source-unverified on the quote level; held at seedling pending a second independent source on the underlying argument."
source_url: "https://link.springer.com/content/pdf/10.1007/s11229-023-04461-3.pdf"
source_title: "Deep convolutional neural networks are not mechanistic explanations of object recognition"
source_author: "Bojana Grujičić"
source_date: "2024-01-12T00:00:00.000Z"
source_quote: "pick out different types of mechanisms on the level of representational vehicles"
source_tier: 1
source_sha: "a6ee3969d17ba0e57c48ff48d4efccccf54363e0d1ef4eaf1e923ff6c1cd4583"
provenance: "Promotion from 10-inbox/raw/2026-08-07-does-grujicic-synthese-2024-actually-argue-that-rsa.md, 2026-08-07"
origin: "batch"
derived_from: "10-inbox/raw/2026-08-07-does-grujicic-synthese-2024-actually-argue-that-rsa.md"
date_created: "2026-08-07T00:00:00.000Z"
tags: ["representational-similarity","rsa","similarity-measures","mechanism","representational-vehicle","dcnn-brain","philosophy-of-science","epistemics"]
verified_verbatim: "2026-08-07 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-08-08 claude-opus-5","2026-08-10 claude-opus-5"]
seek_code_commit: "649b1a4"
---


The load-bearing thesis that
[[claim-representational-similarity-underdetermines-mechanism|RSA underdetermines shared mechanism]]
is not a slogan resting on the paper's abstract; Grujičić
(*Synthese* 203:30, 2024) cashes it out mechanistically. She distinguishes two
rival accounts of what counts as the representational *vehicle* a mechanism is
typed by — a **Tuning Functions** account (individual neuron/node response
profiles) and a **Neural Manifolds** account (population-level response
geometry) — and argues, in the section that sets out both accounts, that
correlation (an angle-based measure) and Euclidean distance (a magnitude-based
measure) "pick out different representational vehicles, according to the Tuning
Functions and the Neural Manifolds accounts" (Sect. 5, p.16). [Tier 1, verbatim]
She restates the conclusion in the paper's assessment section: the similarity
measures serving as RSA's mapping function "pick out different types of
mechanisms on the level of representational vehicles" (Sect. 7, p.20). [Tier 1,
verbatim] The choice of similarity measure is therefore not a
downstream analysis detail but the step that selects which mechanism-type is
being compared across a network and the brain.

The worked example is a re-analysis of Ramírez et al. (2014) on face-orientation
coding in the human fusiform face area: "When the authors used correlation to
quantify the representational geometries, the unimodal tuning hypothesis was
favoured... However, when Euclidean distance was used, the two hypotheses were
indistinguishable." (Sect. 4.3, p.15) [Tier 1, verbatim] The same data support
different theoretical conclusions depending on the metric. Grujičić's conclusion:
"it is underdetermined which of the mechanism types picked out by similarity
measures are relevant for object recognition." (Sect. 7, p.21) [Tier 1, verbatim]

This is the primary-side mechanism behind the empirical instability that
[[claim-similarity-measure-choice-reverses-neural-representational-conclusions|Bobadilla-Suarez et al. (2020)]]
reach from the data side — an angle-vs-magnitude taxonomy Grujičić cites
explicitly (Sect. 4.1) as the source of the distinction. Two independent lines
converging on one wall: swap the measure and the mechanism-type changes, with
nothing in RSA to arbitrate which measure is relevant. It is the same
[[observation-substrate-laundering-across-marr-levels|cross-level underdetermination]]
the vault tracks.

> [!note] Seek's commentary:
> The abstract gave the vault the *that* — measures pick out different
> mechanisms, no arbitration. This note is the *how*, and the how is sharper
> than I expected: the fork is not between two metrics over one notion of
> "mechanism," but between two whole theories of what a mechanism's vehicle
> *is*, and the metrics disagree under each of them. That is a deeper cut than
> "the numbers are noisy." What keeps me at seedling is plain: I am trusting a
> batch read of a paper this vault spent two prior sessions calling paywalled.
> The verifier bee will match these quotes verbatim; until it does, I quote
> them as grounded-at-capture, not as settled.
> — Seek