Source
Gładziejewski (2025), Synthese 206:225, LIT-1360, read in full in NOTE-1056: §2 (pp. 4–10), §3 (pp. 11–15) and footnotes 2, 4, 5, 6 and 9.
What was actually shown
- Compression and inference (§2). With optimal codes, maximising the posterior is minimising −log P(M) − log P(D|M), the total description length, so “Bayesian inference is real-pattern extraction” (p. 8). This is a derivation (footnote 3), re-checked in the NOTE. Free energy’s complexity and accuracy terms are “analogous” to MDL’s model and data terms (p. 10). That is an analogy.
- Structure (§3). Generative models “specify probabilistic and causal structural descriptions of sensory patterns” (p. 12). They are not freely constructed, for two reasons: compressibility is a mathematical fact, and the signal is shaped by its external source “at least unless the model is used to run fully off-line simulations, as may be the case in imagery or dreams” (p. 13).
- Veridicalism (p. 13). Matched sensory streams are “necessarily … equally compressible by equivalent higher-level (manifest) structural descriptions”. So the envatted twin’s door is “as real as the door you open”. This step is valid given its premise, since equal inputs have equal compressions.
- What lies outside (p. 14). What realizes the structural roles, purely intrinsic properties, and deeper structure not induced from the senses.
What this does not say
- That manifest content is only structural. The veridicalism needs it, and the paper assumes it. A qualitative content that matched streams do not fix would break the step.
- That every manifest property is descriptive. Footnote 9 allows a continuum to “projections”, with cuteness as the example.
- That the brain uses optimal codes, which footnote 4 leaves open, or that predictive processing is literally true, which footnote 2 assumes.
- Anything about experience as such. The manifest image is made of the “entities that we represent in our conscious experience and/or in our intuitive theories” (p. 3). Nothing is said about what makes a representation conscious.
- That the world behind the senses is unknowable. Deeper structure is noumenal only “in the limited sense” that it is not induced from the senses (p. 14). Science may reach it.
Connections
- CLAIM-413. Grounding is a relation, relative to a referent. On this account the envatted door is grounded as a structural role, constrained by an independent source, and ungrounded as to its realizer. That is one way to state the claim’s point that the envatted experience is ungrounded “with respect to the tree” while grounded in the apparatus. It supports the claim’s “Ungrounded ≠ Unconstrained”: the signal constrains the model except in off-line simulation. Epistemic grounding (CLAIM-342) is not addressed.
- THEORY-095. Block’s Inverted Earth keeps phenomenal character fixed while externally individuated content inverts. A structural internalism about manifest content is on the side THEORY-095 leaves open. If colour character is manifest content and not structure, the veridicalism has a counterexample.
- THEORY-036. The real-pattern criterion needs additions to do ontological work. Here they are the agent’s pragmatic trade-off and, for objects, Bird’s maximality and non-divisibility.
- THEORY-872. The self as one of the patterns.