Why it is a question
CLAIM-440 is functional: valence is potential of a negotiation by an ordering of the system’s own states. CLAIM-439 adds a candidate for when that regime is felt: when the system also models its potential. Neither is testable until each layer can be measured without the others.
The three layers, with measures already in use elsewhere.
- Potential. Which ranking wins the system’s conflicts, for how long, and whether it returns (QUESTION-268): a persistent potential (TERM-078), reported as the class of orderings the record fits until an intervention narrows it. For animals and people, arbitration weights estimated from choice, in the tradition of Daw, Niv and Dayan’s uncertainty-based competition between controllers (LIT-580). For organizations, a record of decisions on conflict pairs.
- A hedonic marker that does not depend on choice. Orofacial taste reactivity, which Berridge’s programme uses to measure “liking” across species; judgement bias, the reading of ambiguous cues, used for animal affect; and moment-by-moment reports of experienced utility, which Kahneman, Wakker and Sarin separate from the decision utility that choice reveals.
- The system’s model of its own potential. Confidence and metacognitive report, or, for systems that cannot report, an internal variable that tracks layer 1 and is available to other processes, as Brouillet and Friston’s felt fluency is an inference about a change in precision (THEORY-662) and Hesp et al.’s affective precision is a belief about the reliability of one’s own policies.
Two constraints from the readings.
- Layer 3 cannot be validated by retrospective report. Remembered utility is a biased model of experienced utility, by peak-end weighting and duration neglect (LIT-1448, NOTE-1105).
- Layer 3 has a test that needs no report. Hesp et al. (LIT-1468, NOTE-1111) compare an agent with and without the level that models its own action-model precision: behaviour is the same while contingencies are stable and diverges after a reversal. A self-model of potential should show itself in how potential is re-set after a change, not in steady-state choice.
A human candidate for separating layers 1 and 2. Mohammadi, Van De Ville and Vuilleumier (LIT-1444, NOTE-1107) find two networks correlating with valence: one that “may relate to the theoretical construct of “wanting"" (p. 8005) and one “reminiscent of the “liking” rather than “wanting” aspect of valence” (p. 8006). That would let the two layers be measured apart in people. It does not yet show them dissociating: the latent variables are orthogonal by construction and both come from one retrospective report.
Where the layers should come apart.
- On CLAIM-440 as revised, layers 1 and 2 need not coincide at all. Layer 1 is goal conduciveness and layer 2 intrinsic pleasantness, two levels of valence with their own orderings (CLAIM-433), and the claim implies nothing about how they relate. An unpleasant treatment that serves the system’s maintenance is negative on layer 2 and, if it prevails, positive on layer 1. So their covariation is a conditional hypothesis: they covary where the state the system pursues is the state it consumes, where what serves its maintenance is also what is pleasant, and come apart where pursuit and consummation part, as in the treatment and in wanting without liking (THEORY-893, CLAIM-438). The test is whether divergence is confined to such cases. An earlier draft of this question predicted outright that the two layers coincide when a self-maintenance ordering prevails; a review of the pull request that filed it found the revised claim did not imply that, and the prediction was made conditional. Decision utility diverging from experienced utility is no objection either way, provided layer 1 is read within the episode (QUESTION-268).
- On CLAIM-439, felt valence is present where layer 3 tracks layer 1 and absent where layer 1 runs without it, as unfelt fluency does.
- For organizations, layer 3 is the organization’s reporting of what wins its conflicts to those who govern it. CASE-151’s board was told terminations were “in the range of 200-300”, which, if the claim holds, is a mis-modelled potential, not an absent one.
What would count as an answer
A protocol that measures each layer without using the others, applied to at least one system, with a verdict on whether the predicted coincidences and dissociations hold.
What would not answer it
A single measure presented as felt valence, or any measure of layer 3 validated only by reports that are themselves layer 3.