| paper | d4-conf-ecai-2008-0105 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | unclear |
Proposition 1 states a semantic preservation condition for an individual model; it does not claim that an object can be computed or that computing one is hard. The proposed population tally is a coherent aggregate extension, but this paper provides no named computational result to anchor it.
fails bit a — no named computational result to mirror
Propositions 1–8 establish semantic preservation conditions for individual preference change; none supplies a computational result for the proposed population tally.
The strongest positive reading is a population of people in a large deliberative community who receive the same piece of information. A type would be a complete mental state in the paper’s sense: a normality ordering and a preference ordering over the possible worlds. Mass \(\mu_t\) would be the fraction of the community with mental state \(t\). The common revision rule updates each type after learning \(\alpha\). One could then ask what fraction satisfies a chosen formula-level preference \(P(\beta)\) afterward.
That gives a concrete possible extension: given a finite support of such types with rational masses, \(\alpha\), \(\beta\), a revision function, and one of the paper’s preference semantics, compute
\[
\sum_{t:\,M_t\star\alpha\models P(\beta)}\mu_t.
\]
The scenario is plausible when many people share a relatively small number of coarse mental-state types. The update is the action; the requested output is the resulting mass. But this is an aggregate evaluation problem I have added. The paper has no population, collective objective, or optimization over actions.
The anchor requirement is the decisive limit. The paper has no named computational result asserting that a problem is in P, NP-hard, FPT, or similar. Its closest result is Proposition 1, proved in the paper: the preference-satisfaction postulate \((P1)\), “\(P\alpha\to[\star\alpha]P\alpha\),” holds if the revision function satisfies positive and negative uniformity, and it also holds for any lifting operator under the LTW semantics. That is a semantic preservation result for an individual model, not a result about the complexity of a computational problem. Propositions 2–8 have the same character. So there is no qualifying anchor, and no honest anchor-specific continuous problem or complexity prediction to hand to a prover.
The weakest point is also the central one: moving from an individual’s preference change to the distribution of preference changes in a population adds an aggregation question the authors do not study. The proposed extension may interest them, but I would not claim that they would recognize it as the continuous form of a problem established by this paper. The positive case is therefore an unanchored research direction, not a strong mirror case.
The strongest objection is that the paper has no computational result to mirror. Proposition 1 is its closest anchor, but it says when an individual preference-satisfaction postulate holds: if the revision function has the stated uniformity properties, or the LTW semantics is used, then \(P\alpha\to[\star\alpha]P\alpha\). Propositions 2–8 are similar semantic claims. None studies an algorithmic problem, its complexity, or an optimization task.
The best population version of Proposition 1 is straightforward. Treat each complete mental state \(t\) as a type, and let all types receive the same \(\alpha\) under a revision setup covered by the proposition. Then the types satisfying \(P\alpha\) before revision are a subset of those satisfying it afterward, so
\[
\sum_{t:M_t\models P\alpha}\mu_t
\;\le\;
\sum_{t:M_t\star\alpha\models P\alpha}\mu_t.
\]
That is a valid mass-level statement. But it is just Proposition 1 summed over a distribution. Computing either sum for a finite support is a tally over types; the paper supplies no complexity question for which that tally would be a mirror.
One could make the proposal more ambitious: choose a fact or information campaign to maximize the share of a population that prefers some formula afterward. That is a plausible new project, but its campaign options, costs, and collective objective would have to be added. The paper studies how a given belief revision changes one agent’s formula-level preferences; it does not choose inputs to shift a population, define a group decision, or optimize across agents. Any hard computational problem would come from those added choices, not from Proposition 1.
The proposed cohort of people with shared mental-state types is coherent, and high multiplicity is not itself a problem here. The difficulty is the programme’s gate: a sensible population model does not supply a computational-social-choice problem established by this paper. A mass distribution over independent updates gives aggregate evaluation; adding a campaign or group-choice problem may be worthwhile, but it would be a new research direction rather than a continuous mirror of a named computational result.
So the negative case is strong against the offered anchor, but it is not airtight as a universal claim. The paper’s game-theoretic motivation leaves room for a population study its authors might recognize. What the text does not support is treating that possibility as an anchored computational mirror of this paper’s results.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.