| paper | d4-conf-atal-2021-0176 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The named results establish logical soundness and completeness, while the paper leaves satisfiability complexity for future work and offers only a conjecture. A population query or intervention objective would add a computational problem that these results do not anchor.
fails bit a — no named computational result to mirror
No result is covered by a continuous-population mirror; the axiomatization and semantic results are unmirrored, and satisfiability complexity appears only as future work.
I don’t see an honest positive anchor in this paper. Theorem 5.4 and Corollary 5.5 establish soundness and completeness for the quasi-model and model semantics; Theorem 6.4 establishes soundness and completeness for the dynamic logic. These are metatheoretic results, not results about the complexity or algorithms for a computational problem. The paper explicitly leaves satisfiability complexity for future work, conjecturing EXPTIME-completeness.
A possible research proposal would give mass to types of ethical agents—agents with the same knowledge, values, and preferences—and ask what fraction can rationally choose a given outcome. But the paper defines neither that population model nor an aggregate computational problem to solve. Its agents are individually indexed in a finite set, and its stated results concern the logic’s semantics and axioms. Turning those into a continuum would require inventing the problem being mirrored, rather than continuizing one of the paper’s computational results. So there is no prover-ready continuous problem per qualifying anchor here; the positive case is weak because the paper supplies no such anchor.
The strongest objection is that the paper gives no computational result to continue. Theorem 5.4 and Corollary 5.5 establish soundness and completeness for the static logic; Theorem 6.4 does the same for its dynamic extension. They do not give algorithms or complexity bounds for a problem whose input includes a population. Replicating agents or assigning their types masses leaves those metatheoretic claims untouched.
The paper’s closest computational hook is its conclusion’s proposed study of satisfiability complexity. But satisfiability here concerns formulas and evaluative models: it asks whether some model and world satisfy a formula. The finite set of agent identifiers indexes modalities; it is not an input population whose size or distribution affects the question. Extending that set to a continuum would matter only if the language or task also asked about the population—for example, what fraction of agents satisfy a formula. That would be a new aggregate query, not a continuous version of the satisfiability problem the authors identify.
The best applied reframe is plausible: take a large deployment of ethical robots, group robots with the same complete knowledge, values, and preferences into types, and ask what fraction can rationally choose an action, perhaps after a value expansion. But the paper’s choice operator is local to an agent and its evaluative model. For each supported type, one can determine whether the operator holds and sum that type’s mass. The paper supplies no collective rule saying what proportion is enough, no costs for changing a cohort’s values, and no population objective to optimize. Adding a threshold or intervention cost could produce a sensible research question, but the substantive problem would come from that added layer.
The dynamic result has the same limitation. Theorem 6.4 validates an update semantics for a finite coalition. Applying an update to a whole type cohort is conceivable; optimizing over the mass of agents to update would require a cost model and a collective target absent from the paper. If agents of one type are truly indistinguishable for everything the problem uses, their evaluations and responses coincide, so mass only scales the count. If their responses depend on network positions or interactions with particular others, those distinctions must enter the type, and the paper’s finite, named-agent semantics still does not provide a population-level computational problem.
I would not claim that no high-multiplicity ethical-agent scenario could ever be worthwhile; a large robot deployment is a credible setting. The negative case is narrower: none of the paper’s named results anchors such a mirror, and the most attractive population questions require an aggregation or intervention problem the paper does not formulate. That makes a mirror of this paper a substantial new model, rather than a continuization of its computational contribution.
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.