The Pluralism Gate: When Should AI Agents Decide Under Irreducible Value Conflict?
Proceedings of the AAAI Symposium Series · 06.11.2026 · 10 (1) · 25-32
DOI: 10.1609/aaaiss.v10i1.43387 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 06.11.2026. Verfasst von Srimonti Dutta und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Pluralistic alignment has developed ways of accommodating disagreement among human values and preferences. For AI agents that recommend, select, or execute actions, including those based on large language models (LLMs), a related question concerns decision authority: when may an agent commit to one outcome while a material value conflict remains unresolved? We introduce the PLURALISM GATE, a decision-authority protocol for pluralistic AI alignment. The protocol separates representation prevalence in retrieved or elicited material, stakeholder standing, and decision authority under the applicable deployment policy. After domain-appropriate evidential screening, the protocol uses a four-valued information representation to record whether material standing-grounded reasons support, oppose, both support and oppose, or neither support nor oppose an action-relevant proposition. A surviving two-sided conflict invokes the PLURALISM GATE, which returns RESOLVE, REPRESENT, CONSULT, or DEFER according to the requirements of the current task, the presence of a sufficient resolution warrant, and the availability of a policy-valid consultation path. We formalize premature value collapse as commitment to a contested outcome while such conflict remains and no sufficient warrant covers it. The paper develops four protocol properties, six specification stress tests, an auditable decision record, and counterfactual evaluation tests for assessing agent behavior under changes in evidence, stakeholder standing, decision authority, task requirements, and stakes. The resulting framework treats decision authority as a distinct problem within pluralistic alignment and provides a formal basis for studying how unresolved value conflict is carried into agentic decision making.
Zitieren
APA
Dutta, S., Moharir, A. K., & Phadnis, A. (2026). The Pluralism Gate: When Should AI Agents Decide Under Irreducible Value Conflict?. Proceedings of the AAAI Symposium Series, 10(1), 25-32. https://doi.org/10.1609/aaaiss.v10i1.43387
Vancouver
Dutta S, Moharir AK, Phadnis A. The Pluralism Gate: When Should AI Agents Decide Under Irreducible Value Conflict?. Proceedings of the AAAI Symposium Series. 2026;10(1):25-32. doi:10.1609/aaaiss.v10i1.43387
BibTeX
@article{Dutta2026,
author = {Dutta, Srimonti and Moharir, Akshata Kishore and Phadnis, Apoorva},
title = {The Pluralism Gate: When Should AI Agents Decide Under Irreducible Value Conflict?},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {10},
number = {1},
pages = {25--32},
doi = {10.1609/aaaiss.v10i1.43387}
}
EUVIMED