Conflict Resolution in Normative Multiagent Systems

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 270 papers108 conceptsfrozen 2026-10-04

When an autonomous agent enters the human world, it inherits a thicket of overlapping rules. A cleaning robot might be governed by two simple norms: "keep the floor clean" and "do not disturb humans." But if a person is standing directly over a coffee spill, these rules collide. In classical logic, a contradiction like this is a "deadlock"—the system has no valid move.

To move beyond this paralysis, researchers are developing frameworks for normative conflict resolution. This field seeks to move AI from rigid rule-following to "all-things-considered" reasoning, where agents can weigh competing obligations, handle exceptions, and navigate "no-win" scenarios without collapsing into logical inconsistency.

The Limits of Static Logic

Traditional AI often treats norms as hard constraints. However, as Marvin Belzer noted in A Logic of Deliberation, real-world reasoning requires a way to prioritize specific situational requirements over general rules. Without this, systems reach logical contradictions.

The modern approach to this problem is defeasible deontic logic. Unlike standard logic, where a fact is either true or false forever, defeasible logic allows for "rules of thumb" that can be defeated by new information or higher-priority norms. Ezio Bartocci, Agata Ciabattoni, and Guido Governatori have demonstrated in A Normative Supervisor for Reinforcement Learning Agents that this logic can act as a "supervisor" for black-box agents, filtering out unethical actions by reasoning through exceptions in real-time.

Prioritization and the Economy of Violations

When rules conflict, the agent must decide which one to break. This requires a mechanism for utility calibration.

In Norm Conflict Resolution in Stochastic Domains, Daniel Kasenberg and Matthias Scheutz propose treating norms as "suspendable constraints" with associated violation costs. If an agent must choose between two harms, it calculates a compliance score to identify the "least-bad" path.

However, assigning these priorities is difficult. Shuyue Hu and Siyue Ren found that Large Language Model (LLM) agents often struggle with this, frequently assigning "maximum priority" to every rule in their database (Emergence of Social Norms in Large Language Model-based Agent Societies). Without nuanced ranking, the agent remains stuck in a state of "normative resistance," unable to distinguish between a minor social faux pas and a critical safety breach.

Contrary-to-Duty Reasoning

A central challenge in this field is the "sub-ideal" state: what should an agent do after a norm has already been violated?

In classical systems, once a primary obligation is broken (e.g., "do not crash"), the system may have no further instructions. Researchers like Luca Gasparini and Timothy J. Norman address this through Contrary-to-Duty (CTD) reasoning. In Severity-sensitive norm-governed multi-agent planning, they describe secondary obligations—like "if you crash, you must call for help"—that only trigger after a failure. This ensures that an agent continues to seek the "next-best" state of compliance rather than abandoning normative behavior entirely.

Jan M. Broersen and Leendert W. N. van der Torre have formalized this in the BDIOCTL framework, which integrates these external "must-dos" into the agent's internal beliefs and intentions, preventing the logical paradoxes that usually arise when a system tries to reason about its own failures.

Norms as Arguments

One of the most intuitive ways to resolve conflicts is to treat norms not as math, but as a debate. Formal Argumentation Theory allows agents to treat conflicting rules as "arguments" that can attack or support one another.

Beishui Liao and Nir Oren have shown that classic methods for resolving prioritized norms—such as greedy selection or optimization—can be mapped directly onto argumentation frameworks (Prioritized Norms in Formal Argumentation). By identifying "cycles" of conflicting obligations and applying cycle-elimination procedures, agents can derive a consistent set of "all-things-considered" duties even when their underlying rulebook is a mess of contradictions.

Toward Justified Agency

The ultimate goal of this research is what Pat Langley calls justified agency (Explainable, Normative, and Justified Agency). It is not enough for an autonomous vehicle to choose the "least-bad" outcome in a crash; it must be able to defend that decision using the social and legal norms of the society it inhabits.

By combining formal logic with conflict resolution, researchers are building agents that don't just follow rules, but understand the trade-offs they make—providing a foundation for human trust in increasingly complex autonomous systems.

Papers in this version

Showing 60 of 70 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 33 papers
  2. Version 1 · 2026-10-04 · 70 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_normative_conflict_resolution_ebe445bc,
  title = {normative conflict resolution},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/normative-conflict-resolution/rdr_ebe445bc7ee847bd811e293839faf62c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_conflict_resolution_in_normative_multiag_ebe445bc_v1,
  title = {Conflict Resolution in Normative Multiagent Systems},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/conflict-resolution-in-normative-multiagent-systems/rdr_ebe445bc7ee847bd811e293839faf62c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}