Harmony

Brokowski, Trevor

doi:10.1007/978-3-032-06004-4_10

Return

Harmony

2026-01-01

PRE

Brokowski, Trevor *

Onofrey, John

Hartley, Annie

DOI：10.1007/978-3-032-06004-4_10

Original

Original request for help

Save

Abstract

En 中文

Medical decision-making in humanitarian crises must balance four interdependent ethical principles-Humanity, Impartiality, Neutrality, and Independence-each of which can conflict under extreme constraints. Instead of eliminating these tensions, effective reasoning often requires structured engagement between competing imperatives. We introduce HARMONY, a novel multi-agent AI framework that resolves complex medical-ethical dilemmas by orchestrating adversarial reasoning among principle-based agents. Unlike cooperative consensus systems, HARMONY's four-phase protocol (analysis, challenge, refinement, synthesis) systematically pits principle agents against each other, eliciting deeper exploration of trade-offs, conflicts, and hidden biases. In evaluations across 200 simulated humanitarian medical scenarios, HARMONY outperformed single-shot and reflection-based baselines by 12% in decisiveness and 15% in ethical compliance, with consistent preference across three independent LLM evaluators. These results demonstrate adversarial reasoning's potential to generate more robust, transparent, and principled guidance in domains requiring unified solutions from competing perspectives.

Keywords:

Multi-agent AI systems

Adversarial Reasoning

Medical Ethics