Consensus protocols let non-faulty replicas agree on an ordered log despite some failures. They solve replicated state agreement, not truth, fairness, or model hallucination.

Distributed Consensus Orders Shared Decisions

A multi-agent evaluation measures task outcome, handoff loss, coordination overhead, correlated errors, unauthorized actions, and graceful degradation. Strong individual scores do not guarantee a strong team.

Evaluate the Team and Its Interfaces

In this situation—five model agents agreeing on an invented citation—which choice best applies “Consensus Is Not Correctness”?

Agreement among agents may reduce inconsistency but does not prove factual or ethical correctness. Correlated training, shared context, and copied errors can make unanimous systems wrong.

Multiple agents can propose and challenge reasoning, but shared model biases or persuasive style may create false consensus. A judge needs evidence criteria independent of rhetoric.

Debate Diversifies Arguments, Not Ground Truth

A quorum specifies which and how many authorized participants must respond before a collective action proceeds. It should reflect failure tolerance and stakeholder rights, not just speed.

Quorums Define Sufficient Participation

U.S. National Institute of Standards and Technology

Canonical URL: https://doi.org/10.6028/NIST.AI.100-1
License: U.S. government work; public domain in the United States
Restrictions: Do not imply NIST endorsement; cite accurately and use “Republished courtesy of the National Institute of Standards and Technology” for any permitted republication.

Artificial Intelligence Risk Management Framework (AI RMF 1.0) — Graph Engineering Course Source

Consensus Is Not Correctness

In this situation—agents debating whether a scientific claim is supported—which choice best applies “Debate Diversifies Arguments, Not Ground Truth”?

In this situation—requiring safety, legal, and product approval for a high-impact release—which choice best applies “Quorums Define Sufficient Participation”?

Learn Before

Related

Learn After