← All artifacts · All patterns · Multi-Agent

Debate

aka Multi-Agent Debate, Adversarial Debate

category: multi-agent · status: experimental

Intent

Have multiple agents argue different positions on a question and converge through structured exchange.

Context

The answer is contested or the user wants the strongest case for and against.

Problem

Single-agent answers hide reasoning blind spots; the same model giving both answer and critique reinforces them.

Forces

• Genuinely independent positions are hard to engineer with one model.
• Debate length must be bounded.
• A judge is needed to decide; the judge has its own biases.

Solution

Two or more agents are given different positions. They exchange arguments over N rounds. A judge agent (or a tie-break rule) selects the answer or synthesises a position from both.

Constrains

Consequences

Benefits

• Surfaces counterarguments the user can read.
• Higher answer quality on contested questions in benchmarks.

Liabilities

• N-x cost over single-agent.
• Position assignment is itself a prompt-engineering problem.

Known Uses

• Anthropic AI Safety via Debate research
• MIT CSAIL multi-agent debate work

Related Patterns

• alternative-to → inner-committee
• complements → self-consistency
• generalises → swarm
• alternative-to → infinite-debate
• alternative-to → communicative-dehallucination

References

• Improving Factuality and Reasoning in Language Models through Multiagent Debate — Du, Li, Torralba, Tenenbaum, Mordatch (2023) paper