Against Homogeneous Consensus: Why Scientific Discovery Requires Heterogeneous Adversarial LLM Agents

Shuai Wang

Track: long paper (up to 10 pages)

Keywords: LLM Agents, Heterogeneous Adversarial Reasonin, AI Scientist, Auto-research

TL;DR: Autonomous Swarms of AI agents Requires Heterogeneous Adversarial LLM Agents

Abstract:

In this position paper, we argue that current LLM agents, optimized strictly for consensus and coherence, act as epistemic echo chambers that reinforce dominant scientific paradigms. We posit that true discovery requires Epistemic Friction—structured disagreement between heterogeneous explanatory models.
To articulate this vision, we introduce the Triadic Disagreement Framework, a conceptual agent architecture where a consensus-aligned Proposer and a falsification-aligned Challenger engage in sustained adversarial interaction.
Through an illustrative simulation on Alzheimer's disease etiology, we illustrate how this architectural heterogeneity can surface suppressed explanatory pathways (e.g., the Infection Hypothesis) that standard consensus-driven agents ignore.
Our work calls for a shift from helpful assistants to adversarial co-scientists capable of preserving irreducible epistemic conflicts.

Presenter: ~Shuai_Wang34

Format: Yes, the presenting author will definitely attend in person because they attending ICLR for other complementary reasons.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Submission Number: 164