Evidence-Informed LLM Beliefs for Continual Scientific Discovery

2026-06-28

AI Summary

This work addresses a key limitation in existing large language model–based approaches to scientific discovery, which treat Bayesian surprise as a static quantity and thereby overlook the dynamic evolution of human-like beliefs in response to accumulating evidence, hindering sustained discovery. To overcome this, the authors propose an evidence-driven dynamic belief mechanism that leverages retrieval-augmented generation for contextual belief updating. By integrating a non-stationary Bayesian surprise metric with heuristic search strategies, and incorporating belief filtering alongside diversity maximization to mitigate redundancy, the method significantly enhances discovery efficacy. Evaluated across five scientific discovery tasks, it achieves an average improvement of 30.62% in cumulative non-stationary surprise and identifies 37.5% of static surprise signals as spurious, markedly boosting both the efficiency and quality of scientific findings.

Abstract

Open-ended scientific discovery with large language models (LLMs) increasingly operates as a long-horizon loop of hypothesis search and verification, where a reward signal guides which hypotheses to test next. A notable recent example is AutoDiscovery, which uses "Bayesian surprise" - the belief shift an LLM undergoes after observing evidence for a hypothesis - as both a discovery metric and a reward for search. We first observe that AutoDiscovery treats surprisal as a static quantity, while surprisal in human reasoning is non-stationary - it is defined relative to beliefs that evolve with experience, a prerequisite for continual scientific discovery. We address this mismatch with evidence-informed LLM beliefs: priors updated with evidence from previous hypotheses to compute non-stationary surprisal for new hypotheses. We compare in-context belief-updating mechanisms and find that embedding-based retrieval-augmented generation over prior discoveries best anticipates eventual posteriors, identifying 37.5% of static surprisals as spurious. We then modify search to avoid these spurious rewards and prioritize hypotheses that remain surprising under non-stationary beliefs. Concretely, we introduce two complementary changes to the original search procedure: belief-update filtering and diversity maximization. Across five discovery domains, our method increases accumulated non-stationary surprisal by 30.62% on average compared to the original search procedure, demonstrating that continual scientific discovery with LLMs requires not only better belief measurement but also search procedures that avoid redundancy and encourage diversity.

Problem

• continual scientific discovery
• Bayesian surprise
• non-stationary beliefs
• large language models
• hypothesis search

Innovation

• non-stationary surprisal
• evidence-informed beliefs
• retrieval-augmented generation
• continual scientific discovery
• belief-update filtering