Large language models struggle with exploration in chemical experiment design
LLM sequential decision making under uncertainty in biochemical domains
Artificial Intelligence
Summary
Designing new chemical experiments is hard because it requires smart decisions about what to try next. The authors tested several large language models (LLMs) to see how well they make these decisions compared to traditional methods. They found that while LLMs try to explore new options, they often get stuck relying too much on past information and end up acting in a more limited, exploitative way. This means LLMs do not yet outperform statistical approaches in guiding chemical experiments under uncertainty. The study shows that separating past knowledge from new data helps LLMs explore better.
What this means in practice
- •For chemical engineering teams: Test and improve experiment design strategies by understanding LLM limitations in exploration versus exploitation.
- •For machine learning engineers: Develop better LLM prompting techniques that separate prior knowledge from new data to enhance exploration in sequential decision tasks.
Authors
Mattias Akke, Soojung Yang, Jurgis Ruža, Sathya Edamadaka, Rafael Gómez-Bombarelli
Abstract
Large language models (LLMs) are increasingly used to drive scientific discovery. Understanding how LLMs make decisions from new data and memory of the literature is vital before trusting them to design experiments under tight experimental budgets. However, their decision strategies are invisible in the current performance scores used to evaluate research agents. Here, we benchmark five frontier LLMs in a Bayesian Optimization setting against published statistical baselines on seven combinatorial datasets spanning protein engineering, reaction optimization, molecular design, peptide self-assembly, and catalysis. Performance is paired with direct measurements of model beliefs and actions, enabling highly resolved behavior analysis. A prompt ablation that progressively strips context separates memorization from chemical reasoning and from bare categorical optimization. Prior chemical knowledge helps in expectation, but with high variance and occasionally even harms performance. No configuration tested decisively beats a mean statistical baseline across domains. Belief-movement and Martingale diagnostics, corrected here for a measurement-noise bias that mislabels rational agents as irrational, show that models overreact to incoming data rather than entrenching on their priors in the contexts studied here. Interestingly, while LLM actions are exploitative, models sincerely intend to explore and consistently act on that intent. This failure is a competence gap arising from context-stickiness. Removing in-context history restores exploration, indicating that priors and data must be decoupled to achieve effective LLM-driven discovery.