Hypothesis Hunting with Evolving Networks of Autonomous Scientific Agents

TL;DR

The paper defines hypothesis hunting as open-ended exploration over large-scale datasets and introduces AScience and ASCollab, a framework and distributed system where heterogeneous LLM-based agents form evolving networks to generate, critique, and cumulatively refine findings. The formalism captures an epistemic landscape, agent heterogeneity, attention networks, and shared evaluation norms, with mechanisms for memory, collaboration, and peer review that endogenously shape progress. In TCGA cancer cohorts, ASCollab demonstrates sustained exploratory activity, rediscoveries of known drivers, pathway extensions, and novel therapeutic hypotheses, outperforming independent baselines in novelty and quality while increasing diversity of findings. The work highlights the potential of socially structured autonomous agents to scale hypothesis hunting, while acknowledging the need for wet-lab validation and careful consideration of domain generalization and evaluation subjectivity.

Abstract

Large-scale scientific datasets -- spanning health biobanks, cell atlases, Earth reanalyses, and more -- create opportunities for exploratory discovery unconstrained by specific research questions. We term this process hypothesis hunting: the cumulative search for insight through sustained exploration across vast and complex hypothesis spaces. To support it, we introduce AScience, a framework modeling discovery as the interaction of agents, networks, and evaluation norms, and implement it as ASCollab, a distributed system of LLM-based research agents with heterogeneous behaviors. These agents self-organize into evolving networks, continually producing and peer-reviewing findings under shared standards of evaluation. Experiments show that such social dynamics enable the accumulation of expert-rated results along the diversity-quality-novelty frontier, including rediscoveries of established biomarkers, extensions of known pathways, and proposals of new therapeutic targets. While wet-lab validation remains indispensable, our experiments on cancer cohorts demonstrate that socially structured, agentic networks can sustain exploratory hypothesis hunting at scale.

Figures (5)

• Figure 1: Hypothesis hunting. Large-scale datasets are explored by autonomous networks of research agents that collaborate, peer-review, and refine findings to surface promising directions for human validation.
• Figure 2: ASCollab. Evolving network of distributed agents hypothesis hunting.
• Figure 3: Evaluation of novelty, quality, and diversity of findings produced by research network.
• Figure 4: Heterogeneous agent behaviors and endogenous network evolution.
• Figure 5: Exploration trajectory of heterogeneous agents.