Separable Pathways for Causal Reasoning: How Architectural Scaffolding Enables Hypothesis-Space Restructuring in LLM Agents

TL;DR

This paper introduces a compositional architecture for AI agents that enables hypothesis-space restructuring during causal reasoning, improving robustness in problem solving.

Contribution

It presents a novel architecture with context graphs and dynamic behaviors that facilitate hypothesis revision in AI agents, inspired by developmental science.

Findings

Context graphs account for 94% of accuracy gain in reasoning.

Dynamic behaviors detect regime changes and prevent premature hypothesis commitment.

The architecture improves causal reasoning in 1,085 experimental trials.

Abstract

Causal discovery through experimentation and intervention is fundamental to robust problem solving. It requires not just updating beliefs within a fixed framework but revising the hypothesis space itself, a capacity current AI agents lack when evidence demands representations they have not previously constructed. We extend the blicket detector paradigm from developmental science to test this capacity in AI agents equipped with architectural scaffolding that targets hypothesis-space restructuring. Our compositional architecture has two discrete components: context graphs, which structure exploration as typed state machines, and dynamic behaviors, which monitor for evidence that the current hypothesis space is inadequate and expand it at runtime. Across 1,085 experimental trials, these components make orthogonal contributions: context graphs drive reasoning quality within the post-switch hypothesis space, accounting for 94\% of the accuracy gain, while dynamic behaviors drive reasoning eligibility by detecting regime changes and preventing premature commitment to outdated hypotheses.