Curious Causality-Seeking Agents Learn Meta Causal World

2025-06-28

AI Summary

Real-world causal mechanisms often exhibit dynamic shifts across environmental states rather than adhering to a single stationary rule, limiting the generalization of conventional world models in non-stationary environments. To address this, we propose the Meta-Causal Graph (MCG) framework, which formalizes mechanism drift as latent meta-state-triggered switches among causal subgraphs, thereby unifying the representation of time-varying causal structure. We further design a curiosity-driven, intervention-based exploration agent that actively intervenes in the environment, analyzes system responses, and iteratively refines multiple subgraph-level causal models. The method integrates causal discovery, latent-variable modeling, and reinforcement learning. Evaluated on synthetic benchmarks and real-world robotic arm manipulation tasks, MCG achieves significantly improved robustness and generalization to unseen dynamic scenarios. Notably, it constitutes the first approach to enable learnable, meta-level representations of causal structure.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

When building a world model, a common assumption is that the environment has a single, unchanging underlying causal rule, like applying Newton's laws to every situation. In reality, what appears as a drifting causal mechanism is often the manifestation of a fixed underlying mechanism seen through a narrow observational window. This brings about a problem that, when building a world model, even subtle shifts in policy or environment states can alter the very observed causal mechanisms. In this work, we introduce the extbf{Meta-Causal Graph} as world models, a minimal unified representation that efficiently encodes the transformation rules governing how causal structures shift across different latent world states. A single Meta-Causal Graph is composed of multiple causal subgraphs, each triggered by meta state, which is in the latent state space. Building on this representation, we introduce a extbf{Causality-Seeking Agent} whose objectives are to (1) identify the meta states that trigger each subgraph, (2) discover the corresponding causal relationships by agent curiosity-driven intervention policy, and (3) iteratively refine the Meta-Causal Graph through ongoing curiosity-driven exploration and agent experiences. Experiments on both synthetic tasks and a challenging robot arm manipulation task demonstrate that our method robustly captures shifts in causal dynamics and generalizes effectively to previously unseen contexts.

Problem

• Modeling shifting causal mechanisms in dynamic environments
• Identifying latent meta states triggering causal subgraphs
• Learning adaptive world models through curiosity-driven exploration

Innovation

• Meta-Causal Graph as unified world model
• Causality-Seeking Agent with curiosity-driven policy
• Iterative refinement through curiosity-driven exploration