Latent State Design for World Models under Sufficiency Constraints

TL;DR

This paper introduces a functional taxonomy for designing latent states in world models, emphasizing task-specific sufficiency over architecture or domain, and provides an evaluation framework to assess models based on their intended roles.

Contribution

It proposes a new taxonomy categorizing world model methods by their latent state's purpose, and develops an evaluation framework to diagnose model capabilities and limitations.

Findings

A taxonomy that groups methods by latent state roles reveals key distinctions.

Evaluation along seven axes helps diagnose what information is preserved or discarded.

Actionable world models should match the task-specific construction of their latent states.

Abstract

A world model matters to an agent only through the state it constructs. That state must preserve some information, discard other information, and support some future function: prediction, control, planning, memory, grounding, or counterfactual reasoning. This paper treats world-model research as latent state design under sufficiency constraints. We propose a functional taxonomy that groups methods by what their latent state is for, rather than by architecture or application domain: predictive embedding, recurrent belief state, object/causal structure, latent action interface, grounded planning interface, and memory substrate. These roles expose distinctions that architecture-based groupings hide, including the gap between predictive sufficiency and control sufficiency, and the gap between passive video prediction and counterfactual action modeling. The taxonomy supports an evaluation framework that judges a model by the sufficiency constraint its latent state was built to satisfy. We compare methods along seven axes: representation, prediction, planning, controllability, causal/counterfactual support, memory, and uncertainty. We use the resulting matrix as a diagnostic for what a latent state preserves, discards, and enables. The conclusion that follows is that an actionable world model is the one whose state construction matches the task, not the one that preserves the most information.