Artifacts as Memory Beyond the Agent Boundary

arXiv cs.AI / 4/13/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper argues, within reinforcement learning, that an agent’s environment can function as a form of external memory, supporting the “situated cognition” view of cognition.
It introduces a formal mathematical framework showing how certain environmental observations (“artifacts”) can reduce the information needed to represent an agent’s history.
Experiments indicate that when agents observe spatial paths, they can learn effective policies while requiring less internal memory.
The memory-reduction benefit can emerge unintentionally from the agent’s sensory stream rather than from explicit architectural design.
The authors connect their results to qualitative properties used to justify external-memory accounts and propose future methods to deliberately exploit environmental substitution for internal memory.

Abstract

The situated view of cognition holds that intelligent behavior depends not only on internal memory, but on an agent's active use of environmental resources. Here, we begin formalizing this intuition within Reinforcement Learning (RL). We introduce a mathematical framing for how the environment can functionally serve as an agent's memory, and prove that certain observations, which we call artifacts, can reduce the information needed to represent history. We corroborate our theory with experiments showing that when agents observe spatial paths, the amount of memory required to learn a performant policy is reduced. Interestingly, this effect arises unintentionally, and implicitly through the agent's sensory stream. We discuss the implications of our findings, and show they satisfy qualitative properties previously used to ground accounts of external memory. Moving forward, we anticipate further work on this subject could reveal principled ways to exploit the environment as a substitute for explicit internal memory.