A Machine With Human-Like Memory Systems

TL;DR

The paper investigates how to equip AI agents with human-like memory by explicitly modeling semantic and episodic memories within a dynamic environment. It introduces the Room, an OpenAI Gym–style setting where agents encode, store, and retrieve memories to maximize rewards, and demonstrates that dual-memory agents outperform single-memory ones. It shows that pretrained semantic knowledge further improves generalization and that collaboration between multiple agents yields additional gains through complementary memory coverage. The work advances hybrid intelligence by providing both a testbed and empirical evidence that memory structure and collaboration can meaningfully enhance AI question answering in changing environments.

Abstract

Inspired by the cognitive science theory, we explicitly model an agent with both semantic and episodic memory systems, and show that it is better than having just one of the two memory systems. In order to show this, we have designed and released our own challenging environment, "the Room", compatible with OpenAI Gym, where an agent has to properly learn how to encode, store, and retrieve memories to maximize its rewards. The Room environment allows for a hybrid intelligence setup where machines and humans can collaborate. We show that two agents collaborating with each other results in better performance than one agent acting alone.

Figures (3)

• Figure 1: Handcrafted vs. random policies
• Figure 2: Total rewards with respect to different handcrafted policies and memory capacities.
• Figure 3: Total rewards with respect to the number of agents. The lighter and narrower bars account for the single agent.