Memory, Consciousness and Large Language Model

TL;DR

The paper proposes a Tulving memory framework as a duality with Large Language Models, mapping semantic/episodic/procedural memories to pre-training/ input context/ prompting behaviors, respectively. It links the SEM of retrieval to emergent abilities and in-context learning, arguing that consciousness could emerge as an additional emergent ability given sufficient long-context memory. The authors discuss evidence, potential architectures like State Space Models to extend context, and the implications for AI safety. They call for direct experiments to quantify memory threshold curves and propose naming the duality the Tulving–LLM memory duality.

Abstract

With the development in cognitive science and Large Language Models (LLMs), increasing connections have come to light between these two distinct fields. Building upon these connections, we propose a conjecture suggesting the existence of a duality between LLMs and Tulving's theory of memory. We identify a potential correspondence between Tulving's synergistic ecphory model (SEM) of retrieval and the emergent abilities observed in LLMs, serving as supporting evidence for our conjecture. Furthermore, we speculate that consciousness may be considered a form of emergent ability based on this duality. We also discuss how other theories of consciousness intersect with our research.

Figures (2)

• Figure 1: Schematic diagrams of synergistic ecphory model of retrieval corresponding to (a) general emergent ability and (b) in-context learning.
• Figure 2: Schematic diagrams of synergistic ecphory model of retrieval corresponding to some unknown emergence ability.