Association Memory Model Using The Notion Of "Importance"

M. Weinzweig

Abstract

This paper, authored by M.N. Weinzweig of the USSR Academy of Sciences, proposes a theoretical framework for human memory that diverges from the rigid hierarchical structures common in early computer science. The model introduces a dual-memory system—Principal and Contextual—where information retrieval is governed by "importance," defined as the subjective probability that a memory will be useful in a given context. By grounding this theory in a proposed molecular mechanism, the author explains how the brain achieves rapid, parallel associative search through thermal motion and surface complementarity rather than sequential scanning.

Introduction

In the early development of artificial intelligence and cognitive modeling, memory was frequently conceptualized as a hierarchical tree. In such systems, retrieving a specific piece of information required sequential checks at every branching node, a process that becomes increasingly inefficient as the database grows. Human memory, however, is remarkably fast and associative. We do not appear to "scan" our entire life history to recall a friend's name; rather, the information seems to surface spontaneously.

Weinzweig addresses the "mechanism of paying attention," seeking to explain how the brain decides which specific memory to pull into the field of consciousness at any given moment. The challenge lies in explaining how a massive database of associations can be navigated near-instantaneously while accounting for the shifting influence of current context.

The Dual-Memory Framework

The proposed model relies on two distinct types of memory:

  1. Principal Memory (Long-term): A vast, permanent storehouse of statements, notions, and experiences.
  2. Contextual Memory (Short-term): A temporary buffer containing information derived from immediate sensory input and recent thoughts.

The bridge between these two stores is the mathematical treatment of "Importance." Every element within the memory system is assigned an importance value. This is not a vague qualitative feeling but a functional parameter representing the subjective probability that the element will be necessary to solve the current problem or navigate the current situation.

Departing from digital indexing metaphors, Weinzweig suggests that memory is physically realized through molecular interactions. Information is stored as "words-molecules" (such as DNA, RNA, or proteins). In this "molecular soup," retrieval is powered by random thermal movement.

When molecules collide, they may stick together if their surfaces are "complementary" (a lock-and-key mechanism). This physical binding triggers the retrieval of the memory. This allows for a massive, parallel search where the "right" answer literally bumps into the "question" and adheres to it. High "importance" effectively increases the likelihood or frequency of these successful collisions, allowing relevant information to reach the field of attention faster.

Context and the "Pete" Example

To illustrate the role of context, the author uses the example of being asked, "Have you seen Pete?" Under a hierarchical model, the brain might have to sort through every individual named Pete ever encountered. In Weinzweig’s model, the current context—the identity of the speaker, the physical location, and recent events—acts as a filter. This context assigns a significantly higher importance value to the "correct" Pete, making that specific image surface immediately while irrelevant Petes remain in the background.

Cognitive Implications and Results

The model provides a qualitative framework for several observed cognitive phenomena:

  • Associative Efficiency: It explains why biological entities favor associative memory over hierarchical structures, as it avoids the "hopeless ways" of searching through irrelevant branching trees.
  • Frustration of Attention: The model accounts for moments when attention is captured by irrelevant notions. This occurs when a random thought or external stimulus happens to have a high temporary importance or a strong accidental molecular connection.
  • Physical Plausibility: By citing the "molecular computer" as a basis, the author provides a potential bridge between psychological theories of attention and the physical reality of cellular chemistry.

Limitations and Future Directions

As a theoretical and conceptual paper, this work does not rely on empirical datasets or benchmarks. Its primary limitation is the lack of experimental data to confirm that memory is stored or retrieved via molecular complementarity. Furthermore, the author acknowledges a scale gap: while memory storage may be molecular, the "consciousness mechanism" that processes this information likely operates at the neuronic level. The specific interface between molecular records and the electrical firing of neurons remains a primary question for future exploration.

Summary updated

BibTeX
@inproceedings{weinzweig1975association,
  author = {M. Weinzweig},
  title = {{Association Memory Model Using The Notion Of "Importance"}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1975},
}
Figure