Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Human-Like Lifelong Memory: A Neuroscience-Grounded Architecture for Infinite Interaction

Diego C. Lerma-Torres

Abstract

Large language models lack persistent, structured memory for long-term interaction and context-sensitive retrieval. Expanding context windows does not solve this: recent evidence shows that context length alone degrades reasoning by up to 85% - even with perfect retrieval. We propose a bio-inspired memory framework grounded in complementary learning systems theory, cognitive behavioral therapy's belief hierarchy, dual-process cognition, and fuzzy-trace theory, organized around three principles: (1) Memory has valence, not just content - pre-computed emotional-associative summaries (valence vectors) organized in an emergent belief hierarchy inspired by Beck's cognitive model enable instant orientation before deliberation; (2) Retrieval defaults to System 1 with System 2 escalation - automatic spreading activation and passive priming as default, with deliberate retrieval only when needed, and graded epistemic states that address hallucination structurally; and (3) Encoding is active, present, and feedback-dependent - a thalamic gateway tags and routes information between stores, while the executive forms gists through curiosity-driven investigation, not passive exposure. Seven functional properties specify what any implementation must satisfy. Over time, the system converges toward System 1 processing - the computational analog of clinical expertise - producing interactions that become cheaper, not more expensive, with experience.

Human-Like Lifelong Memory: A Neuroscience-Grounded Architecture for Infinite Interaction

Abstract

Large language models lack persistent, structured memory for long-term interaction and context-sensitive retrieval. Expanding context windows does not solve this: recent evidence shows that context length alone degrades reasoning by up to 85% - even with perfect retrieval. We propose a bio-inspired memory framework grounded in complementary learning systems theory, cognitive behavioral therapy's belief hierarchy, dual-process cognition, and fuzzy-trace theory, organized around three principles: (1) Memory has valence, not just content - pre-computed emotional-associative summaries (valence vectors) organized in an emergent belief hierarchy inspired by Beck's cognitive model enable instant orientation before deliberation; (2) Retrieval defaults to System 1 with System 2 escalation - automatic spreading activation and passive priming as default, with deliberate retrieval only when needed, and graded epistemic states that address hallucination structurally; and (3) Encoding is active, present, and feedback-dependent - a thalamic gateway tags and routes information between stores, while the executive forms gists through curiosity-driven investigation, not passive exposure. Seven functional properties specify what any implementation must satisfy. Over time, the system converges toward System 1 processing - the computational analog of clinical expertise - producing interactions that become cheaper, not more expensive, with experience.

Paper Structure

This paper contains 43 sections, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Context expansion is insufficient.
  3. Related work.
  4. Scope.
  5. Neuroscientific and Clinical Foundations
  6. Complementary learning systems.
  7. Emotion as compressed computation.
  8. The CBT belief hierarchy as emergent pattern.
  9. Curiosity-driven learning and dual-process expertise.
  10. Conviction tracking and reconsolidation.
  11. Reconstructive memory and graded epistemic states.
  12. Architectural Overview
  13. Executive Function and Working Memory
  14. Memory Service: Knowledge Graph
  15. Implementation notes.
  16. ...and 28 more sections

Figures (1)

  • Figure 1: Architecture overview. Working memory resides within the executive---it is the LLM's active workspace, not part of the memory store. Identity emerges from high-weight gists that the gateway selects frequently into working memory. The memory service exposes two access levels: gists/valence vectors (cheap O(1) lookup used by System 1 via the gateway) and the full graph (expensive traversal used by System 2 directly). System 2 bypasses gateway thresholds for deliberate search, analogous to direct PFC--hippocampal circuits. A compression mechanism (implementation-dependent) degrades content resolution over time while preserving valence vectors.