Position: Hippocampal Explicit Memory Is the Cornerstone for AGI

The gist The underlying learning mechanism of LLMs is highly analogous to human implicit memory, and realizing Artificial General Intelligence should involve the integration of explicit memory.

Explicit and Implicit Memories

Human memory is structured to store and utilize multiple types of information, enabling various cognitive functions. Explicit memory handles information related to facts, events, or experiences, while implicit memory operates mainly by influencing performance and behavior through unconscious learning. Explicit memory is typically divided into two main categories: episodic memory and semantic memory. Episodic memory preserves personal experiences tied to specific times and places, while semantic memory refers to the memory of general knowledge, concepts, language, and facts.

Formation Mechanism

The formation of explicit memory occurs through strong synaptic strengthening in the main neural circuits of the hippocampus and medial temporal lobe when new events are encountered. The process involves several stages:

1. In the dentate gyrus (DG), neural signals are sparsely encoded, selectively activating a small subset of neurons and thereby strongly inducing long-term potentiation (LTP).

2. Next, the CA3 region binds these sparsely active neurons through its recurrent collateral circuit, forming an autoassociative network.

3. Finally, the CA1 region serves as the output stage of the hippocampus (Nakazawa et al., 2004), integrating information received from CA3 and transmitting it to the cerebral cortex.

Implicit memory is primarily formed in subcortical structures such as the basal ganglia, cerebellum, and amygdala. The main stages of learning occur as follows:

1. Prediction: Anticipating a reward based on performing a specific behavior in the current context.

2. Action: When the behavior is performed, signals from the cortex are densely transmitted to the striatal MSNs.

3. Outcome Evaluation: Assessing whether the behavior resulted in a reward, failure, or punishment.

Retrieval Mechanism

Explicit memory retrieval centers on the hippocampal-cortical loop, where contextual cues are essential. The process is outlined as follows:

1. A contextual cue or intention initiates the hippocampalcortical loop.

2. Signals from the EC activate specific neurons in the CA3 region.

3. The CA3 region’s recurrent circuitry reconstructs the entire memory pattern based on the activated neurons.

4. The reconstructed pattern is relayed through the CA1 region to the cerebral cortex, synchronizing the elements of the memory.

Implicit memory retrieval happens automatically (Graybiel, 2008). When a familiar stimulus appears, stored stimulus-response patterns in the striatum are activated via MSNs. In short, implicit memory retrieval reactivates learned behaviors automatically in response to familiar cues.

Nature of Learning in LLMs

The learning and reasoning processes of LLMs reveal key distinctions from hippocampal explicit memory, indicating that the core characteristics more closely resemble implicit learning and habituation. Key similarities include:

** Gradual Learning:**

Explicit memory can encode new episodes vividly with just one exposure when strong stimuli are present or attention is engaged. In contrast, implicit memory relies on gradually accumulating associations between stimuli, actions, and rewards (or punishments). Similarly, LLM training follows a comparable pattern where model parameters adjust their weights cumulatively by repeatedly observing numerous examples within a vast corpus.

Dense and Distributed Coding:

Explicit memory is characterized by sparse coding to ensure pattern separation. In contrast, subcortical circuits receive dense signals from the cortex. LLMs exhibit a similar structure, with their network broadly interconnected.

Error-Driven Learning:

Implicit learning involves dopamine neurons strengthening or weakening synapses based on reward prediction errors. In contrast, learning in explicit memory relies on the associations of events or pieces of information, rather than on an error-based mechanism. The loss function in LLMs works in a similar way to dopamine signals, adjusting model parameters through backpropagation based on the difference between the actual output and the ground truth across large datasets.

Functional Specificity of Memory

Explicit and implicit memory serve distinct cognitive roles. Key differences include:

** Pattern Recognition:**

Pattern recognition is the ability of humans to identify specific rules or consistencies in a complex environment, which is rooted in the implicit memory system.

Language Ability:

Fluent communication heavily relies on implicit memory, while semantic memory supports language processing and categorization. Language ability can be understood as relying on a dual memory system, where explicit memory processes new information during the initial stages of learning, and repeated use transitions it into implicit memory for automation.

Logical Reasoning:

Mathematical and logical reasoning rely heavily on semantic memory, a subset of explicit memory. The difficulty of logical reasoning lies in the fact that the core logical concepts and their relationships are so abstract that the possible variations of corresponding low-level representations can be practically infinite.

Executive Function:

Explicit memory is a crucial factor in executive functions through the retrieval of past experiences, which supports complex cognitive processes such as planning and decision-making. Executive functions heavily rely on short-term storage, often referred to as working memory.

Metacognition:

Metacognition entails the ability to distinguish between what one knows and does not know, and it relies on explicit mechanisms in predicting memory retrieval.

Need for Explicit Memory

Current LLMs face limitations that closely align with the inherent constraints of implicit memory, which lacks mechanisms for rapid and context-rich learning. Explicit memory offers a potential pathway to overcome these challenges by enabling higher-order cognitive abilities of humans. Key limitations include:

Higher-Order Learning:

Humans utilize dynamic learning, continuously acquiring new knowledge over time, which is not strictly confined to the training stage of LLMs.

Computational Requirements for Artificial Explicit Memory System

An explicit memory system is defined as a function fmemory: fmemory: I → O, where I = (E, M) and O = (∆M, Y). The internal structure of M is specified incrementally through the following requirements:

1. Sparse Indexing: The memory system generates a sparse index from the dense input, where most elements are zero.

2. Associative Construction: The memory state M further includes an associative matrix A ∈ R n×n that captures relationships among sparse indices <ref:2606.

1. Bold Header: Explicit Memory Integration for Higher-Order Cognition

This improvement involves integrating an explicit memory module defined by a function fmemory: I → O, which includes sparse indexing (S = sparsify(E)) and associative construction (A), allowing the LLM to store and retrieve episodic and semantic information to enable dynamic learning, long-term reasoning, and metacognitive tasks essential for AGI.

2. Bold Header: Enhanced Logical Reasoning Stability

The system can achieve robust mathematical reasoning by moving beyond implicit statistical learning through explicit memory, addressing the paper's finding that LLMs struggle with problems that involve unfamiliar formulations or require flexible abstraction beyond their training distribution. This allows the model to apply the fundamental arithmetic principle of addition explicitly rather than relying on implicit, probabilistic associations.

3. Bold Header: True Metacognitive Self-Correction

By implementing an explicit memory system capable of tracking the source and context of one’s knowledge, the AI can move beyond pattern-learned self-correction (a pattern learned through multitask learning and RLHF processes) to achieve true metacognition, allowing it to perform self-assessment even in unfamiliar environments.

4. Bold Header: Contextual Knowledge Retrieval via Pattern Completion

The system can solve complex, novel knowledge retrieval tasks by utilizing the explicit memory mechanism where the CA3 region’s recurrent circuitry reconstructs the entire memory pattern based on the activated neurons, allowing it to perform pattern completion on partial input instead of relying solely on implicit stimulus-response mappings.

5. Bold Header: Substrate-Independent Cognitive Blueprint

The improved AI system will be designed using an architecture that abstracts the proven structural blueprint of the human brain—specifically the hippocampus and explicit memory—allowing it to realize AGI regardless of the underlying physical substrate, as supported by the principle that higher cognitive functions emerge from computational operations rather than from any specific physical substrate.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various tasks, raising expectations for Artificial General Intelligence (AGI). This position paper argues that integrating explicit memory is the cornerstone for advancing LLMs toward AGI. The key reason is that the underlying learning mechanism of LLMs is highly analogous to human implicit memory. However, higher-order cognitive functions necessary for AGI, such as long-term strategic planning, metacognition, and symbolic reasoning, heavily rely on hippocampal explicit memory and cannot arise solely from implicit statistical learning. Drawing on findings from neuroscience, I advance this perspective and complement it with computational requirements for artificial explicit memory systems, hoping to foster further research and lay the groundwork for explicit memory integration.