PapersSelected73°

Neural Networks Symbolic Structure

The Emergent Symbolic Structure of Artificial Neural Networks

Why we picked it

The research bridges traditional symbolic AI with modern neural networks, offering new interpretability methods for developers working with LLMs in ASEAN tech sectors.

The paper demonstrates that neural network vector representations can be approximated with symbolic structures across various models.

Body

This holds true for both small-scale neural networks trained on list manipulation and large language models in arithmetic, logic, computer code, and language domains.

The symbolic approximation allows precise interventions on LLM behavior, showing reliance on identified symbolic structures.

Original · arXiv cs.AI

The Emergent Symbolic Structure of Artificial Neural Networks

Modern systems in artificial intelligence (AI) somehow excel in domains for which they seem poorly suited. Intelligence has traditionally been modeled as operating over structured combinations of symbols, such as logical formulas. However, the strongest modern AI systems are based on neural networks, which instead represent information in continuous vectors. Vectors seem inadequate for capturing the structure of language, logic, and other cognitive domains, yet neural networks achieve impressive performance in these areas. How do they do it? In this work, we propose a potential answer: Despite appearances, perhaps the internal representations of neural networks implicitly realize symbolic structure. In support of this hypothesis, we show that the vector representations of a variety of neural networks can be closely approximated with symbolic structures: we can replace the network's entire representation-generating process with a closed-form equation instantiating a symbolic structure, and the network's behavior remains largely unchanged. This finding holds for both small-scale neural networks trained to manipulate lists as well as large language models (LLMs) operating in four domains that are central in symbolic traditions: arithmetic, logic, computer code, and language. Further, our symbolic approximation allows us to modify an LLM's behavior in targeted ways via precise interventions on its internal representations, showing that the LLM's behavior is reliant on the symbolic structures we have identified. This work provides a potential way to reconcile longstanding symbolic conceptions of intelligence with the vector-based nature of modern AI.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include neural networks, large language models, cognitive science.

Source attribution: curated from arXiv cs.AI (@R. Thomas McCoy, Paul Soulos, Tal Linzen, Paul Smolensky).