Whither symbols in the era of advanced neural networks?

2025-08-07

AI Summary

This paper revisits the foundational cognitive science question—whether human thought must be grounded in symbolic systems—in light of growing evidence that neural networks exhibit human-like cognitive capacities, including compositional reasoning, novel generation, and few-shot learning. Method: Adopting an interdisciplinary approach, the study empirically evaluates state-of-the-art AI models on these capabilities while critically examining their underlying mechanisms through the lens of cognitive theory. Contribution/Results: First, it demonstrates that purely connectionist systems can spontaneously exhibit symbol-like operations, challenging the necessity of explicit symbolic representations for high-level cognition. Second, it reveals that neural network training data is deeply rooted in human symbolic practices, indicating that symbolic systems continue to shape intelligent behavior indirectly—as meta-representations that define task structure and cognitive boundaries. Consequently, the paper proposes a new research agenda that moves beyond the traditional “symbolic vs. connectionist” dichotomy toward a symbiotic symbol–neural cognitive modeling paradigm.

Technology Category

• Cognitive Modeling & Cognitive Systems: Symbolic Representations
• Machine Learning: Neuro-Symbolic Learning
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Social networks and social learning

Abstract

Some of the strongest evidence that human minds should be thought about in terms of symbolic systems has been the way they combine ideas, produce novelty, and learn quickly. We argue that modern neural networks -- and the artificial intelligence systems built upon them -- exhibit similar abilities. This undermines the argument that the cognitive processes and representations used by human minds are symbolic, although the fact that these neural networks are typically trained on data generated by symbolic systems illustrates that such systems play an important role in characterizing the abstract problems that human minds have to solve. This argument leads us to offer a new agenda for research on the symbolic basis of human thought.

Problem

• Examining symbolic systems' role in human cognition
• Challenging necessity of symbols in neural networks
• Proposing new research agenda for symbolic thought

Innovation

• Neural networks exhibit symbolic-like abilities
• Training data from symbolic systems is crucial
• New research agenda for symbolic cognition