Answers are grounded in this paper and cite their sources.
Many domains of cognition, such as language and reasoning, have traditionally been modeled with symbolic systems. However, in both biological and artificial intelligence, some of the most advanced systems in these domains are neural networks, despite the fact that neural networks appear fundamentally different from symbolic systems. What, then, is the relationship between neural computation and symbolic behavior? In this paper, we review recent literature that analyzes the internal mechanisms used by neural networks from AI. We conclude that this line of work supports a quasi-symbolic view of cognition, in which the mind uses computation that is approximately symbolic but with some important non-symbolic properties that derive from the mind’s neural underpinnings.
No takes yet. Share an insight, caveat, or question.
McCoy et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: