Neuromorphic and Neuro-Symbolic Computing

These two terms sound similar, but they describe different approaches.

ApproachCentral IdeaExamples
Neuromorphic ComputingBuild hardware and algorithms that imitate features of biological nervous systems, often using spikes, local memory, and event-driven computation.Spiking neural networks, Intel Loihi, IBM TrueNorth
Neural AILearn statistical patterns from large amounts of data.Deep learning, transformers, large language models
Symbolic AIRepresent knowledge explicitly and reason with rules.First-order logic, theorem provers, Cyc
Neuro-Symbolic AICombine learned neural models with explicit symbols and logical reasoning.An LLM connected to Cyc or another theorem prover

Neuromorphic systems are primarily about how computation is physically and algorithmically organized. Cyc and LLM combinations are better described as neuro-symbolic: the LLM supplies neural language capabilities, while Cyc supplies explicit knowledge and logic.

These ideas can eventually be combined—a neural component could run on neuromorphic hardware while a symbolic component performs logical inference—but Cyc itself is not a neuromorphic computer.