Gary Marcus Defends Neurosymbolic AI: Pure Neural Nets Won't Reach AGI

GaryMarcus · x · 2026-08-02

Gary Marcus responds to recent debates on AI reasoning, defending his long-standing neurosymbolic approach. He argues that any system combining neural networks with symbolic operations (like tools and harnesses) is a hybrid, not a pure neural network.

He emphasizes that if OpenAI or any other lab achieves AGI entirely without tools, harnesses, or symbolic computation, it would completely refute his theories. However, he notes that current systems (e.g., using the Lean prover for math) still fundamentally rely on symbolic components.

Related event: AI Developers Clash with Gary Marcus Over LLM Capabilities and Neurosymbolic AI(13 posts)→

Original post →

More from AGI Musings

AGI Musings channel →