Deep Learning Alone Won't Lead to AGI

GaryMarcus · x · 2026-07-15

The shared post argues that deep learning researchers generally misunderstand the limits of statistical learning, believing it cannot achieve true general intelligence.

The core thesis is that AGI won't emerge merely from deep learning training based on "interpolation within bounded domains," because this mechanism fundamentally cannot learn abstraction and true extrapolation. It cites Gary Marcus's 2018 arguments as support.

Original post →

More from AGI Musings

AGI Musings channel →