The essay opens by acknowledging the impressive strides of generative AI, which have prompted some researchers to declare that artificial general intelligence is just around the corner. But the author pushes back, arguing that this optimism rests on a mistaken assumption: that thought can be reduced to language. Citing computer scientist Terry Winograd, the piece warns that projecting language back as the model for thought causes us to lose sight of the tacit, embodied understanding that underpins real intelligence.
According to the author, language models operate purely on symbolic patterns, with no connection to the physical world or the sensory-motor experiences that shape human reasoning. This gap is not a minor limitation but a fundamental one. Even if a model can process text, images, and audio simultaneously, it still lacks the lived, bodily engagement that gives meaning to those signals. The essay concludes that reaching AGI will require more than scaling up multimodal data—it will demand a new architecture that incorporates embodiment and action.