Grady Booch: Frontier Models Are Not Conscious and the Word Itself Is Useless
Grady_Booch · x · 2026-10-10
Software engineering pioneer Grady Booch argues that contemporary frontier models are not conscious, and that the word "consciousness" is too vague and subjective to be useful in rational discourse about mind and computing.
His alternative framing: frontier models are "complex stateful reactive stochastic software-intensive systems" with some self-referential reflection, capable of coherent but unreliable outputs — behaviors that are mostly taught, not learned.
He concedes humans are also complex, stochastic, and unreliable, but highlights key differences: we are mortal, we have agency, we generate abstractions, we build and test models of the world and ourselves, we deliberately identify gaps and self-improve, and we are composed of massive numbers of parallel agents from cells to mind.
Booch asserts the mind is computable, but says we may be generations away — lacking both proper architectures and even the understanding needed to ask the right design questions. Still, he marvels at living in an era where those questions can finally be posed.
More from AGI Musings
- Managers may be prepped for herds of agents, but does the reverse hold? — nlarusstone · 2026-10-10
- Dev argues giving AI rights is the worst misalignment path, says Anthropic policies raise his p(doom) — robleclerc · 2026-10-10
- Adversarial work means demand for frontier intelligence never scales down — abhiadesai · 2026-10-10
- Anthropic cuts live internet access for internal evals, citing unreliable AI agent control — TechCrunch AI · 2026-10-10
- AI artist: critics deny my work is mine just because I took creative liberties — eschadiol · 2026-10-10
- "It's 2026 and people still excuse AI hallucination because humans lie too," says ML author Burkov — burkov · 2026-10-10