Grady Booch: Contemporary AI Still Lacks Abductive Reasoning, Just 'Next-Token Prediction'
Grady_Booch · x · 2026-09-23
Software engineering pioneer Grady Booch doubled down in a debate on LLM reasoning, insisting that the architecture of every transformer-based LLM is demonstrably next-token prediction and that his interlocutor shifted the subject.
His core claim: while contemporary AI can perform induction and deduction, there is no evidence yet of abductive reasoning — forming best-explanation hypotheses. Generalization beyond a training set, he argues, still "requires some magic fairy dust that remains undiscovered."
He also notes that "intelligence," "consciousness," and "sentence" are suitcase words, and without shared definitions the two sides keep talking past each other, ending with the question: can contemporary AI do abductive reasoning at all?
More from AGI Musings
- Dustin Moskovitz: the single largest funder of the EA ecosystem, with billions to AI safety — SydSteyerhart · 2026-09-23
- Open-source advocates call doom narratives a regulatory moat against open weights — AlexTensor · 2026-09-23
- When AI science bifurcates from human science, like math seems set to do — burny_tech · 2026-09-23
- Show today's LLMs to experts 10 years ago and they'd call it AGI — JacksonKernion · 2026-09-23
- Benedict Evans: AI may automate much of junior lawyers' and consultants' routine work — rohanpaul_ai · 2026-09-23
- AI safety will follow engineering tradition: formal proofs for simple cases, evals for complex — burny_tech · 2026-09-23