Floridi et al. prove AI can't have guaranteed correctness and open-ended generality at once
rvp · x · 2026-09-27
- Luciano Floridi, Alberto Messina et al. publish "What Generality Costs in Artificial Intelligence" on SSRN.
- They propose the Certainty–Scope Conjecture, using information theory to mathematically show that with a finite computational budget, an AI system can achieve guaranteed correctness or an open-ended operational scope — but not both.
- The implication: hallucinations and similar systemic errors are not temporary engineering hurdles but an unavoidable mathematical trade-off of open-ended generality.
- The authors frame this as a reality check against the narrative of all-powerful universal AI models.
More from AGI Musings
- Gary Marcus: Everyone would freak out if Chinese AI did what OpenAI is doing — GaryMarcus · 2026-09-27
- Judea Pearl's Caltech talk: Can Computers Have Free Will? — yudapearl · 2026-09-27
- Researchers debate interpretability as a hedge against latent reasoning architectures — ChrisGPotts · 2026-09-27
- Inside Berkeley's conference on whether AI can feel pain — and why one philosopher says hold off on AI personhood — NathanpmYoung · 2026-09-27
- Terry Tao went from calling o1 a 'mediocre grad student' to fearing AI will devour math academia — aran_nayebi · 2026-09-27
- Tech insiders privately concede AI may 'kill billions' while the public assumes life goes on — birchlse · 2026-09-27