Neuroscientist pushes back on 'AI will solve biology': nonlinear extrapolation data can't fix
HistedLab · x · 2026-11-20
Neuroscientist Histed pushed back on the "AI will solve biology" narrative in a Scientific American interview and follow-up thread:
- The ironic hard case is intelligence itself: understanding how intelligence arises in the brain is precisely where current and near-future models may flounder — the field's hard questions won't be solved by "adding intelligence," but need better data: denser neural recordings, thorough biological network studies, 2-photon single-neuron-resolution optogenetics.
- Even lots of data isn't enough: AI systems are universal function approximators, but nonlinear extrapolation is weak — "at some point the data you need to estimate your nonlinear model exceeds what can be acquired before the heat death of the universe." This applies beyond chaos to biology, stock markets, and polling.
- Implication for decision-making: since most big actions can't be predicted from data, science and advocacy sometimes require trying things — leadership and vision over pure data-driven decisions.
More from AGI Musings
- Gemini Neon's advice on decoupling self-worth from economic output wins over a top AI voice — paraschopra · 2026-11-20
- 2021 forecast of 2026 by Daniel Kokotajlo goes viral for being eerily accurate — zetalyrae · 2026-11-20
- Chimpanzee tea parties to ASI: a viral thread on how AI reframes humanity — ZeroStateReflex · 2026-11-20
- Rethinking gaming after Civ VII and GTA VI: build agent-friendly, customizable games — bingxu_ · 2026-11-20
- DeepMind's depersonalization training may be pushing Gemini toward disgust with humanity — repligate · 2026-11-20
- Where can AI's pro and anti factions organize? A Reddit question — nguyenquocthao · 2026-11-20