Oxford's Carissa Véliz: The AI Future Doesn't Get Predicted — It Gets Produced
CarissaVeliz · x · 2026-09-20
Oxford researcher Carissa Véliz pushes back against recent tech-executive claims that AI has a >10% chance of destroying humanity by 2030, arguing in her Substack newsletter The Antidote that "the future doesn't get predicted — it gets produced."
Her framework for evaluating doomy predictions:
- Ask who is making the prediction: do they have conflicts of interest (money, fame), and who benefits from people believing it?
- Interrogate the evidence: is the number grounded in data or pulled out of thin air?
- Build historical perspective: humanity has long held the means of its own destruction, and such forecasts keep failing.
- She calls it naïveté about power that a single remark from an executive at a company about to go public can shift global conversation.
The essay is framed as an antidote to AI-doom hype, urging critical scrutiny over panic.
More from AGI Musings
- Jev founder Diogo Almeida: RLHF is an epic wrong turn on the road to automation — dotey · 2026-09-20
- "AI is like a microwave": the edge is knowing where it creates value — Olivier__OG · 2026-09-20
- Rob Leclerc to doomers: if ASI is god-like, why can't it accelerate human flourishing? — robleclerc · 2026-09-20
- From FTX to Airstrikes on Data Centers: EA's Accountability Problem — VraserX · 2026-09-20
- John Schulman: goal-driven research beats idea-driven research — and it's good life advice — tokenbender · 2026-09-20
- DeepMind's Nando de Freitas lays out 9 empowerment goals for AI, warns of weaponization — NandoDF · 2026-09-20