Survey of 2,778 AI researchers puts 50% odds on machines outperforming humans by 2047
gleech · x · 2026-09-10
A debate over what counts as a "doomer" cites the largest AI researcher survey on arXiv (2,778 authors at top venues):
- ≥50% chance of milestones like autonomously building a payment site or fine-tuning an LLM by 2028;
- 10% odds by 2027 and 50% by 2047 for unaided machines beating humans at all tasks — 13 years earlier than the previous year's survey;
- Full automation of all human occupations: 10% by 2037, 50% by 2116;
- 68.3% see good outcomes as more likely, yet 48% of those optimists still give ≥5% to extinction-level bad outcomes.
The poster notes 58% of respondents put ≥5% on extinction, so under a "≥1% extinction probability = doomer" definition most researchers qualify.
More from AGI Musings
- Daniel Faggella's 'Good Monster': optimizing any human 'good' may imply humanity bowing out — danfaggella · 2026-09-21
- RL pioneer Csaba Szepesvari: humans never had control, and AI deepens the illusion — CsabaSzepesvari · 2026-09-21
- RL veteran Csaba Szepesvári: the illusion that things are under our control was always false — CsabaSzepesvari · 2026-09-21
- Most Researchers Still Treat Transformers as Black Boxes, and Public Understanding Is Decades Away — gerardsans · 2026-09-21
- In the Age of AI Slop, a 1939 Essay Argues Useless Knowledge May Not Be So Useless — xkonjin · 2026-09-21
- Csaba Szepesvári: AI Risk Stems From Unknown Unknowns, Not Ignorance of How Models Work — CsabaSzepesvari · 2026-09-21