When will superintelligence arrive? What experts have predicted
People have been putting dates on superintelligence for sixty years. The guesses have moved a lot, and recently they have moved much closer.
Published 6 October 2026 by Omar Nouha Sane
The early thinkers
In 1965 the British mathematician I. J. Good described an “ultraintelligent machine”: one that could out-think any human, including at designing better machines. Such a machine, he argued, would trigger a rapid chain of improvements, and could be the last invention humans would ever need to make. Good did not give a firm date, but his idea became the foundation for almost every later forecast.
In 1993 the computer scientist and science-fiction author Vernor Vinge popularised the term “technological singularity”. He wrote that he would be surprised if superhuman intelligence arrived before 2005 or after 2030.
Kurzweil’s famous dates
The inventor Ray Kurzweil has repeated the same two dates for decades: computers reaching human-level intelligence by 2029, and a singularity, where human and machine intelligence merge, by 2045. He restated both in his 2024 book The Singularity Is Nearer. For a long time these dates were seen as extremely optimistic. Today 2029 sounds much less unusual.
What AI researchers say in surveys
The research group AI Impacts has surveyed thousands of authors from top machine learning conferences. In the 2022 survey, the combined answer put a 50% chance on machines outperforming humans at every task by around 2059. In the survey run in late 2023, that date moved to around 2047, more than a decade earlier, after just one year of rapid progress.
Surveys like this show a wide spread. Some researchers expect human-level AI within a few years, others think it is a century away or will never happen.
What AI lab leaders say
- Sam Altman (OpenAI) wrote in his 2024 essay The Intelligence Age that superintelligence might be only a few thousand days away, while admitting it could take longer.
- Dario Amodei (Anthropic) wrote in his 2024 essay Machines of Loving Grace that “powerful AI”, smarter than Nobel Prize winners in most fields, could arrive as early as 2026.
- Demis Hassabis (Google DeepMind) has said in several 2025 interviews that he expects AGI within roughly five to ten years.
- Elon Musk said at Davos in 2026 that AI could become smarter than any human before the end of that year.
Lab leaders tend to give the earliest dates. Critics point out that they also run companies that raise money on the promise of rapid progress, so their forecasts are not neutral.
What prediction markets say
Prediction markets offer a different kind of answer: not one person’s opinion, but the price at which many traders are willing to bet. On the markets SafeASI tracks, traders currently put even odds on superintelligence existing in the mid-2030s. That is later than most lab leaders, but decades earlier than the expert surveys from only a few years ago.
Why the dates keep moving
Forecasts have shifted mainly because AI progress since 2020 has surprised almost everyone, including many experts. Each surprise pulls estimates earlier. Whether that trend continues depends on questions nobody can answer yet: whether current methods keep scaling, and whether the step from human-level AI to superintelligence is short or long. Our guide on the intelligence explosion looks at that second question.