What does a 36% chance actually mean?
Probabilities are easy to misread. A short guide to understanding the numbers on the tracker, and what they do not say.
Published 6 October 2026 by Omar Nouha Sane
36% is not “no”
A 36% chance means that, if you could replay the world many times, the event would happen in about a third of them. That is far from unlikely. A dice roll of five or six has a 33% chance, and nobody would be surprised to see it.
“By 2040” includes everything before
The yearly markets on SafeASI ask whether superintelligence will exist by a certain year. The chances are cumulative: 75% by 2040 already includes the 36% by 2030. The chance that it first arrives between 2030 and 2040 is therefore roughly 75% minus 36%, about 39%.
Why 75% by 2040 and 76% by 2050 is not a mistake
At first glance it looks odd that a whole extra decade adds only about one percentage point. Read cumulatively, the markets are saying: if superintelligence has not arrived by 2040, traders do not expect it to arrive soon after. Possible reasons include a belief that progress either happens fast or stalls for a long time, a feeling that the question might never resolve clearly, or simply that these long-dated markets have few traders and are not perfectly consistent with each other.
The 50% point
The big year on the tracker is where the curve crosses 50%: the point where traders think superintelligence is as likely to exist as not. It is not the most likely single year, and it is not a deadline. It is a midpoint. The methodology page shows how it is calculated.
Common mistakes
- Reading a forecast as a promise. A 75% forecast that does not happen was not necessarily wrong. One in four times, it should not happen.
- Ignoring the definition. Each market defines superintelligence in its own words. Two similar-looking questions can measure different things.
- Over-reading small moves. A market with twenty traders can move a few points because of a single bet. Look at trends, not daily wiggles.
- Forgetting the condition. These numbers assume no catastrophe happens first. They are not estimates of everything that could happen.
Calibration
A forecaster is well calibrated if events they call 70% likely happen about 70% of the time. That is the best test of forecasting quality, but it needs many resolved questions. For superintelligence there are none yet, so the numbers here are best understood as today’s collective best guess. More on that in What is a prediction market?