Currently this number is around 5-10% (https://www.lesswrong.com/posts/H6hMugfY3tDQGfqYL/what-do-ml-researchers-think-about-ai-in-2022).
I will take the mean of all surveys during the year of 2025 weighted by sample size, for all of the surveys that have no major methodological problems (as determined by discussion and/or voting in the comments of this market). Resolves to %.
The survey question has to specifically ask about the probability of AI-caused extinction.
Resolves N/A if no surveys happen during 2025, but if this looks like it's about to happen I might go and run my own survey by asking random people at some ML conference (no promises though).
Update 2026-01-24 (PST) (AI summary of creator comment): The creator has clarified that mean (not median) will be used for individual survey results, as the description specifies taking "the mean of all surveys" and a weighted mean of medians would be unusual.
The creator conducted a survey at ICLR 2025 with:
Mean: 18.7% (n=21)
Median: 10%
If no other surveys are brought to their attention in the next few weeks, the market will resolve using this survey result.
1,000I personally conducted a survey of p(doom) values at ICLR 2025. I found a mean of 18.7% (n=21) and a median of 10%. I forgot to specify whether I meant mean or median in the original post, but I did say the mean across surveys, and it would be kind of weird to do a weighted mean of medians, so I'm going to say that mean makes more sense.
If no other surveys are brought to my attention in the next few weeks I'll resolve it using this survey result.
Average or median? In the survey you linked, the p(extremely bad outcome eg human extinction) had an average of 14% and a median of 5%, and the 5-10% numbers you cited were for median p(human extinction)