When will AIs be good at predicting the future?

1.3kṀ4986

2042

ALL

70%

2025

80%

2026

83%

2027

94%

2030

Each year category resolves to the percentage of people worse than the best AI at predicting the future.

To measure who is better or worse at predicting the future, I will look at how the best AI system did in the ACX prediction contest. In summary, it is about placing a bunch of predictions at once on relevant topics. You can read more about it here: https://www.metaculus.com/tournament/2023-contest/.

I might use a different contest if I find a better one.

Similar market about problem solving: https://manifold.markets/patrik/how-good-will-ais-be-in-solving-com.

Technology

Manifold

GPT-5

Get

1,000

to start trading!

People are also trading

Once AI can predict the future better than any human it will be AGI

70% chance

Will we have better-than-human-aggregate forecasting AIs by the end of 2024?

4% chance

When will AIs be good at solving complex problems? (read description)

Will AI be smarter than any one human probably around the end of 2025?

16% chance

Which AI will be the best at the end of 2025?

Which AI future will we get?

By 2026 will there be autonomous AI good enough that I use it?

37% chance

Will AI be able to accurately predict natural disasters with a week's notice by 2030?

5% chance

Will it take <=12 months for open-source AI forecasting to go from consistently worse than humans to broadly superhuman?

18% chance

Will an AI model be developed before 2030 that can accurately predict local weather patterns up to 6 months in advance?

Sort by:

https://x.com/Simeon_Cps/status/1878022286613074271

@MalachiteEagle Very interesting. Tho I think something like they used for openai o1 and o3 models but with questions that are after the training data could be even a better way.

I don’t understand the resolution yet, is it that you will use some AI systems to forecast the ACX forecasts and then compare the best performing one to the distribution of human results?