https://www.astralcodexten.com/p/introducing-ai-2027
This market resolves in January 2027. It resolves YES if the AI Futures Project's predictions seem to have been roughly correct up until that point. Some details here and there can be wrong, just as in Daniel's 2021 set of predictions, but the important through-lines should be correct.
Resolution will be via a poll of Manifold moderators. If they're split on the issue, with anywhere from 30% to 70% YES votes, it'll resolve to the proportion of YES votes. Otherwise it resolves YES/NO.
@YuxiLiu This market is only on the up through end of 2026 portion of the scenario, and more importantly, it doesn't have to be precisely accurate to resolve yes. The prediction written in 2021 that got "the important through-lines ... correct" had incorrect details such as massive changes to our epistemic landscape online due to automated propaganda.
I mostly see this market as similar to predicting if the METR trendline of increasing task horizons plays out through EoY 2026, & reasoning paradigm models continue making impressive improvements on coding/math/verifiable problem solving. There's a bit more that has to happen, but not a ton.
People making these kinds of AI predictions dramatically overrepresent even the current capabilities of AI and ignore the physical limitations of the computer systems that AI models run on. We are far more likely in an AI bubble that will burst as a ceiling of processing capability is reached in the next 1-2 years.
I made a related market for individual 2026 predictions. Since they claim that uncertainty increases beyond that I think its a good indicator of how everything is going to turn out. Help adding questions wanted!
https://manifold.markets/BayesianTom/which-ai-2027-predictions-will-be-r
I’m conflicted between:
1. This sounds extremely outlandish and science fiction when you read all the conclusions (post-scarcity society, solving medicine/nuclear fusion/gene editing, datacenter of geniuses
2. It’s done by some of the smartest people I’ve read who make truth their number one goal, and have thought for a long time about what’s the most true possible reality
@MingCat Eli AND Daniel (plus Scott thrown in for free) is definitely making me take this super seriously. Which is awkward because I have a lot of bets implying AGI by ~2027/2028 is significantly less than 30% likely. XD
😆 great comment.
@MingCat Do note that Eli's median timelines are a bit longer than the scenario: see the first graph in https://ai-2027.com/research/timelines-forecast. (and I agree with him fwiw).
@MingCat note that January 2027 is still early/normal, lots of the contingent things happen later in the scenario.