Predicting AI takeoff speed using Manifold consensus
Other prediction markets have taken a less scientific approach of predicting when AGI will be achieved, because they use arbitrary resolution criteria chosen by a single person rather than a community consensus.
Every month, a poll is posted asking whether Manifold users believe that weak AGI has been achieved. The poll does not define "artificial general intelligence," and only states that a "weak" system does not need to physically manipulate the world. At the time of market creation, one third of respondents believed that weak AGI existed, with the trendline projecting consensus that AGI exists in February 2025.
Beginning in September 2024, a second question, worded nearly identically to the first, will be posted, asking whether weak superintelligence has been achieved. The question will not define superintelligence, except again to state that the system need not be able to manipulate the physical world.
On the first month that the poll about "artificial general intelligence" has more YES respondents than NO respondents, a clock will begin ticking. If seven or more months elapse before the poll about "superintelligence" has more YES respondents than NO respondents, this question will resolve to NO. Otherwise, this question will resolve to YES.
This market will also resolve to YES if the poll involving superintelligence achieves a majority first in a contradiction.
RELATED: