Superforecasting appears to be an innate talent, but one that can be significantly improved through training and by organising the best performers into elite teams. The biggest gains appear to be from reduction in noise (filtering out erroneous or irrelevant information) with significant contributions from improved information-gathering and reduction in bias (incorrectly prioritised information).2
Research is ongoing into how effectively superforecasting can be taught,3 how successfully it can be applied to domains other than geopolitics, and whether predictions can be further improved through integration with expert knowledge. The ability of superforecasters to assess extreme, disruptive risks — “black swans” such as pandemics, for example — is also an area of active debate.4
Prediction markets also harness the human ability to synthesise many types of information, but do so by aggregating the predictions made by many individuals (some of whom may be superforecasters).5 Participants are offered financial or reputational incentives to make accurate predictions, and market mechanisms are used to establish consensus positions. On political events,6 technological developments7 and corporate strategy.89
The optimal design of such markets is an area of active investigation.10 One emerging model is “tournaments”, in which participants first make individual forecasts before being asked to collaborate in teams, then to assess other teams’ rationales and update their own predictions. One such tournament, focusing on existential risks, found that the predictions made by expert and superforecaster participants continued to be significantly different.11
Because the underlying mechanisms of superforecasting remain somewhat obscure, more research and validation are needed to legitimise its use in real-world decision-making. The same is true of the application of artificial intelligence to detect patterns which can be used for prediction, whether independent of human prediction or integrated into it.
