Forecasting AI Progress — Long-horizons, Out-of-Sample Questions, and the Dreyfus Prize

Forecasting AI Progress is a large-scale, comprehensive forecasting tournament dedicated to predicting advances in artificial intelligence. It's an excellent opportunity to make sense of one of this century’s most important technologies, and for you to continue developing their forecasting abilities in a tournament with real-world stakes and prize money on the line.

Out-of-sample Questions

Each round of the Forecasting AI Progress features questions that are out-of-sample—by which we mean that these questions resolve after the tournament is judged and the prizes disbursed. You’ll be able to readily identify these questions, as these usually span time-frames in excess of two years.

Your performance on these questions will not affect the score in the tournament, nor your tournament ranking. However, despite this, these questions might be the most important to answer to contribute to an understanding of the future of AI. Moreover, these predictions will feature in the analyses produced for the Dreyfus Prize (prizes awarded to five authors of best analyses of the long-term future of Artificial Intelligence).

Like we do for all forecasts, we will still be tracking everyone’s performance on these questions even though these take more than three years to resolve. This way, you’re able to see how well you can forecast at long time-scales, and we can train our systems on long-horizon forecasting data.

Answering All Questions

To qualify for prizes, you will need to answer all the questions in the tournament, including the out-of-sample questions. This is important to us because we ultimately want to produce a dataset of forecasts spanning many time-frames, so that we can evaluate how predictive accuracy varies over time-horizons.

Dreyfus prize [Coming Soon]

The Dreyfus prize is awarded to five authors of best analyses of the long-term future of Artificial Intelligence. A total of $6,500 will be awarded to the top five authors. Each participant is limited to a single entry.

Analyses might take the form of well-reasoned arguments and/or appropriate models for predictions on questions related to the long-term future of AI that are run in the tournament. Other analyses we would like to see are those that synthesize the predictions that are made in the course of the tournament, that draw out their key implications, and assemble these into a coherent view of the future of AI. Judges will look for analyses that are insightful, well-reasoned, and elucidating of some aspects of the future of AI. Authors should demonstrate a keen understanding of both forecasting and AI.