Identify which timing-question framing is more accurate

Ascertain whether the fixed‑years framing or the fixed‑probabilities framing yields more accurate forecasts for timing questions about AI milestones in the Expert Survey on Progress in AI, and quantify any systematic bias introduced by each framing.

Background

The survey used two framings for eliciting timelines: (1) fixed‑years (asking probabilities by specified future years) and (2) fixed‑probabilities (asking years to reach specified probabilities). Prior and current surveys observe that fixed‑years consistently produces later predictions.

Despite replicating this effect in 2023, the authors explicitly state that they do not know which framing, if either, is more accurate, motivating the need to determine the relative accuracy and potential bias of each approach.