Through a glass darkly

Abstract

Three methods of forecasting transformative AI (TAI)––expert surveys, prediction markets, and biological anchoring––are evaluated for accuracy and consistency. Expert surveys, while seemingly accurate in 2016, proved unreliable in a 2022 replication, often misjudging already achieved milestones. Prediction markets, exemplified by Metaculus, demonstrate some predictive power but exhibit volatility and potential overreaction to specific events. The biological anchoring model, based on comparing the computational power of AI with the human brain, offers a more structured approach but still relies on uncertain parameter estimations, leading to revisions over time. While all three methods suggest TAI is likely within 10-40 years, the limitations of each highlight the inherent difficulty of predicting such complex technological advancements. – AI-generated abstract.