Forecasting transformative AI: the "biological anchors" method in a nutshell

Abstract

This essay summarizes Ajeya Cotra’s “Forecasting Transformative AI with Biological Anchors” (Bio Anchors) and discusses its merits and demerits in predicting the timeline of transformative AI. Bio Anchors posits that the cost of training an AI model scales with its size and the complexity of the tasks it learns. The author extrapolates from the costs of training existing AI models to predict the cost of training a model the size of a human brain to perform tasks such as scientific research. The author argues that this model places a high probability on transformative AI being developed this century, despite its inherent complexity. The author then discusses several arguments for and against this model’s accuracy. The author concludes that while Bio Anchors likely errs on the conservative side, the lack of an established expert consensus on the timeline of transformative AI means that further research is needed. – AI-generated abstract.