Are we "trending toward" transformative AI? (How would we know?)

Abstract

This article investigates the forecasting of transformative AI, defined as AI capable of bringing about a qualitatively different future. The author argues that forecasting transformative AI differs from traditional forecasting methods due to the extended time horizons involved, the lack of readily available data, and the qualitatively unfamiliar nature of the anticipated future. The author acknowledges that extrapolating trends in AI capabilities may not be reliable due to the unpredictable nature of AI progress and the subjective interpretations of “impressive” or “capable” AI systems. The author also discusses the limitations of surveying AI experts for forecasting, citing potential biases and inconsistencies in their responses. The author concludes by emphasizing the importance of considering underlying factors and trends, such as the growth of AI models, to predict the potential arrival of transformative AI. – AI-generated abstract.