Alex Lawsen on forecasting AI progress

Abstract

This podcast episode features an interview with Alex Lawsen, an advisor at 80,000 Hours, about his approach to forecasting AI progress. Lawsen, who previously created a YouTube series about forecasting, discusses the importance of calibration training, the limitations of extrapolating past trends to predict future outcomes, and how to develop an “inside view” of AI alignment. He argues that, despite recent advances in AI capabilities, it is still difficult to accurately forecast the timing of general artificial intelligence (AGI) and that current methods of forecasting, such as extrapolating economic growth curves, are insufficient to make confident predictions. Lawsen also advocates for a more nuanced approach to forecasting, emphasizing the importance of considering different possible scenarios and carefully defining the criteria for resolving forecasts. He concludes by offering advice on how to identify opportunities for impactful work in the field of AI, suggesting that podcasting may be an effective way to reach a wider audience and contribute to the ongoing discourse about AI safety. – AI-generated abstract