Report on semi-informative priors

Abstract

Many STEM fields have the explicit goal of developing highly ambitious technologies. One such goal is artificial general intelligence (AGI), a hypothetical type of AI theorized to possess intellectual abilities indistinguishable from those of a human. Researchers employ a variety of methods to estimate the timeline for AGI development, including the application of uninformative priors and inside view approaches that consider the specific challenges that need to be solved. This report instead proposes a semi-informative prior that combines elements of both approaches. Using this framework, the authors estimate that there is a 5-35% chance AGI will be developed by the year 2100, with a central estimate of 20%. The authors also find that more rapid growth in the AI research community and of computation used in AI R&D boosts the probability of earlier AGI development. – AI-generated abstract.