Forethought researcher Tom Davidson discusses a research-taste evaluationEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. and cautions that its measured tasks do not establish that AI can perform every kind of frontier research. He considers how automating more routine research workWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort. could free researchers and accelerate further automation.
Davidson distinguishes recursive improvementRecursive self-improvement is the proposed process in which an AI system helps improve its own capabilities, then uses those improvements to support further advances. from an explosive takeoff, considering expensive experiments, diminishing returns and hardware production as potential constraints. He suggests that labs could study internal algorithmic progress and the relationship between research effort and compute.
The conversation also considers research diversity, jobs, political responses and concentration of power. Davidson presents possible futures and uncertainties rather than a settled forecast.
Watch on YouTube
