Near‑future pace of AI capability progress

Ascertain whether AI capability progress over the next decade will slow down, maintain its current pace, or accelerate, by rigorously evaluating scaling returns and data constraints, the prospects for algorithmic and architectural breakthroughs, and the relevant economic and political drivers, in order to calibrate the timing and urgency of AI welfare assessments and safeguards.

Background

The authors’ analysis of near-term AI welfare depends partly on how quickly AI capabilities advance. They outline reasons progress could slow or stall (e.g., diminishing scaling returns, data bottlenecks, cost constraints, political resistance) and reasons it could sustain or accelerate (e.g., continued scaling gains, breakthroughs, AI-for-AI research feedback loops, economic incentives, international dynamics).

Because the path and pace of progress materially affect when welfare-relevant capacities might emerge, establishing the trajectory of AI progress is necessary for determining when and how precautionary policies should be implemented.