Modeling Progress in AI

AI progress claims need rigorous modeling: The paper identifies that discussions about AI's future impact—from intelligence explosions to job displacement—often rely on vague theories rather than systematic, quantifiable models. This makes it difficult to evaluate whether predictions are actually grounded in evidence.

Hardware and algorithms both matter, but differently: The paper emphasizes that AI progress depends on both faster computers and smarter algorithms, and these factors interact in complex ways. Understanding this relationship is crucial for predicting how quickly AI capabilities will advance.

Human expertise remains essential: The research highlights that human inputs—such as researcher knowledge and effort—play a significant role in enabling AI capabilities. This suggests AI progress isn't purely automatic and depends on continued human investment and guidance.

Different AI fields progress at different rates: Rather than treating "AI progress" as a single phenomenon, the paper argues that different sub-fields of AI (like computer vision versus natural language processing) advance at different speeds and through different mechanisms, requiring tailored analysis.

Better models could improve policy discussions: By developing more rigorous frameworks for modeling AI progress, researchers can provide more reliable insights on consequential issues like technological unemployment, helping policymakers make better-informed decisions about AI's societal impacts.