Marginal Revolution: Are we witnessing an intelligence explosion?

Marginal Revolution: Are we witnessing an intelligence explosion?

Luis Garicano and Alex Imas argue that AI capabilities improved 1.91 times faster after October 2025, while R&D spending grew 6.19 times and the technical workforce 3.10 times. Their accounting comparison estimates that measured capability gains per dollar fell by approximately 69%, and measured gains per technical employee-year decreased by 38%. They stress that these figures do not establish causation: falling returns could reflect congestion as organizations expand effort, or advances becoming harder to find. With enough congestion, the same figures could even fit a pattern in which each advance makes the next easier—the possibility required for an intelligence explosion. The authors conclude that their finding is consistent with ideas getting harder to find in AI, requiring ever larger increases in inputs to maintain the rate of improvement. They say the idea of an intelligence explosion remains to be demonstrated.

Publication: 64086dfb-f56e-44dc-8943-42b7540607b1

attribution — pass: The Marginal Revolution post presents the passage as quoted from Luis Garicano and Alex Imas. The reading attributes the figures, the caveats and the conclusion to them, and the title credits the publisher. It does not present their claims as established fact.

From Luis Garicano and Alex Imas:

summary — pass: The summary leads with the authors' accounting comparison and keeps their caveat that the figures show no causation. It includes both explanations they offer, congestion and progressively harder advances. It also keeps their notable point that strong congestion could still fit an explosion. In plain terms, it explains congestion and what an intelligence explosion requires, then ends with their conclusion. The question comes directly from the source's stated open issue. Omitting the Bloom reference does not distort the meaning.

To know whether a break is needed, we must know what an unchanged relationship would have delivered

date — pass: The originalPublished date of 2026-11-13T18:39 UTC comes from the publisher page and matches the feed entry for this Marginal Revolution post. The reading adds no conflicting date claim. The 'after October 2025' reference is quoted from the source.

support — pass: Each claim is backed by an exact, matching excerpt. The phrase 'grew more rapidly' accurately describes 6.19x and 3.10x against 1.91x. The limitations correctly note the authors' disclaimer about causation and the missing underlying data and method. The reading adds no unsupported benefits, costs or forecasts.

So measured capability gains per dollar fell by approximately 69%, and measured gains per technical employee-year decreased by 38%.

media — pass: The selected media is the verified direct link to the original Marginal Revolution article. It is the most direct way to reach the source.