Reproduce “A Model of Artificial Jagged Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Joshua Gans
Joshua S. Gans developed an economic model of Artificial Jagged Intelligence (AJI) where AI's uneven performance across tasks is treated as an information problem, and users cannot initially observe local reliability. The model demonstrates that users are statistically overexposed to AI weaknesses due to an 'inspection paradox', highlighting the economic value of calibration tools and specific investments beyond mere scaling.
Key insights
- The 'inspection paradox' causes users to be statistically overexposed to an AI model's weaknesses; the mean experienced error is double what a naive average would suggest.
- Investing in 'calibration' (tools that reveal local reliability and allow abstention) is a powerful economic complement to raw AI model improvements, enabling users to selectively delegate tasks and capture more value.
- AI 'jaggedness' is scale-invariant; increasing model size (more knowledge points) improves average performance but does not eliminate the relative unevenness, explaining persistent 'surprising' failures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.