Artificial intelligence and economic growth
An evaluator found an error in a proof supporting a central result of Aghion, Jones, and Jones. The authors confirmed a corrected proof; the public record shows what was wrong and what still holds.
Why this, and why now
How fast will AI capabilities grow and replace human work? How will gains flow to labor versus capital? Will revenue and compute concentrate in a few firms and countries? Will regulation change which models are released where? These questions bear on AI-safety strategy, philanthropic funding, regulation, public spending, and international controls.
Economists, forecasters, and policy analysts are moving into this area fast. Growth theorists are co-authoring scenario models with frontier labs. Forecasting researchers are eliciting beliefs from economists, AI-company staff, and superforecasters. Policy organizations are producing quantitative estimates of compute smuggling and regulatory delays. Some of this work is already being used: in fiscal scenarios, in commentary on monetary policy, and in legislative fights.
Little of it has had careful, public, expert scrutiny. Academic journal review would provide that, but it takes years and reveals only which journal accepted the paper. By then much of this work is out of date. Informal critique in blogs and threads is faster, but it is self-selected and rarely carries legible expertise. We aim to fill the gap: neutral, paid, expert evaluation, published quickly, with ratings and claim-level assessment.
How it works
We are neither a journal nor a forum thread. Evaluators write a referee-style report with quantified ratings and uncertainty, and we publish it alongside the authors' response and a management synthesis.
1. Prioritize
We screen for decision relevance, apparent influence, checkable claims, and timing, with input from an advisory group and public ratings.
2. Commission
We match each paper with balanced expertise from our pool of 200+ vetted evaluators and beyond, check conflicts, and pay for careful critique with incentives for promptness.
3. Publish
Evaluations, ratings, and author responses appear publicly with DOIs, usually within about a month of commissioning for this stream.
4. Synthesize
We summarize which claims hold up, which assumptions drive the results, and what a decision-maker should and should not rely on.
We ask whether research is credible and useful, not whether it is novel or "publishable." We have no stake in which way a result comes out: we do not produce AI research, hold a policy agenda, or take funding from AI labs.
Where we are
We have briefs on 34 candidate papers, generated with AI and each given some human review: what each claims, why it matters, how checkable it is, and how much scrutiny it has already had. Two are prioritized for evaluation now; more follow as evaluators and funding allow.
Focus area: the economic effects of AI, covering forecasts, labor-market evidence, effects across countries, and who captures the gains, with our related past evaluations.
- Forecasting the Economic Effects of AI Karger et al., Forecasting Research Institute. Structured forecasts from economists, AI-company staff, superforecasters, and the public on growth, labor-force participation, and more. Already used in fiscal scenarios and policy commentary. Evaluation focus: sampling, question design, aggregation, and economic interpretation.
- Economic Scenarios for Transformative AI Korinek et al., Anthropic Institute. Maps AI-capability paths onto GDP, labor share, wages, and unemployment to 2030. A likely anchor for the redistribution debate. Evaluation focus: model structure, sensitivity to assumptions, calibration, and scenario selection.
- Potentially next Candidates under consideration include the economics of recursive self-improvement; estimates of compute smuggling to China; game-theoretic models of AI race dynamics; AI liability and insurance.
Scope: AI governance and regulation, the rate of AI progress, and AI's effects on growth, labor, and the concentration of economic and political power. Not technical AI safety or interpretability. The briefs are AI-generated, checked against sources, and reviewed by at least one person on our team; corrections are welcome.
Unjournal track record
114public expert evaluations published, with over 1,000 structured ratings
59research papers and projects evaluated, mostly in economics and quantitative social science
200+vetted evaluators in our pool, and we recruit beyond it for each paper
The returns to science in the presence of technological risks
Three evaluations of Matt Clancy's model of whether accelerating science is worth it given catastrophic risks, with a detailed author response on discount rates and the treatment of AI risk.
Forecasting existential risks: the XPT
Evaluations of the tournament design, elicitation, and aggregation behind widely cited forecasts of catastrophic risk from AI and other sources.
Towards best practices in AGI safety and governance
Evaluations of a survey of AGI-lab practices that has informed governance proposals, assessing its methods and the strength of its claims.
Our team and network lean toward formal economics and social science. Governance, law, and compute expertise is thinner, which is one reason we are asking for evaluator suggestions.
How to help
- Tell us which claims you rely on. If you or your organization use this research in decisions, funding, or agenda-setting, we want to know which claims matter and what would change if they were wrong.
- Suggest evaluators. We need people in expert elicitation and survey methods, macroeconomics and growth theory applied to AI, labor economics, and the technical side of AI. Send suggestions or self-nominations privately to contact@unjournal.org, or join the evaluator pool.
- Rate and nominate papers. Rate and discuss the shortlist, or submit research we have missed. A low priority score is not a judgment of quality.
- Fund or partner. Evaluator compensation and speed depend on funding. If you want independent evaluation of research your organization produces or relies on, write to contact@unjournal.org.