AI economics & governance

Research briefs for independent evaluation. Research scopeWe cover economics, social science, policy, and forecasting, in line with The Unjournal’s focus. The emphasis is on AI’s trajectory and social consequences rather than work centered on machine learning, computer science, and engineering, such as technical alignment or frontier-model benchmarking. Quantitative models and methods remain in scope..

Read paper discussions A simpler list with full discussions, evidence, and feedback for prioritized papers across research areas.
Overview, audio, and help
ListenAn audio walk-through of the bottom line, the live policy debates and every paper (about 16 minutes, synthesised voice). Working notes, not Unjournal decisions. Read as an overview · Plain transcript · Download MP3

Unjournal team members: internal notes on these papers (possible conflicts, team discussion, outreach) are on a Notion page. Open it if you have access, or ask contact@unjournal.org.

Your rater profile and saved ratings
Optional for public ratings; name and email are required for team ratings and saved privately in this browser. Use your team email for matching, or use a pseudonymFor extra confidentiality, use a pseudonym and a separate email address, then email contact@unjournal.org privately to tell us which entries are yours. Leave public attribution off if you want to remain anonymous. The team can match your entries manually; this does not authenticate team membership. and contact us privately.

Focus on your strongest 5–10 candidates. Why?Spend most of your time on papers with the strongest potential impact: we’re likely to commission evaluations of only a few. Distinguishing candidates around the 80th–95th percentiles matters more than fine distinctions around the 50th–70th percentiles, and much more than sorting papers below the median. Use a quick first pass to find your leading candidates; there’s no need to examine the whole list in detail.

Open a title for its brief. Use Discussion for the full paper page.
Or choose a rating in the right-hand column.

Your rating is saved in this browser. Human totalsEveryone’s published submissions, including yours after the next refresh. Your saved rating remains visible in every display mode. Earlier Keep/Maybe/Drop picks count on the same scale; a detailed rating replaces your earlier pick. combine published ratings; AI scoresModel screening scores (0–100), with their own provenance and uncertainty. They are distinct from human priority percentiles and research-quality judgments. Choose human ratings, AI scores, both, or just your saved ratings above. The human/AI average combines the published weighted human mean with the stored AI screening score. Quick ratings have weight 1; full ratings have weight 2, or 3 for public raters and 4 for team raters with detailed discussion (at least 120 characters or 20 words). The AI score has weight 3. When one source is missing, the available source is used. This is an experimental evaluation-priority aid. provide a separate screening view.

These ratings are early and provisional, often from one or two people, and will change as more input comes in. A low priority for evaluation is not a judgment of a paper’s quality: some excellent papers are poor candidates because they’ve already had heavy scrutiny or have little to check.

Candidate papers. Open a title to read its brief. Compare and rate shows attention, recency, and published ratings.
?Sort order: selected for evaluation, working shortlist, getting attention now, alternates, other candidates, then older papers. Topic and source groups share the other-candidate level; their labels alone do not establish proximity to evaluation. Selected papers retain any outstanding objection condition. Click the heading again to reverse. ? ?Paper dates use the latest release or revision recorded in each brief, not when we added it. Month-only dates are approximate; the three-month filter includes months that overlap its window. In window: released or substantially revised since April 2026. Theory: older, but exempt because the contribution is theoretical. Borderline: just outside. Older: fails the recency rule but is still in use. Read ?Everyone’s published ratings, including yours if it has reached the latest refresh. AI ratings are available through the Ratings shown control. Quick ratings use -- 10 · - 30 · ~ 50 · + 70 · ++ 90; exact percentiles are shown at the nearest step, with the plain average. Earlier Keep/Maybe/Drop picks count as + / ~ / -. Try to form your own view before looking. This browser ?
Live policy debates these papers feed