Welcome to The Unjournal drop-in
Tonight's theme: scaling up our AI economics and governance evaluations.
| Time (UK) | Item |
|---|---|
| 19:30 | Welcome and introductions |
| 19:34 | News: the AI economics and governance stream, and choosing papers 3 and 4 |
| 19:41 | Progress on other projects |
| 19:46 | Where you can help, including a research lead role |
| 19:54 | Open discussion |
| 20:07 | Coming up, newsletter, feedback |
| 20:10 | Management team check-in (management team only) |
Recorded with an AI notetaker and shared with the team; ask any time to pause or keep something out. Questions go in the Zoom chat, and we'll answer all of them.
Good news: the AI economics and governance stream is funded
Lightcone Commons has awarded us a grant (announced 1 Nov) for a 12-month stream of evaluations of research on AI's economic effects and governance. The case for it is in our EA Forum post (20 Oct, with David Manheim):
- Economists and policy analysts are producing influential work on AI growth, labour, chip smuggling and regulation, and it is already shaping fiscal scenarios, policy commentary and legislation.
- Little of it has had careful, public, expert scrutiny: seminar critiques are private, blog responses are self-selected, and journal review takes years.
- We aim for public credibility signals within about a month of commissioning, with quantified ratings, claim-level assessment and author responses.
- We have briefs on 34 candidate papers, and we want your input on which matter most. We can go beyond those, of course: suggestions welcome.
"We hope to provide more rigor without accepting the (far longer) timelines for traditional review."
Where the first two papers are
- Economic Scenarios for Transformative AI: authors notified; first evaluator on board; recruiting a second
- Forecasting the Economic Effects of AI: authors notified; recruiting evaluators
- Selected under our simplified rule: two strong management approvals, then a week for anyone to object
Be careful not to say who has agreed or declined to evaluate.
Choosing papers 3 and 4
Selection call tomorrow, Tuesday 3 Nov, 15:00–16:00 UK. Candidates under discussion include:
- Recursive self-improvement and AI-driven growth: The Economics of Recursive Self-Improvement
- AI race dynamics and existential risk: The AGI Race and Existential Risk
- AI and the labour market, for example Still Waters, Rapid Currents or Canaries in the Gold Mine
AI governance paper briefs · Our EA Forum post: which research most needs evaluation?
Views before tomorrow? Put them in the chat, or rate papers on the briefs page.
Progress on other projects
Cultivated meat cost (Pivotal Question)
- September review corrected the cost model's accounting; revised since in response to critiques and comments
- Late October: dry cell mass now the default cost basis, side-by-side chicken-meat-equivalent costs, a 25%-cell hybrid product comparison
- Evidence log updated with new regulatory approvals and productivity claims
- Proposed "modeling hack": teams and AI tools each model 2036 cost per kg from one shared parameter sheet (date not set)
Value of existential risk mitigation
- Interactive model replicating Rethink Priorities' 2023 report, with switches for the alternatives raised in our evaluation (off by default)
- Unofficial rebuild of Rethink Priorities' 2026 global catastrophic risk fund model; a Donor Compass page reproducing their six ways of combining worldviews
- The report is now on Zenodo, with the author's and Rethink Priorities' permission
- Exploratory and largely AI-built; not yet checked by a person or reviewed by Rethink Priorities
Model site (draft) · Fund model rebuild · Report (Zenodo) · Our evaluation
Institute for Replication webinar
Fri 13 Nov, 14:00–15:30 UK (9:00–10:30 am US Eastern, 7:30 pm India). "Open Science Futures in Social & Behavioral Sciences", hosted by Monk Prayogshala with the Institute for Replication and The Unjournal: open science, replication, and open and AI-assisted evaluation. A first step toward in-person evaluation and replication games in India.
Where you can help
Suggest evaluators
The pool for serious AI economics evaluation is small: macro and growth theory, labour economics, forecasting and expert elicitation. PhD students and postdocs welcome. Send names privately to contact@unjournal.org.
Advise an evaluation
Short advisory roles on the current papers: especially someone strong on expert elicitation and aggregation.
A research lead role
We're planning a part-time lead for the stream (next slide). Interested, or know someone?
Also: rate papers on the briefs page, and share the work.
Planned: part-time research lead, AI economics and governance
- Purpose: run our 12-month stream, jointly with David Reinstein and others on the team, evaluating research on the economics and governance of AI: labour markets, growth, concentration of wealth and power, regulation, state capacity.
- Main work: find and prioritize the most useful research; commission expert evaluators and oversee evaluations; turn open debates into Pivotal Questions, where better evidence could change decisions.
- Also: run workshops with researchers, practitioners and funders that map disagreements and record how views shift; write short, timely syntheses for research users; work with our field specialists and advisors.
- Time: about half-time for about 12 months, remote.
- Background: research experience in economics, quantitative social science or policy; familiarity with AI economics or governance; comfortable judging research quality and managing several people and deadlines; good communication skills.
Interested, or know someone? Email contact@unjournal.org or use our expression-of-interest form. Details are still being finalised; a full description will follow.
More detail
The stream plans about 15 evaluation packages and two Pivotal Questions, each with a workshop. Applied AI work goes out of date quickly, so the job is getting careful public evaluation out fast, well within journal timelines. The lead would work closely with the co-directors and our field specialist team.
Open discussion
Questions, suggestions, challenges: Zoom chat or just speak up.
Some starting questions
- What would make a strong researcher take on a fast AI economics evaluation? Purpose, recognition, prizes, pay?
- How should evaluators use AI? Our current policy is evolving: since we want fast, high-quality turnaround on AI-relevant work, we may encourage AI use more. Should we run strong AI evaluations first and ask humans to check them, or keep them separate and compare? Examples of AI review workspaces for our first two papers (AI-generated, not checked by a person, not Unjournal evaluations): Forecasting the Economic Effects of AI · Economic Scenarios for Transformative AI. We'd also consider trialling refine.ink.
- AI governance work goes stale in weeks. How fast do we need to be, and what formats should we evaluate (dashboards, datasets, living reviews)?
We'll respond to every comment, including those we don't reach live, and add them to the record.
Coming up
- Tue 3 Nov, 15:00 UK: AI economics paper selection call
- Fri 13 Nov, 14:00 UK: joint webinar with the Institute for Replication and Monk Prayogshala on open science futures (register)
- Wed 18 Nov, 12:30 UK (7:30 am US Eastern): next drop-in
- Mon 7 Dec, 19:30 UK (2:30 pm US Eastern): first-Monday drop-in
- Newsletter: unjournal-newsletter.netlify.app/newsletter
- Anonymous feedback: info.unjournal.org/feedback.html
- Talk to David: calendly.com/daaronr
Thanks! Management team: please stay on.