Apart Research's Economics of Transformative AI Sprint is a weekend research hackathon (co-organised with BlueDot Impact) that challenges participants to produce original economic analysis of how transformative AI will reshape labour markets, productivity, and inequality — with a hard requirement that outputs have direct relevance to AI safety governance. It exists because AI safety research has been dominated by computer scientists; Apart wants economists, domain experts, and empirical modellers to fill the governance and distributional-risk gap before advanced capabilities arrive. Strong projects are invited into a 12–24 week Apart Lab Fellowship for publication support, compute, and travel funding.

• THREE scored dimensions (inferred from published reviewer comments on the prior Africa LMIC submission): 1. Innovation & Literature Foundation — engagement with the empirical AI-economics literature (Acemoglu/Restrepo task-exposure framework, Otis et al. 2024, Björkegren, etc.); novelty of framing; citation quality (reviewers penalised only 3 citations heavily). 2. Practical Impact on AI Risk Reduction — explicitness of the causal chain from findings → AI safety / governance outcomes; does the paper explain HOW inequality in LMICs feeds back into global catastrophic risk, alignment funding, or model governance? (The past Africa submission scored lowest here — 1.5/5 — for skipping this linkage.) 3. Methodological Rigor & Scientific Quality — quantitative models with specified parameters; code repository; released data; actual empirical analysis completed (not a project plan). The 2nd-prize winner ran 4 million Claude prompts mapped to occupation data in Brazil — that is the bar. • Format compliance: PDF required (past submission penalised for DOCX); appendices must contain what they claim. • AI safety relevance: reviewers explicitly ask how findings 'feed back into global catastrophic risk, alignment funding, or governance of advanced models.' This is the key differentiator from a development-economics submission. • Openness: code repository + open dataset release are strong positive signals ('if delivered, would add value for future researchers'). • Eligibility: open to all analytical backgrounds; no formal credentials required; remote-first; LMIC perspective explicitly valued.

From the April 2025 Economics of TAI Sprint (8 entries total): 1st Prize — 'The Early Economic Impacts of Transformative AI: A Focus on Temporal Coherence' (formal theory on how goal-directed behaviour over time determines automation potential); 2nd Prize — 'Evaluating the risk of job displacement by transformative AI automation in developing countries' (empirical, used 4 million Claude prompts mapped to occupation data in Brazil's RAIS labour market — actual analysis, not a plan); 3rd Prize — 'Economics of AI Data Center Energy Infrastructure: Strategic Blueprint for 2030' (technical-economic bottleneck analysis); 4th Prize — 'Economic Feasibility of Universal High Income in an Age of Advanced Automation' (fiscal modelling of wealth taxes + AI dividends). The non-winning Africa LMIC submission (Kirumira/Macharia) received scores of 1 / 1.5 / 2 across the three rubric dimensions — its core failure was being a project plan with no empirical analysis, too few citations, and an indirect AI-safety linkage. The pattern: winners delivered completed quantitative work, not promises.

A quantitatively skilled researcher — economist, data scientist, or domain expert from an adjacent field — who can execute empirical or formal-model analysis in a compressed weekend, ground it in the Acemoglu-Restrepo task-exposure literature, release the code and data, and explicitly close the loop from their distributional findings to an AI safety or governance implication. Extra weight goes to LMIC/African perspectives that are grounded in real data rather than described as future deliverables. The platonic applicant is someone like the 2nd-prize winner: an empiricist who maps a specific labour market to AI automation risk using actual data, publishes the pipeline, and names the governance lever their findings inform.