Reviewing The Evidence On Worker Retraining Programs
A new review coauthored by David Roodman and Anthropic's Maxim Massenkoff assesses worker retraining programs, concluding they offer modest benefits but are likely insufficient for large-scale AI-driven job displacement.
Intelligence analysis by Gemini 2.5 Flash
Anthropic's economic research team has published a review of 56 US and European studies on worker retraining programs, finding that while these initiatives yield positive but modest effects, they are unlikely to effectively address significant labor market disruption caused by AI. The authors recommend immediate investment in demonstrating, evaluating, and scaling the most promising p…
Imagine if robots started doing lots of jobs, and people needed to learn new skills. This report looked at how well special classes, called retraining programs, help people get new jobs. It found that while these classes help a little bit, like giving someone a small boost, they probably aren't strong enough to help everyone if many jobs disappear all at once. So, we need to find better ways to help people learn new things quickly.
Analysis
David Roodman
Independent researcher David Roodman, alongside Anthropic's Maxim Massenkoff, coauthored the comprehensive review examining the effectiveness of worker retraining programs. Their collaboration is part of Anthropic's broader Economic Research team's initiative to understand AI's effects on the economy, building on previous work like the Economic Index and a framework for measuring AI's labor market impact.
The review's findings are central to Anthropic's Economic Policy Framework, which explores various policy responses to AI-driven disruption. By scrutinizing the evidence behind worker retraining, Roodman and Massenkoff aim to provide a data-driven foundation for future policy decisions, ensuring that proposed solutions are grounded in empirical reality rather than assumption.
56 Randomized US Studies
The review synthesizes evidence from 56 randomized US studies, augmented by experimental data from Europe, to provide a robust meta-analysis of retraining program outcomes. This extensive dataset allows for a comprehensive assessment of average program effectiveness, cost-efficiency, and the government's return on investment.
On average, the studies indicate that job training programs lead to a modest increase in employment by two to three percentage points and an annual earnings boost of approximately $1,000 per participant. While these gains are positive, they come at an average cost of about $13,000 per person, with governments recovering more than half through increased tax revenue and reduced benefit payments, leading to programs roughly breaking even overall.
Economic Futures Research Fund
The authors' central recommendation is to invest significantly in demonstrating, evaluating, and scaling the most promising retraining programs, including through efforts to rapidly expand leading programs for specific worker groups and rigorously measure their results. This forward-looking approach is crucial given the anticipated scale of AI-induced labor market changes.
Anthropic's Economic Futures Research Fund is specifically designed to support investigations into such critical questions, aiming to foster research that can inform effective policy responses to AI's economic impacts. The fund seeks to identify and support innovative solutions that can bridge the gap between current retraining capabilities and the future demands of an AI-transformed economy, ensuring that policy development is proactive and evidence-based.
Key points
- Worker retraining is a popular policy option for mitigating AI-driven labor market disruption.
- A review of 56 US and European studies found that current job training programs produce positive but modest effects.
- On average, programs increase employment by 2-3 percentage points and earnings by roughly $1,000 annually, costing about $13,000 per person.
- A small set of 'sector programs' that partner with employers show larger gains, but attempts to replicate them have often failed.
- The authors conclude that existing retraining programs would likely be insufficient if AI displaces workers at scale.
- The central recommendation is to invest now in demonstrating, evaluating, and scaling the most promising programs.
The research identifies 'sector programs' as potentially more effective, suggesting that focused investment and rigorous evaluation of these models could lead to scalable solutions. By acting now to demonstrate and expand successful programs, societies could proactively mitigate the negative labor market impacts of AI.
The report warns that existing retraining programs are likely to fall short if AI displaces workers at scale, potentially leading to widespread unemployment and economic instability. The difficulty in replicating successful 'sector programs' also poses a significant challenge, suggesting that effective solutions may be harder to implement broadly than hoped.
