Economics team of AI research company Anthropic has published a new working paper, “Economic Scenarios for Transformative AI“, alongside an interactive scenario explorer that translates expectations about AI capability and adoption into projected paths for GDP, wages, employment, and the labor share through 2030.
The framework, built on a task-based macroeconomic model, does not offer predictions. Instead, it converts a small set of measurable parameters, the share of tasks AI can perform, how widely it is used, productivity gains per task, and the balance between automation and augmentation, into comparable economic outcomes.
The paper illustrates three scenarios. In the modest scenario, AI’s economic footprint resembles the internet’s: GDP by 2030 is 1.6% above its no-AI path, growth reaches 2.4% annually, and unemployment rises by only a tenth of a point. The substantial scenario corresponds to forecasts circulated by financial institutions in 2023: AI performs roughly half of knowledge work by 2030 with 8.3% higher GDP, growth of 5.4%, and knowledge-worker wages essentially flat while wages elsewhere rise. The extreme scenario assumes recursively self-improving AI adopted rapidly: GDP growth hits 15.4% per year, the economy is 32% larger, and society is far wealthier, but the labor share falls from 60% to 45%, cognitive wages drop 11.5% below trend, cognitive unemployment reaches 17.9%, and overall unemployment hits 11.9%, beyond postwar records.
Notably, the model finds AI’s boost to innovation-driven growth is small even in the extreme case, since research remains bottlenecked by physical tasks.
Complementing the model, Anthropic surveyed 10,980 US adults in August on AI capabilities, adoption, productivity effects, and re-employment prospects. The median respondent expects AI to handle six of eight benchmark tasks, from routine business correspondence to building software, by 2030, expects deployment on 40% of feasible tasks, and believes a displaced worker would need roughly eight months to find work in a new occupation.
Fed through the model, these median answers produce outcomes close to the substantial scenario: GDP 8.6% above the no-AI path and unemployment around 4.6%. Views are highly dispersed, however; about 10% of respondents hold expectations consistent with the extreme scenario, while 40% say Nobel-level AI-driven discoveries will never occur.
The authors emphasize that almost all divergence between scenarios materializes after 2027, since the paths share current measurements of AI use. Sensitivity analysis underscores two swing factors: the elasticity of capital supply, which determines whether workers or capital owners capture the gains, and wage rigidity, which determines whether cognitive workers bear costs through lower wages or joblessness.
In the extreme scenario, total labor income is roughly unchanged despite GDP being a third larger, meaning nearly all gains accrue to capital and compensating knowledge workers would require transfers of about 9% of GDP, roughly the scale of Social Security and Medicare combined, with no historical precedent for technology-driven redistribution of that magnitude.
According to the company, the framework is intended to inform Anthropic’s research funding and policy proposals aimed at ensuring AI’s economic benefits are broadly shared.
The post Anthropic’s Economic Scenarios For 2030: AI Could Double Growth Or Push Unemployment Past Postwar Records appeared first on Metaverse Post.
