Scenarios for our Economic Future

gist

Anthropic Institute’s Econ Scenario Explorer and companion working paper map AI capability and adoption paths into 2030 GDP, wages, labor share, reallocation, and unemployment. Jobs are bundles of tasks that AI can augment, automate, or leave alone while also creating new ones. Across modest, substantial, and extreme scenarios, growth always rises; only extreme change—near-full autonomous knowledge work, likely with recursive self-improvement—pushes unemployment beyond historical norms and cuts knowledge-worker wages while capital takes a larger share. A survey of about 11,000 US adults lands near the substantial path.

ideas

• Tasks, not jobs, are the unit. AI augments, automates, leaves alone, or creates tasks inside occupations (O*NET-style bundles); aggregate those millions of task instances and you get GDP, the labor market, and the labor share.
• Three calibrated futures. Modest ≈ internet-scale (+1.6% GDP by 2030, ~$34.1T); substantial ≈ twice-normal growth with half of knowledge work AI-capable but incompletely adopted (+8.3%, ~$36.3T); extreme ≈ RSI-speed autonomy (+32.4%, ~$44.4T, ~15% annual GDP growth).
• Displacement is an occupational switch cost. Knowledge work shrinks while manual/less-exposed work expands; switching is slow, so extreme automation spikes unemployment among cognitive workers even as other occupations tighten.
• Wages and capital diverge. Average wages can rise while knowledge-worker pay stays flat (substantial) or falls >10% (extreme); more of GDP accrues to capital than to labor income (labor share 60→45% in the paper’s extreme path).
• The model is a thinking tool, not a forecast. It omits policy, demand shocks, robotics, and financial crises; reviewers pushed capital returns and occupation-wage divergence into the scenarios, and many other criticisms remain open for later versions.

quotes