The Korinek–Stiglitz Collaboration

In the AI Story

The substitutability analysis is the analytical core. A productivity-enhancing technology can be deployed in two modes: labor-augmenting, in which each worker becomes more productive and retains her role at expanded capability; or labor-saving, in which the same output is produced with fewer workers. Claude Code is genuinely ambiguous between these modes — Segal's Trivandrum deployment is augmenting, while a competing firm's calculation that five engineers can replace fifty is saving. The technology does not determine the mode. The incentive structure does. And the current incentive structure, Korinek and Stiglitz demonstrate, strongly favors saving because saving produces margin expansion that capital markets reward, while augmenting produces capability investment that capital markets cannot see.

The unemployment projection follows. If AI's labor-saving mode dominates — as current institutional arrangements incentivize — and if displaced workers cannot be absorbed into new productive activity because the institutional infrastructure for absorption does not exist, the resulting unemployment rate substantially exceeds the levels experienced in the Great Depression. The economy would be simultaneously more productive and more unequal, generating more output with fewer workers while displaced workers bear the full cost of displacement. This is not a pessimistic forecast; it is the central-case projection of a formal model whose assumptions track current institutional conditions. Deviating from the projection requires deviating from the conditions.

The steering framework is the policy conclusion. Korinek and Stiglitz argue that when redistribution is politically constrained — as it is under the current inequality spiral — the next-best intervention is influencing the direction of technological progress itself. Tax incentives favoring labor-augmenting AI over labor-saving AI. Public research funding directed toward applications that expand rather than substitute for human capability. Regulatory preferences for AI deployment that creates rather than eliminates productive opportunities. These are interventions in innovation direction, economically justified when social safety nets are inadequate — which is to say, justified now.

The collaboration's influence extends beyond academic economics. Korinek's policy work with the IMF has introduced elements of the framework into multilateral governance discussions. Stiglitz's public advocacy has brought the analysis to policymakers, journalists, and civil society organizations engaged with AI governance. The framework is now a standard reference in serious discussions of AI's distributional consequences, though implementation of its policy recommendations remains politically obstructed by the same forces the framework identifies as obstacles.

Origin

Korinek, trained at Columbia where Stiglitz remained after his Nobel, began developing AI-economics analysis in the mid-2010s as part of broader work on financial stability and inequality. The collaboration produced major papers in 2017, 2019, and 2022, each extending the framework's scope. The 2019 NBER working paper Artificial Intelligence and Its Implications for Income Distribution and Unemployment is the canonical reference; the 2022 Steering Technological Progress paper articulated the policy framework most fully.

Key Ideas

Labor-saving vs labor-augmenting decomposition. The same technology produces radically different distributions depending on deployment mode; mode is determined by incentive structure, not technology.

Market bias toward labor-saving. Quarterly earnings metrics reward margin expansion over capability investment, systematically selecting labor-saving applications even when labor-augmenting would produce higher social welfare.

Fifteen-percent unemployment projection. Under current institutional arrangements, the central-case forecast substantially exceeds Great Depression levels.

Steering as second-best policy. When redistribution is politically constrained, shaping innovation direction through tax incentives, research funding, and regulatory preferences becomes the effective lever.

Formal rigor as political resource. The mathematical precision of the framework makes its distributional warnings harder to dismiss as mere ideology.

Debates & Critiques

The debate over the steering framework is substantial. Critics argue that governments lack the information to distinguish labor-augmenting from labor-saving applications, that such distinctions are often context-dependent, and that intervention in innovation direction risks producing worse outcomes than unconstrained markets. Korinek and Stiglitz respond that perfect information is not required — broad categories suffice — and that the current unconstrained market is already producing outcomes worse than any plausible policy error, as evidenced by the projected unemployment rates.