Jack Clark Urges IMF to Abandon Standard Growth Models Over Automated Science

Anthropic co-founder Jack Clark has warned the International Monetary Fund that prevailing macroeconomic models cannot account for the economic disruption caused by automated scientific discovery.

Speaking to economists and policymakers at the Washington-based lender, Clark argued that classical growth frameworks are obsolete. Existing economic tools assume that human labour remains the fundamental bottleneck to technological progress. That premise collapses, Clark said, as artificial intelligence transitions from narrow task execution to autonomous research and development.

Multilateral institutions routinely analyse AI through the lens of labour-market displacement. Central banks and finance ministries model productivity gains as incremental shifts in labour efficiency, assessing how models substitute for office workers, paralegals, or software engineers. Clark told the IMF that this lens mischaracterises the core trajectory of the technology.

The decisive macroeconomic shock will not stem from automating routine services. It will stem from automating the scientific method itself. When frontier models design experiments, analyse complex biological systems, synthesise materials science literature, and formulate novel hypotheses, the production function for technological progress alters permanently.

Orthodox endogenous growth models, such as those pioneered by Paul Romer, treat ideas as non-rival goods produced by human researchers. In these frameworks, the rate of economic expansion is constrained by the supply of human minds devoted to research. Technological progress scales roughly with the population of scientists and engineers, mediated by institutional funding and educational pipelines.

Clark pointed out that frontier AI decouples the generation of ideas from human biological constraints. When research output is driven by compute clusters rather than human headcounts, the inputs to innovation shift to electricity, silicon availability, and algorithmic efficiency. The feedback loop between software advances and scientific discovery threatens to produce non-linear growth dynamics that standard dynamic stochastic general equilibrium models cannot absorb.

Clark framed this transition as the economic realisation of the singularity. Stripping the term of speculative science fiction, he defined it as an empirical economic inflection point. In this regime, software-driven research reduces the time required to achieve scientific milestones from decades to months. Feedback loops compound when AI models assist in developing better algorithms, specialised hardware, and more efficient energy systems for subsequent model generations.

The IMF relies on long-range forecasts to evaluate debt sustainability, structural adjustment programmes, and balance-of-payments risks across sovereign borrowers. Clark argued that baseline projections grounded in historical capital-labour ratios risk steering global economic policy blind. If scientific discovery accelerates exponentially, current assumptions regarding capital depreciation, intellectual property value, and national competitiveness will degrade rapidly.

To manage this transition, Clark urged multilateral institutions to overhaul their analytical infrastructure. The fund must stop relying solely on traditional metrics such as total factor productivity and labour force participation. Instead, surveillance frameworks should track compute capacity, energy infrastructure, and the velocity of AI-driven research outputs.

Under this formulation, sovereign compute capacity emerges as a primary determinant of macroeconomic stability. Nations lacking access to advanced semiconductor clusters or the power grids required to run them will face acute technological divergence. Sovereign risk assessments may soon need to incorporate domestic AI infrastructure alongside foreign exchange reserves and fiscal deficits.

Clark also addressed the widening gap between technological development and institutional response times. Frontier labs are deploying models that compress scientific workflows at a pace that exceeds the legislative and regulatory cycles of democratic states. The lag leaves international financial institutions relying on linear analytical methods to manage exponential underlying shifts.

For international policymakers, the dilemma is no longer whether AI can substitute for individual jobs. The broader challenge is restructuring macroeconomic governance around an economy where cognitive labour is abundant, scalable, and machine-driven. If the rate of scientific invention is freed from human biological limits, can global financial institutions rebuild their analytical foundations before their existing baselines become entirely irrelevant?