DeepMind Institute Maps 11 AGI Economic Interventions Using 51 AI Agent Economists
The DeepMind Institute, launched yesterday, has published one of its inaugural essays: an evaluation of 11 household-facing policy interventions for managing AGI-driven economic disruption. The authors, Julian Jacobs and Alex Imas, frame the work as a conditional policy framework rather than an endorsement of any single universal program.
The core methodology is unusual. In addition to literature reviews and surveys, the essay uses 51 AI agent raters, each modeled on survey data from actual economists, to deliberate on policy tradeoffs. The AI agents were used to simulate how the economics profession might assess different interventions across a standardized set of dimensions.
The four-dimension rubric
The essay introduces a unified framework for comparing policy options across four dimensions:
- Welfare and Resilience: impact on material living standards, sense of meaning, and economic stability
- Agency and Voice: individual economic choice, direct ownership of AI-driven gains, and democratic participation
- Feasibility and Efficiency: political support, administrative simplicity, economic cost, and speed of rollout
- Durability: suitability across different AGI economic scenarios, from mild disruption to comprehensive cognitive automation
The durability dimension is the mechanism for conditionality. Rather than ranking policies by a single score, the framework asks which interventions hold up across multiple possible futures.
The challenge case for acting now
The essay makes a specific argument for acting pre-emptively rather than reactively. It notes that during the Industrial Revolution and early 20th-century economic transitions, policy typically came after the disruption had occurred. It identifies three structural challenges that make proactive AGI policy harder than past transitions: the lack of granular and timely labor-market data that can detect AI-driven shifts in real time, difficulty determining which policies distribute AGI’s benefits while preserving worker agency, and the absence of a common rubric for comparing interventions on standardised dimensions.
The framework is positioned as an answer to the third challenge. The AI agent raters approach is presented as a way to make comparative policy analysis tractable at a scale that would be impractical using only human expert panels.
What the essay does not do
The essay does not prescribe a single policy or rank the 11 interventions by a single score. It does not claim that AGI economic disruption is inevitable: the authors note that economists are currently not observing definitive evidence of systemic employment or wage impacts. The exercise is explicitly preparatory, mapping conditional responses to scenarios that may or may not materialise.
The essay is part of the DeepMind Institute’s inaugural set of publications, which also includes essays on reasoning transparency and principles for a new utopianism, all aimed at making complex academic research on AGI accessible to the public and policymakers.