Conference Report
A two-day conference in Oct 2024 bringing together 30 leading economists, AI policy experts, and professional forecasters to rapidly evaluate the economic impacts of frontier AI technologies by 2030.
Promising Future Research Ideas
Our report authors identified nine plausible research questions, split into four categories, stemming from upcoming challenges identified by attendees.
1. Diffusion & Market Structure
How will changes in market structure for AI (e.g. a commoditized, competitive industry with little differentiation, vs. a small oligopoly of leading firms) shape the pace of AI adoption? What are the economic outcomes of concentrated vs. broad-based AI ownership?
Underlying many of the predictions of our attendees in the worldbuilding exercises is the assumption that AI systems will be owned and deployed by a smaller set of leading technology firms. This raises crucial questions about how different market structures would affect AI diffusion rates and economic outcomes. Understanding these dynamics is essential for developing effective policy responses to economic power concentration.
What benchmarks or AI evaluations will predict the elasticity of substitution between AI capital and human labor across different industries? How will these predict labor demand and hiring trends?
Attendees noted that AI models are rapidly surpassing all well-scoped quantitative benchmarks. Performance on scoped cognitive tasks is rapidly approaching or surpassing expert-level performance. However, AI systems are still not yet capable of performing the vast majority of human cognitive labor roles. Existing benchmarks do not correspond well to AI system’s capability to perform real-world human jobs. New, qualitative benchmarks must be developed to measure the critical skills and tasks necessary to fully automate human jobs.
2. Labor Market Dynamics
How will AI-driven automation affect the labor share of income, and which policy interventions will be most effective in mitigating negative labor externalities?
Though conference debates revealed significant uncertainty around labor share trajectories, our attendees discussed how declining labor share will drive income inequality, social instability, and a lack of purpose for workers. Research is needed on a wide variety of policy interventions, including UBI, wage subsidies, re-skilling programs, taxation of AI-driven profits, unemployment benefits, and more.
How will AI-driven automation affect employment and wages in developing countries, especially in sectors traditionally reliant on low-skill labor? What policies might facilitate a more equitable transition?
Our report discusses how developing countries may face reduced global demand for traditionally outsourced goods and services, even while simultaneously providing essential inputs such as natural resources or data to the global AI supply chain. We need research to model the plausible economic impacts and propose equitable economic transitions in the Global South.
What mechanisms support effective human capital development in an AI-dominated economy, and how can education and training systems be redesigned to promote workforce resilience?
AI job reskilling, improved education, and job transition funds were all suggested by attendees as plausible economic policy responses to AI automation. However, there is limited existing research describing how to adjust educational curriculums and vocational programs. Researchers should explore what types of skills and educational programs would improve human outcomes in a post-AI society.
3. Income and Wealth Distribution
What tax and transfer mechanisms best ensure equitable distribution of AI-driven gains while maintaining robust incentives for innovation and investment?
Policymakers will face significant trade-offs between fostering AI innovation (e.g. R&D incentives) and mitigating externalities such as income inequality, labor displacement, and social instability. Tax mechanisms will play a large role in this challenge. Policy researchers should conduct modeling on the tax base distortions of AI, and evaluate the role of corporate profit taxes, automation taxes, capital gains taxes, subsidies, or robot taxes on societal outcomes in different future scenarios.
What are the socioeconomic consequences of a shift from labor to capital from AI systems? How will this dynamic change the concentration of wealth over time?
Group 3's Palma Inequality Model described plausible methods by which AI could exacerbate wealth concentration. If AI capital is highly concentrated, wealth gaps could expand. Classical and neoclassical theories suggest rising capital share could exacerbate inequality. Beyond inequality, replacement of labor with AI could lead to socioeconomic consequences such as reduced investment in human capital, reduced incentive mechanisms for education or lower social mobility.
4. Quality of Life and Social Welfare
What are the potential implications of AI for social welfare systems, and how can these systems be adapted to provide adequate support for individuals and families in an AI-driven economy?
Attendees anticipate that widespread automation may lead to increased demand for social safety nets such as UBI, unemployment benefits, negative income taxes, or job transition funds. Concrete research into social welfare systems and policies that are adaptable, responsive to local demands, and effectively protect key human outcomes will permit governments to enact effective response strategies to significant unemployment.
What are the appropriate metrics to capture the impact of AI adoption on quality of life?
Group 5’s Quality of Life Model describes how AI will impact various dimensions of human well-being beyond economic indicators, such as mental health, social cohesion, and physical safety. Demonstrating a causal relationship between quality of life outcomes and AI diffusion by identifying specific measurable observables would significantly increase awareness for the social impact of advanced AI systems.