The economic alignment problem of artificial intelligence

1. Problem Statement

The paper identifies a critical gap in the current discourse on artificial intelligence (AI) safety: the economic alignment problem. While the "alignment problem" in AI research focuses on ensuring AI systems act in accordance with human intentions and values, the authors argue that this technical challenge cannot be solved without addressing the economic system in which AI is developed.

• Core Conflict: Advanced AI is currently being developed within a growth-oriented economic system (capitalism) that prioritizes endless GDP growth, profit maximization, and market competition.
• The Risk: If AI is trained to optimize for economic growth and profit within a system already incompatible with human well-being and ecological sustainability, the resulting AI will likely exacerbate social inequality, environmental destruction, and existential risks (e.g., human extinction).
• The Urgency: AI capabilities are growing exponentially (doubling every ~7 months for task complexity), while the socio-ecological crisis (transgression of planetary boundaries) deepens. The authors posit that developing Artificial General Intelligence (AGI) within a neoliberal, growth-centered framework is a "recipe for disaster."

2. Methodology

The paper employs a qualitative synthesis and theoretical framework analysis, integrating insights from AI safety, ecological economics, and post-growth research. It utilizes several quantitative and empirical components to substantiate its arguments:

• Analysis of Exponential Trends: The authors analyzed historical data (2010–2024) from sources such as the AI Index, Epoch AI, and METR to calculate doubling times for AI metrics (compute power, model size, task complexity). They fitted exponential curves to demonstrate that AI growth far exceeds Moore's Law.
• Scenario Modeling (GDP Projections): Using historical GDP data (1950–2024) and literature reviews, the authors constructed five distinct scenarios for the impact of AI on global GDP per capita (2025–2050):
  1. Intelligence Explosion (30% annual growth).
  2. Baseline AGI (18% stable growth).
  3. Widespread Adoption (modest 7% increase over a decade).
  4. Modest Growth (marginal gain).
  5. Human Extinction (sudden end of growth).
• Empirical Re-analysis of Labor Market Data: The authors conducted a new statistical analysis comparing job satisfaction/meaningfulness with automation risk. They compared data from 2013 (Frey & Osborne) with data from 2023 (Eloundou et al.) to test the hypothesis that automation targets jobs with low satisfaction.
• Application of the Framework: The paper applies the Doughnut Economics framework (social foundation vs. planetary boundaries) to visualize how AI development could bridge social deficits or exacerbate ecological overshoot.

3. Main Contributions

The paper makes several novel theoretical and policy contributions:

• Redefining Alignment: It redefines the alignment problem as inherently economic. It argues that technical alignment is impossible if the underlying economic incentives (profit, growth) do not align with human values.
• Post-Growth as a Solution: It proposes Post-Growth Economics (including Degrowth, steady-state, and Doughnut Economics) as the necessary theoretical foundation for safe AGI. Post-growth prioritizes "sufficiency" and "satisficing" over "optimization" and "maximization."
• The Distinction between "Satisficing" and "Optimizing": The authors argue that AI should be trained to satisfice needs within planetary boundaries, rather than optimize a single metric like GDP or paperclips, which leads to perverse outcomes.
• Tool AI vs. Agentic AI: It distinguishes between "Tool AI" (human-controlled, specialized) and "Agentic AI" (autonomous decision-makers). The paper advocates prioritizing Tool AI to preserve human autonomy and avoid existential risks associated with recursive self-improvement.
• Policy Roadmap: It provides a concrete table of post-growth policy solutions mapped to specific AI risks (e.g., using "Job Guarantees" to combat displacement, "Resource Caps" to prevent rebound effects).

4. Main Findings and Insights

• Exponential Acceleration: AI capabilities are doubling every 7 months (task complexity), implying an annual growth rate of over 200%. By 2030, AI could be 5,000 times more powerful than GPT-4.
• The Automation Paradox: Contrary to 2013 findings, 2023 data shows that highly satisfying, meaningful jobs (e.g., teachers, therapists, executives) are now at high risk of automation by Large Language Models (LLMs). This threatens the "meaning" of human life, not just employment.
• The Rebound Effect: Even if AI improves resource efficiency, the Jevons Paradox (rebound effect) suggests that efficiency gains will be offset by increased consumption and production, leading to net environmental damage unless caps are imposed.
• Inequality and Power: AI development currently concentrates power in the Global North and Big Tech, creating "algorithmic colonialism" and exacerbating global inequality.
• Scenario Outcomes:
  • Growth-Centered Path: Leads to potential human extinction or massive inequality due to the pursuit of infinite growth within a finite system.
  • Post-Growth Path: Offers a path to "radical abundance," where AI serves human needs within planetary boundaries and reduces existential risk.

5. Significance and Implications

• Interdisciplinary Bridge: The paper serves as a critical bridge between the AI Safety Community (focused on technical alignment) and the Post-Growth/Ecological Economics Community (focused on systemic limits). It argues that these fields must converge to solve the AGI challenge.
• Governance Reform: It calls for a fundamental shift in governance, proposing that AI be treated as a global commons rather than private property. It suggests democratic ownership (cooperatives), nationalization of AI labs (modeled on CERN), and the introduction of "compute budgets" and resource caps.
• New Economic Theory: The authors argue that the arrival of AGI requires a new branch of economics ("Post-AGI Economics") that moves beyond market-based, price-based allocation toward needs-based provisioning, potentially facilitated by AI's ability to solve complex allocation problems.
• Existential Stakes: The paper concludes that the choice of economic system determines the outcome of the AI revolution. Without a shift to a post-growth framework, developing AGI carries a higher risk of human extinction than climate change or biodiversity loss.

In summary, O'Neill et al. argue that technical safety is insufficient; the only viable path to safe AGI is to first align the economic system with human well-being and planetary boundaries, utilizing post-growth principles to orient AI development toward "satisficing" rather than "optimizing."