The 2024 article by Charles I. Jones, published in the American Economic Review Insights, explores the ethical trade-off between accelerated economic growth driven by AI and the existential risk it poses to humanity. The author proposes a mathematical model to analyze the optimal level of AI adoption, weighing potential increases in consumption and life expectancy against the probability of human extinction.
The model highlights the role of the utility function in decision-making. With a logarithmic utility function, the model suggests a high tolerance for risk, potentially accepting a one-in-three chance of extinction in exchange for significant consumption gains. However, when using a CRRA utility function with a coefficient of relative risk aversion of two or higher, the model becomes significantly more conservative regarding existential threats.
The analysis also addresses singularity scenarios where self-improving AI leads to infinite consumption. Even in these cases, if the utility function is bounded, the model maintains a cautious approach toward existential risk. A critical exception occurs when AI-driven innovations significantly increase life expectancy, as this benefit is measured in the same units as survival, potentially justifying higher risks.
The author defines the trade-off using parameters such as the initial consumption level, the economic growth rate—potentially reaching 10 percent annually—and the intensity of AI usage. The risk of catastrophe is modeled as a Poisson process, where the probability of survival decreases exponentially the longer society relies on advanced AI. The study aims to clarify these economic forces rather than provide a definitive policy answer.
