How Much Should We Spend to Prevent an AI Catastrophe? | Diving In
Summary
Justin Wolfers applies economic frameworks to the AI safety debate, using airbags and the value of statistical life to quantify the cost of reducing existential risk. He analyzes how utility functions and medical innovation alter the trade-off between rapid AI growth and catastrophic danger, concluding that society likely underinvests in safety by a factor of 30. Drawing on work by Chad Jones, Wolfers argues for massive public spending on alignment research and independent oversight, comparing the challenge to the nuclear age but noting AI’s unique speed and accessibility.
Topics discussed
Sponsor: MaintainX AI maintenance system
Sponsor: Alma for therapists
Introduction: The AI safety debate and economics
Episode roadmap and host introduction
Defining the risks of powerful AI
The airbag analogy: Cost vs. lives saved
Value of a Statistical Life (VSL) explained
Utility functions and material abundance
Log utility: Growth vs. existential risk
Sponsor: IXL online learning platform
Sponsor: Rosetta Stone Sapphire language app
Ceiling on utility: When growth stops helping
Medical innovation and the trade-off for lives
Ethics, global inequality, and moral weight
Sponsors: NerdWallet, RingCentral, T-Mobile
Calculating the cost of AI safety investments
Regulatory models: The 'chaperone' approach
Comparison with nuclear weapons history
Conclusion: The case for caution
Sponsor: Jewelers Mutual insurance
Listen ad-free on Castria