THE SIMPLE MACROECONOMICS OF AI

2024

Open publication workspace · Sign in to read the full PDF.

AI-generated summary

1) This paper examines how different regulatory learning assumptions about potential harms impact the socially optimal rate of artificial intelligence adoption.
2)
- Compares "learning by doing" (where real-world AI deployment is required to surface harms) with "lab learning" (where safety is evaluated independently of adoption over time).
- Demonstrates that learning by doing generally favors accelerating AI adoption to uncover and react to potential harms quickly, whereas lab learning and irreversible adoption can favor delays.
- Explores policy interactions, including harm mitigation investments, insurance mechanisms, and the promotion of diverse research architectures.

3) Tags: artificial intelligence, economics, regulation, technology adoption, public policy

Check the original publication for accuracy and context.