Executive Summary

While regulation can and does solve market failures, the knowledge problem makes it hard to regulate efficiently. Regulators don’t and can’t know for sure whether regulation will work. As a consequence, regulators tend to rely on heuristic theories to decide whether and when to regulate. While those heuristics are better than nothing, they often cause both under- and over-regulation. Antitrust scholars debate whether to promote consumer welfare or competition, rather than whether regulation would make the market more efficient. And copyright scholars debate copyright theory, rather than the outcomes of copyright policy. Maybe AI can help solve the knowledge problem in regulation by not only providing more reliable predictions about whether regulation will work, but also by identifying potential changes to existing regulation. But if we want AI to help us improve regulation, we’ll have to ensure it’s possible to train AI models.