How learning about harms impacts the optimal rate of artificial intelligence adoption

SUMMARY This paper examines recent proposals and research suggesting that artificial intelligence (AI) adoption should be delayed until its potential harms are fully understood. Conclusions on the social optimality of delayed AI adoption are shown to be sensitive to assumptions about the process by which regulators learn about the salience of particular harms. When such learning is by doing – based on the real-world adoption of AI – this generally favours acceleration of AI adoption to surface and react to potential harms more quickly. This case is strengthened when AI adoption is potentially reversible. This paper examines how different conclusions regarding the optimality of accelerated or delayed AI adoption influence and are influenced by other policies that may moderate AI harm.