An MIT paper argues that the best way to regulate potentially risky technologies like AI is through gradual adoption while monitoring for negative impacts. via Marginal Revolution.