No, you'd want to bootstrap the process with some kind of machine learning to categorize humor and non-humor. It would have poor accuracy, no doubt, but it wou

No, you'd want to bootstrap the process with some kind of machine learning to categorize humor and non-humor. It would have poor accuracy, no doubt, but it would be better than random generation. I agree that it would be less efficient than in humans but with learning it might be able to converge on a set of strategies, possibly "cheats" that are likely to produce laughs.