The good and bad news is that publishing’s social proof is about to lose 98 percent of its value once NLP gets to the point of having as much predictive value as publishing’s signals. We might be there already. That doesn’t mean a program can decide what is and is not great literature. It doesn’t have to. It just to be better than the curators we have. That could probably be written in a hackathon with a decent labeled dataset.

The good news is that this will make it a lot easier for good writers without connections or preexisting celebrity to be discovered. The bad news is that there will be a few people who reverse engineer the AI graders and get millions of readers despite not being all that good.