Basically the same as the alignment problem in AI. You need to be very careful of how you define your rewards, because you'll end up incentivizing exactly what

Basically the same as the alignment problem in AI. You need to be very careful of how you define your rewards, because you'll end up incentivizing exactly what you define.