Yeah, it was a bit frustrating for me just how many people didn’t already have some grasp of that problem, but as an AI programmer I’ve had to deal with this all the time so it’s hard for me to judge how widespread it should be. Coming from that background, too, their solutions are the lazy solution – you’re just avoiding making the model deeper because you think it will be harder to communicate that than some fuzzy probabilities, but you miss opportunities for a richer experience (and, counterintuitively, it often ends up being easier to convey to player when you’re adding rules to the model than when you’re adding a range of probabilities).

I find it really interesting and a little frustrating that the talk Rob Pardo of Blizzard game talked about the EXACT same issue about players’ perceptions of randomness, and how Blizzard adjusts their probabilities to work with how players THINK randomness works instead of how it actually does. Nobody seemed bothered by this talk.