The author of one of my favorite roleplaying game blogs — he uses the handle "Noisms" and the blog is called Monsters & Manuals — recently tried an interesting experiment to see how well ChatGPT could run a solo roleplaying session using his own published game setting "Yoon-Suin." Short answer: not well.
It's worth reading his post to see how the chatbot failed.
One problem was that its scenario seemed too generic, but that could be the result of asking it to use a somewhat obscure and quirky game setting. It might have done a better job using something with abundant on-line documentation like the Forgotten Realms or Golarion. (And, to be cranky, both of those settings are pretty generic anyway, so who would notice?)
The second problem was that the GM-bot simply assented to whatever the player decided to do — while simultaneously doing its best to "railroad" the player into a specific course of action. I was very amused by this, because this sounds exactly like the style of "cooperative storytelling" gamemastering which exists at one vertex of the Game-Narrative-Simulation triangle. Amused, because the flaws that ChatGPT displays are exactly the things some gamers find unsatisfying about that style of roleplaying: no meaningful challenges, and you're stuck in a plot you can't escape.
"Artificial Intelligence" (a term which gets overused even more egregiously than "force field") has always been unsatisfying at running games. In online fantasy combat games the AI-controlled NPCs are notorious for only being able to draw from a limited menu of canned responses. ChatGPT simply expands that menu to the entire online world, but ultimately it's still just spitting back something vaguely appropriate based on keywords.
In early roleplaying games the GM was often called the Referee or the Judge, a holdover from tabletop wargames, but it's worth remembering that all gamemasters still have that job when running a game. Judging requires judgement, and that's something which ChatGPT just doesn't have.
So for now, at least, gamemasters are safe.
Recent Comments