Training AI To Create Encounters

[Kaodi]

I was wondering yesterday whether in the future an AI could be trained to make custom balanced encounters and even whole schedules of encounters. Probably not the fluff, but the crunchy parts that make things interesting. I imagine it could be a huge time saver and could also make more freeform activities a lot easier to run.

Right now we roughly gauge encounter difficulties with things like challenge ratings and relative level but an AI could easily crunch the numbers to gauge difficulty based on odds of success, character deaths, and the like. So if the AI had the stats and resources of the characters it could possibly fine tune challenges for several criteria. Like if it is the final battle you could target a much lower success rate.

It might also be funny to see what sort of wacky combinations of creatures it might come up with, and to see what sort of explanations you could scrape together to justify them.

 

[Jacob Lewis]

We've had that kind of thing for years! Never had a problem.

 

[MNblockhead]

Sure. I could see someone like Roll 20 eventually having AI-run games. It would have to be very advanced before I would be interested. Video games are one thing but I like my TTRPGs and board games to help reconnect with other natural persons.

Where I can see AI being useful at the table is as a DM's assistant. Actually, in my last game one of the players was testing an Amazon Alexa app for looking up D&D rules. I was pretty awful, but I fully expect that it will rapidly improve and asking Siri or Alexa or Google for a D&D rule will be quite useful.

AIs could also be helpful for tracking combat and generating random encounters.

 

[Sword of Spirit]

The other day I was thinking that it would be great if an AI could convert content from previous editions of D&D to the current edition. In particular, giving it thousands of monsters and having it pop out 5e appropriate conversions would be awesome, since that's one of the biggest mechanical roadblocks for continuing older campaigns with newer rules.

 

[Blue]

Being able to Monte Carlo the numbers of an encounter in a white room we can do now - and does nothing in regards to actual tactics. Knowing that "oh, that Hypnotic pattern got a lot of them so I can go more aggressive here" sort of thing adds in so many decision points though. You'd need to model behavior, and not just "perfect tactical" but also how players like to run their characters (cautious, etc.) and the party dynamic. That'll take doing.