Live data from Hacker News

The Real Challenge Is Teaching AI to Play D&D

wired.com

31–40 of 83 posts

Re: The Real Challenge Is Teaching AI to Play D&D

#31
post #26

This is sort of like asking: If you had Star Trek replicators, what would be the optimal configuration to use those replicators as paperweights? D&D is not a game with rigorously defined rules and winning conditions, instead it's a means to an end, the end being: have fun with friends through a mutual fantasy. If you had a genuine post-singularity superhuman AGI, it would be silly to use it for D&D, because you could…

> genuine post-singularity superhuman AGI

Except we don’t have that. AI research is mostly about taking a problem that’s just a little to difficult to actually do and working on it. A great DM AI May not need to understand what’s going on any more than a jeopardy AI, it just needs to be good enough to work.

A great many things like image classification initially look like they need AGI, but you can make something useful without it.

Re: The Real Challenge Is Teaching AI to Play D&D

#32
post #19

I think creating an AI dungeon master has a wide range of difficulty depending on how open ended the game is intended to be. The more structured and rules based the game is, the easier it is to create a dungeon master - the more free form and "on the fly" the game is the harder the task. So while creating a dungeon master may be "easy" if the rules and story are heavily enumerated, it likely won't be very fun as it d…

D&D isn't even the most free form RPG out there, there are plenty of systems that run the gamut from "highly tactical and rules based" to "almost freeform collaborative story telling."

D&D at least has well defined combat rules. I've ran games that use cinematic combat rules, where the books have suggestions and some math, but a lot of things are left up in the air with the general guidance of "whatever is awesome."

So there aren't rules for shooting grenades out of mid air, but it made for an awesome climax to the story as the team raced onto the escape ship and the hanger bay door was slamming down, so of course I allowed it.

I was playing one game where we the players objected to how the military was prepping for an AI invasion so we rebelled, next session the DM brought in a giant map and it turned into a base building tower defense session.

D&D can go off the rails, other systems can go off the rails then continue until things are off planet!

Re: The Real Challenge Is Teaching AI to Play D&D

#33
post #26

This is sort of like asking: If you had Star Trek replicators, what would be the optimal configuration to use those replicators as paperweights? D&D is not a game with rigorously defined rules and winning conditions, instead it's a means to an end, the end being: have fun with friends through a mutual fantasy. If you had a genuine post-singularity superhuman AGI, it would be silly to use it for D&D, because you could…

"dangerous"? Edit: I personally know a good amount of people for that is a life goal.

Depends on your definition of dangerous. Dangerous in the same sense that a button you could press to directly stimulate your brain's pleasure centers would be dangerous, because you'd probably end up pressing it to the exclusion of everything else (personal hygiene, family, eating...)

Re: The Real Challenge Is Teaching AI to Play D&D

#34
post #13
post #2

I didn’t realise D&D was this challenging. On the flip side, this is what people said. Forget Chess - Go is the one. Forget Go - D&D is the one.

I haven't played D&D in a looong time, but making a human-like dungeonmaster seems like the textbook example of a hard problem for AI.

It's a hard enough problem for many humans... I completely agree with you.

If someone can make even a decent AI DM, I'd say turn that one loose on NP-complete for a followup softball question.

Re: The Real Challenge Is Teaching AI to Play D&D

#36
post #33

Earlier quoted context omitted.

"dangerous"? Edit: I personally know a good amount of people for that is a life goal.

Depends on your definition of dangerous. Dangerous in the same sense that a button you could press to directly stimulate your brain's pleasure centers would be dangerous, because you'd probably end up pressing it to the exclusion of everything else (personal hygiene, family, eating...)

This is called the wireheading problem [1]. It was solved (for AI) by Stuart Russell in his new book Human Compatible [2].

[1] https://wiki.lesswrong.com/wiki/Wireheading

[2] https://slatestarcodex.com/2020/01/30/book-review-human-comp...

Re: The Real Challenge Is Teaching AI to Play D&D

#37
post #32
post #19

I think creating an AI dungeon master has a wide range of difficulty depending on how open ended the game is intended to be. The more structured and rules based the game is, the easier it is to create a dungeon master - the more free form and "on the fly" the game is the harder the task. So while creating a dungeon master may be "easy" if the rules and story are heavily enumerated, it likely won't be very fun as it d…

D&D isn't even the most free form RPG out there, there are plenty of systems that run the gamut from "highly tactical and rules based" to "almost freeform collaborative story telling." D&D at least has well defined combat rules. I've ran games that use cinematic combat rules, where the books have suggestions and some math, but a lot of things are left up in the air with the general guidance of "whatever is awesome."…

Forget D&D. I won't be satisfied until the computer is GM for a nice game of Paranoia.

Re: The Real Challenge Is Teaching AI to Play D&D

#38
A lower bar might be a Paranoia[0] GM. For those that haven't played, Paranoia is a tabletop RPG that takes place in a dystopian future ruled by an insane/evil AI. It is hilarious fun. I could see it being easier for an AI to RP an insane AI than a D&D DM. :D

[0] https://en.wikipedia.org/wiki/Paranoia_(role-playing_game)

Re: The Real Challenge Is Teaching AI to Play D&D

#39
post #30
post #18

Earlier quoted context omitted.

This is kind of the point I was making with my question. Playing DnD is not about winning. Being a DM is not about reaching some sort of conclusion. So it's essentially impossible to make an AI good at this without achieving true general AI (not within our lifetime for sure)

True, but I can imagine a world where AI DMs are in vogue, and are curated not for their one-size-fits-all perfection, but for their aesthetic or functional qualities. The hardest thing to capture will be truly surprising off-screen storytelling moments that role playing games enable. Otherwise, you can probably optimize for 'challenging combat,' 'fun puzzles,' 'social interaction' etc... Have a hard-nosed, Old Schoo…

interesting, i didnt know about that setting in rimworld. the game has been high on my list due to the fact ive had much FUN with DF, so im gonna bring rimworld up on that list!

Re: The Real Challenge Is Teaching AI to Play D&D

#40
post #33

Earlier quoted context omitted.

"dangerous"? Edit: I personally know a good amount of people for that is a life goal.

Depends on your definition of dangerous. Dangerous in the same sense that a button you could press to directly stimulate your brain's pleasure centers would be dangerous, because you'd probably end up pressing it to the exclusion of everything else (personal hygiene, family, eating...)

If you have a superhuman AGI willing to do what you ask of it, just make sure you ask it to take care of that other stuff too.
Post reply on HN