As a precursor to this discussion, I'll remind folks that per the forum rules, discussion of this topic is allowed (and keep it civil), but the posting of any AI-generated content is not and will be warned accordingly.
That being said, I've no experience with it and I don't think I'd ever want to.
Rollback Post to RevisionRollBack
Your Friendly Neighborhood Community Manager (she/her) You can call me LT. :)
CM Hat On| CM Hat Off Generally active from 9am - 6pm CDT [GMT-5]. Thank you for your patience if you message me outside of those hours!
It’s come up here before, and using the existing generative software program it’s nowhere near viable for anything close to a typical D&D experience. The “AI” cannot formulate and track any kind of ongoing narrative- particularly across multiple separate prompts- so the longer it runs the more it’s going to contradict itself, lose the thread, or suddenly introduce a new element from nowhere. It cannot interpret the scope of a spell effect, magic item, etc. A common point I see when people relate stories of trying to get an AI to run D&D is that AI training makes them very permissive, and it’s simply not going to tell someone “no, X action cannot be attempted” even when X is “flap my arms hard enough to fly”. The current software is good at repeating facts and putting fairly generic pictures together, but it simply cannot handle the kind of critical analysis and thinking a DM needs.
I'd say if someone put in the time and effort to actually train a GPT GM, it would be absolutely stellar at it.
Whether it's legal is somewhat beyond me. It would require you to feed the GPT all the rulebooks, adventures, maps and pictures - and so on, all you can imagine, the more the merrier. But if you do all that, there's every reason to believe it would do it very well indeed.
If you could further feed it some actual play - steal uh, borrow some from Youtube, record your own sessions, find whatever seems to be under fair use or in the public domain.
Top that off with some strategic use of the memory you get for it, and .. yea, I'm quite certain you'd get a very good GM indeed. Frankly, way better than most human ones. Better than me, certainly, at least at the rules lawyer part of the game.
Rollback Post to RevisionRollBack
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
It cannot independently parse the rules, a situation, or a narrative. How on Earth is it going to function as a DM? It’ll just regurgitate a mish mash of scenes it’s been trained on, with no capacity for consistent roleplay.
Roleplay really is the province of the players. NPC's do have 'roles', but they are mostly quite brief, and end poorly. So in terms of running the GM side of an adventure, I have little doubt in it's ability.
Further, I think you underestimate it. I use AI for a campaign involving a barbarian highland kinda place. It remembers narrative details far better than I do - that's a part of what I use it for.
And that's just a ... bog standard GPT. Not trained for it specifically.
Obviously, I'm not saying it can create campaigns. It's as creative as falling down a well.
But it can run a premade adventure. Easily.
It may require the players to play nice with it. Draw too far outside the lines, and my conviction faltes considerably.
As an aside, GPT agrees with me - even when I ask for it's most conservative estimation - but comes up with a list of things it might struggle with, that I can't just dump here. But I can paraphrase: It would struggle with long term memory (plot details from sessions past), rules improvisations, players drawing outside the lines. It isn't capable of: Running the map! Keeping track of details of PC backstories. And player emotions.
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
As an aside, GPT sometimes has the greatest possible conviction it can do stuff it absolutely cannot even get close to doing.
I once asked it for a bakery near where I was. It failed at that so eloquently it's hard to even begin to describe. It just plucked up quality bakeries from across the country (made some up too) and dumped them right near my location. 'Oh, right over on Main Street, there's this excellent bakery that's rated 100 at Rate Site' except there's zero bakeries on main street. I know, that's why I asked.
Rollback Post to RevisionRollBack
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
It may require the players to play nice with it. Draw too far outside the lines, and my conviction faltes considerably.
As an aside, GPT agrees with me - even when I ask for it's most conservative estimation - but comes up with a list of things it might struggle with, that I can't just dump here. But I can paraphrase: It would struggle with long term memory (plot details from sessions past), rules improvisations, players drawing outside the lines. It isn't capable of: Running the map! Keeping track of details of PC backstories. And player emotions.
Okay, so…why would I want to play with a DM like that? Why would I want a DM who can’t keep track of PC backstories or player emotions? I genuinely do not understand the appeal of this.
AI can certainly assist DMing, there's a great deal of filling in of minor details that it's quite competent at, but current generation AIs aren't really capable of managing an extended plot or parsing a rules text.
I don't understand why anyone would want to play D&D without other people.
Because not everyone can get four others to come play with them as a group. My wife and I can't do it - with young kids, schedules are volatile and so getting people over or going to visit them for 3hrs give or take is impossible to do on a weekly basis. An AI DM would be ideal, if it were viable. Such as it is, we have to make do with DMPCs, sidekicks and other things to try and make things work. Or play different TTRPGs that are less volatile to small parties.
Rollback Post to RevisionRollBack
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Just wondering has anyone consider this as plausible? If so, how would it look like?
Right now? Not really. You could get some semblance of a game going and it will be on, but really, it would not be up to the standard of a real DM for the reasons TAoR gave. It just has so many issues and problems that ultimately, it will become an exercise in frustration.
In the future? It'll work. In principle, DMing is right up AI's alley. It can use the data it's fed to create interesting stories and so forth, and then it's just applying the rules - which again, is in theory what AI should be good at.
The AI at the moment is technically impressive but in practice is pretty poor. However, it's advancing very quickly. A couple of years ago when it started becoming mainstream, it was ridiculously easy to spot AI content. The pictures were just freaky to look at. A person would have a foot instead of a hand, for example. Now, it generates video so well, that you have to be consciously looking for the signs to know that it's AI and not real. The signs are still there of course, but it's not so obvious. Odd behaviours of subjects instead of the subjects being weirdly depicted, for example, or odd reflections rather than extra limbs.
So the next few years, AI DMs will remain pretty poor substitutes. But the In the next couple of decades? I think AI will surpass humans in ability and skill.
Rollback Post to RevisionRollBack
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Yes some people have considered and pitched it as possible/plausible, and my opinion of those people can only be expressed by words i am not allowed to say in these forums.
suffice to say you would burn a lot of power for poor results.
Rollback Post to RevisionRollBack
He/Him. Loooooooooong time Player. The Dark days of the THAC0 system are behind us.
"Hope is a fire that burns in us all If only an ember, awaiting your call To rise up in triumph should we all unite The spark for change is yours to ignite." Kalandra - The State of the World
I don't understand why anyone would want to play D&D without other people.
Because not everyone can get four others to come play with them as a group. My wife and I can't do it - with young kids, schedules are volatile and so getting people over or going to visit them for 3hrs give or take is impossible to do on a weekly basis. An AI DM would be ideal, if it were viable. Such as it is, we have to make do with DMPCs, sidekicks and other things to try and make things work. Or play different TTRPGs that are less volatile to small parties.
Sure, I’m familiar with the difficulties of scheduling a D&D session. I just don’t see the appeal of this as a solution. All the other things you said sound way more fun.
Because not everyone can get four others to come play with them as a group. My wife and I can't do it - with young kids, schedules are volatile and so getting people over or going to visit them for 3hrs give or take is impossible to do on a weekly basis. An AI DM would be ideal, if it were viable.
If you're willing to be using a computer to do your playing (which would be necessarily the case for an AI DM), why not just play on a VTT? Between removing the need for commute times, and the ability to get up in the middle of the game if it's essential, it really does make scheduling much easier (and it allows playing with people who are non-local, one of the players in my current game lives 1500 miles away). However, if you want a computer DM with what's available with current tech, I suggest a cRPG.
The capacities of AI are vastly oversold, to track a campaign would take a significant amount of resources that modern AI generally does not have. Most AI these days relies on Humans to do the memorisation for them and to pick up and correct inconsistencies. Going from a Town, to a Dungeon and back to Town, most AI systems would completely forget how the town looked, what NPCs exist there, the promised rewards for the Dungeon quest. As an Aid to DMing then AI maybe able to assist a lot but given that AI is just mimicking things produced by Humans, it creates a serious number of issues in it's usage.
In the future? It'll work. In principle, DMing is right up AI's alley. It can use the data it's fed to create interesting stories and so forth, and then it's just applying the rules - which again, is in theory what AI should be good at.
It's not going to get meaningfully better. Large Language Models, however they dress them up, have fundamental limitations that cannot be solved without some kind of breakthrough.
For instance, they cannot track hidden state. The entirety of their state is the conversation. There are parts of the conversation that are hidden from you at the beginning, but the LLM cannot decide "the noble is a vampire", and drop subtle hints throughout the adventure, while making sure they don't, for instance, go out in the sun. If somebody were to set up an LLM DM with a scenario where the noble is a vampire, and that's hidden in the setup prompt, it's likely either going to leak, or not get revealed when it should be.
Similarly, they can't "apply the rules", because they don't know what rules are.
LLMs are a remarkable technology, but they're not actually capable of doing many of the things they're hyped as being good for, and whether the things they can do are worth doing at the expense of the resources they consume is questionable, at best.
LLMs are a remarkable technology, but they're not actually capable of doing many of the things they're hyped as being good for
This is a big part of the problem. Often the people promoting these things are either deliberately and maliciously misconstruing what they're actually capable of, or they're genuinely unaware of how the things they're promoting work and think they can do a lot of things that they will never be able to do. It often feels like they're completely disconnected from reality.
Just wondering has anyone consider this as plausible? If so, how would it look like?
-
View User Profile
-
View Posts
-
Send Message
D&D StaffAs a precursor to this discussion, I'll remind folks that per the forum rules, discussion of this topic is allowed (and keep it civil), but the posting of any AI-generated content is not and will be warned accordingly.
That being said, I've no experience with it and I don't think I'd ever want to.
Your Friendly Neighborhood Community Manager (she/her)
You can call me LT. :)
CM Hat On | CM Hat Off
Generally active from 9am - 6pm CDT [GMT-5].
Thank you for your patience if you message me outside of those hours!
Useful Links: Site Rules & Guidelines | D&D Educator Resources | Change Your Nickname | Submit a Support Ticket
It’s come up here before, and using the existing generative software program it’s nowhere near viable for anything close to a typical D&D experience. The “AI” cannot formulate and track any kind of ongoing narrative- particularly across multiple separate prompts- so the longer it runs the more it’s going to contradict itself, lose the thread, or suddenly introduce a new element from nowhere. It cannot interpret the scope of a spell effect, magic item, etc. A common point I see when people relate stories of trying to get an AI to run D&D is that AI training makes them very permissive, and it’s simply not going to tell someone “no, X action cannot be attempted” even when X is “flap my arms hard enough to fly”. The current software is good at repeating facts and putting fairly generic pictures together, but it simply cannot handle the kind of critical analysis and thinking a DM needs.
I don't understand why anyone would want to play D&D without other people.
pronouns: he/she/they
I'd say if someone put in the time and effort to actually train a GPT GM, it would be absolutely stellar at it.
Whether it's legal is somewhat beyond me. It would require you to feed the GPT all the rulebooks, adventures, maps and pictures - and so on, all you can imagine, the more the merrier. But if you do all that, there's every reason to believe it would do it very well indeed.
If you could further feed it some actual play -
stealuh, borrow some from Youtube, record your own sessions, find whatever seems to be under fair use or in the public domain.Top that off with some strategic use of the memory you get for it, and .. yea, I'm quite certain you'd get a very good GM indeed. Frankly, way better than most human ones. Better than me, certainly, at least at the rules lawyer part of the game.
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
It cannot independently parse the rules, a situation, or a narrative. How on Earth is it going to function as a DM? It’ll just regurgitate a mish mash of scenes it’s been trained on, with no capacity for consistent roleplay.
Roleplay really is the province of the players. NPC's do have 'roles', but they are mostly quite brief, and end poorly. So in terms of running the GM side of an adventure, I have little doubt in it's ability.
Further, I think you underestimate it. I use AI for a campaign involving a barbarian highland kinda place. It remembers narrative details far better than I do - that's a part of what I use it for.
And that's just a ... bog standard GPT. Not trained for it specifically.
Obviously, I'm not saying it can create campaigns. It's as creative as falling down a well.
But it can run a premade adventure. Easily.
It may require the players to play nice with it. Draw too far outside the lines, and my conviction faltes considerably.
As an aside, GPT agrees with me - even when I ask for it's most conservative estimation - but comes up with a list of things it might struggle with, that I can't just dump here. But I can paraphrase: It would struggle with long term memory (plot details from sessions past), rules improvisations, players drawing outside the lines. It isn't capable of: Running the map! Keeping track of details of PC backstories. And player emotions.
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
As an aside, GPT sometimes has the greatest possible conviction it can do stuff it absolutely cannot even get close to doing.
I once asked it for a bakery near where I was. It failed at that so eloquently it's hard to even begin to describe. It just plucked up quality bakeries from across the country (made some up too) and dumped them right near my location. 'Oh, right over on Main Street, there's this excellent bakery that's rated 100 at Rate Site' except there's zero bakeries on main street. I know, that's why I asked.
Blanket disclaimer: I only ever state opinion. But I can sound terribly dogmatic - so if you feel I'm trying to tell you what to think, I'm really not, I swear. I'm telling you what I think, that's all.
Okay, so…why would I want to play with a DM like that? Why would I want a DM who can’t keep track of PC backstories or player emotions? I genuinely do not understand the appeal of this.
pronouns: he/she/they
AI can certainly assist DMing, there's a great deal of filling in of minor details that it's quite competent at, but current generation AIs aren't really capable of managing an extended plot or parsing a rules text.
Because not everyone can get four others to come play with them as a group. My wife and I can't do it - with young kids, schedules are volatile and so getting people over or going to visit them for 3hrs give or take is impossible to do on a weekly basis. An AI DM would be ideal, if it were viable. Such as it is, we have to make do with DMPCs, sidekicks and other things to try and make things work. Or play different TTRPGs that are less volatile to small parties.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Right now? Not really. You could get some semblance of a game going and it will be on, but really, it would not be up to the standard of a real DM for the reasons TAoR gave. It just has so many issues and problems that ultimately, it will become an exercise in frustration.
In the future? It'll work. In principle, DMing is right up AI's alley. It can use the data it's fed to create interesting stories and so forth, and then it's just applying the rules - which again, is in theory what AI should be good at.
The AI at the moment is technically impressive but in practice is pretty poor. However, it's advancing very quickly. A couple of years ago when it started becoming mainstream, it was ridiculously easy to spot AI content. The pictures were just freaky to look at. A person would have a foot instead of a hand, for example. Now, it generates video so well, that you have to be consciously looking for the signs to know that it's AI and not real. The signs are still there of course, but it's not so obvious. Odd behaviours of subjects instead of the subjects being weirdly depicted, for example, or odd reflections rather than extra limbs.
So the next few years, AI DMs will remain pretty poor substitutes. But the In the next couple of decades? I think AI will surpass humans in ability and skill.
If you're not willing or able to to discuss in good faith, then don't be surprised if I don't respond, there are better things in life for me to do than humour you. This signature is that response.
Yes some people have considered and pitched it as possible/plausible, and my opinion of those people can only be expressed by words i am not allowed to say in these forums.
suffice to say you would burn a lot of power for poor results.
He/Him. Loooooooooong time Player.
The Dark days of the THAC0 system are behind us.
"Hope is a fire that burns in us all If only an ember, awaiting your call
To rise up in triumph should we all unite
The spark for change is yours to ignite."
Kalandra - The State of the World
Sure, I’m familiar with the difficulties of scheduling a D&D session. I just don’t see the appeal of this as a solution. All the other things you said sound way more fun.
pronouns: he/she/they
If you're willing to be using a computer to do your playing (which would be necessarily the case for an AI DM), why not just play on a VTT? Between removing the need for commute times, and the ability to get up in the middle of the game if it's essential, it really does make scheduling much easier (and it allows playing with people who are non-local, one of the players in my current game lives 1500 miles away). However, if you want a computer DM with what's available with current tech, I suggest a cRPG.
The capacities of AI are vastly oversold, to track a campaign would take a significant amount of resources that modern AI generally does not have. Most AI these days relies on Humans to do the memorisation for them and to pick up and correct inconsistencies. Going from a Town, to a Dungeon and back to Town, most AI systems would completely forget how the town looked, what NPCs exist there, the promised rewards for the Dungeon quest. As an Aid to DMing then AI maybe able to assist a lot but given that AI is just mimicking things produced by Humans, it creates a serious number of issues in it's usage.
It's not going to get meaningfully better. Large Language Models, however they dress them up, have fundamental limitations that cannot be solved without some kind of breakthrough.
For instance, they cannot track hidden state. The entirety of their state is the conversation. There are parts of the conversation that are hidden from you at the beginning, but the LLM cannot decide "the noble is a vampire", and drop subtle hints throughout the adventure, while making sure they don't, for instance, go out in the sun. If somebody were to set up an LLM DM with a scenario where the noble is a vampire, and that's hidden in the setup prompt, it's likely either going to leak, or not get revealed when it should be.
Similarly, they can't "apply the rules", because they don't know what rules are.
LLMs are a remarkable technology, but they're not actually capable of doing many of the things they're hyped as being good for, and whether the things they can do are worth doing at the expense of the resources they consume is questionable, at best.
This is a big part of the problem. Often the people promoting these things are either deliberately and maliciously misconstruing what they're actually capable of, or they're genuinely unaware of how the things they're promoting work and think they can do a lot of things that they will never be able to do. It often feels like they're completely disconnected from reality.
pronouns: he/she/they
It is not plausible for me to play any game with stolen property and a known thief.
That's a hard no from me. Sorry.
Anzio Faro. Protector Aasimar light cleric. Lvl 18.
Viktor Gavriil. White dragonborn grave cleric. Lvl 20.
Ikram Sahir ibn-Malik al-Sayyid Ra'ad. Brass dragonborn draconic sorcerer Lvl 9. Fire elemental devil.
Tayn of Darkwood. Human Life Cleric. Lvl 10.