• The VOIDRUNNER'S CODEX is coming! Explore new worlds, fight oppressive empires, fend off fearsome aliens, and wield deadly psionics with this comprehensive boxed set expansion for 5E and A5E!

RPG Evolution: The AI DM in Action

How might WOTC launch an AI-powered DM assistant?

How might WOTC launch an AI-powered DM assistant?

technology-4256272_1280.jpg

Picture courtesy of Pixabay.

We know Wizards of the Coast is tinkering with Artificial Intelligence (AI)-powered tools for its multiple properties, including Dungeons & Dragons. But what might that look like in practice?

Interactive NPCs​

Large Language Model (LLM) AIs have been used extensively to create non-player characters of all stripes on Character.AI. It's not a stretch to imagine that Wizards might have official NPCs included as part of the digital purchase of an adventure, with the rough outline of the NPC acting as parameters for how it would interact. DMs might be able to create their own or modify existing NPCs so that the character drops hints or communicates in a certain way. Log outputs could then be available for DMs to use later.

There are several places today where you can create NPC bots powered by AI that are publicly available, although the DM might need to monitor the output in real time to record the conversation. Character.AI and Poe.com both provide the ability to create publicly available characters that players can interact with .

Random Generators​

There are already dozens of these in existence. What's particularly of note is that AI can go deep -- not just randomize what book is in a library, but provide snippets of text of what's in that book. Not just detail the name of a forgotten magic item, but provide stats for the item. For WOTC products, this could easily cover details that no print product can possibly encompass in detail, or with parameters (for example, only a library with books on necromancy).

AI RPG companion is a great example of this, but there are many more.

Tabletop Assistants​

Hasbro recently partnered with Xplored, with the goal of developing a "new tabletop platform that integrates digital and physical play." Of particular note is how Xplore's technology works: its system "intelligently resolves rules and character behaviors, and provides innovative gameplay, new scenarios and ever-changing storytelling events. The technology allows players to learn by playing with no rulebook needed, save games to resume later, enables remote gameplay, and offers features like immersive contextual sound and connected dice."

If that sounds like it could be used to enhance an in-person Dungeons & Dragons game, Xplored is already on that path with Teburu, a digital board game platform that uses "smart-sensing technology, AI, and dynamic multimedia." Xplored's AI platform could keep track of miniatures on a table, dice rolls, and even the status of your character sheet, all managed invisibly and remotely by an AI behind the scenes and communicating with the (human) DM.

Dungeon Master​

And then there's the most challenging aspect of play that WOTC struggles with to this day: having enough Dungeon Masters to support a group. Wizards could exclusively license these automated DMs, who would have all the materials necessary to run a game. Some adventures would be easier for an AI DM to run than others -- straightforward dungeon crawls necessarily limit player agency and ensure the AI can run it within parameters, while a social setting could easily confuse it.

Developers are already pushing this model with various levels of success. For an example, see AI Realm.

What's Next?​

If Hasbro's current CEO and former WOTC CEO Chris Cocks is serious about AI, this is just a hint at what's possible. If the past battles over virtual tabletops are any indication, WOTC will likely take a twofold approach: ensure it's AI is well-versed in how it engages with adventures, and defend its branded properties against rival AI platforms that do the same thing. As Cocks pointed out in a recent interview, WOTC's advantage isn't in the technology itself but in its licenses, and it will likely all have a home on D&D Beyond. Get ready!
 

log in or register to remove this ad

Michael Tresca

Michael Tresca

cbwjm

Seb-wejem
I absolutely love using chatgpt and Bing image generators for my games. I've used chatgpt to come up with adventures and I even used it to generate a campaign setting, it came up with some cool stuff using my inputs.

Using the Bing image generator I've created portraits for my PCs, some images of the gods, and backgrounds to help set a scene. It's all great stuff.

Not sure about using it to replace a DM, though perhaps in the future. There are a lot of people that want to play but can't put a group together, maybe there's only two players, both I could see using an AI to run a game.
 

log in or register to remove this ad

You can get better responses through prompt engineering, but that only goes so far, and at a certain point you're spending more time talking to the LLM than you are actually getting the work done.
Yes I remember realizing this when I was trying to get a list of team names for fantasy sport out of it, that I'd spent far long asking it to come up with better stuff/refine what it was doing than I could haver just come up with the names myself.
 

I've got to say if you think all current AI art looks "absolutely terrible", your standards are either exceedingly high, or you're going in with such an expectation it ll be you're projecting on it. Art AIs can produce too wide a variety of work for that to otherwise make sense. The most you can say is that versions that don't have style prompts have a tendency to look very similar, probably from having been trained on mostly the same material.
I disagree Thomas, not entirely but even if you do use quite elaborate style prompts, AI still strongly tends to do certain things that make it look both bad and same-y - one particular one is, if it's doing anything non-photographic, to use multiple different approaches to shading in the same image, something that real artists very rarely do, and where they do, it's very selective and careful - whereas with AI art it's absolutely routine. And it doesn't look good unless except to dimwits who just like the most things to be going in an image. Their idea of quality is just a lot of things happening.

And with only "normal" levels of style prompt, it tends to look more than just "same-y", it looks bland and low-quality in a peculiar "detailed but crap" way that absolutely screams "AI ART!!!" at the top of its lungs. It's very identifiable because normal artists just don't spend that level of effort (fake effort in this case) on a crap-looking, uninspired image - not even bad ones!

It's really a bit like rolling a d20 hoping for a natural 20, because every so often, with the exact same prompts, it'll produce something, that so long as no-one looks at it too closely, is passable and not immediately obvious as AI art. But much as "prompt engineers" like to pretend it it's a skill, it demonstrably isn't. You can use the same prompt a dozen times and get a dozen very very different images, and maybe one of them, if you're lucky, is okay - and this is just what people who claim to be good with prompts do - pay enough to be able to spam images, and get it to spam until it gets one that isn't ghastly - ironically they are showing a small skill - they can usually identify a less-awful image on the pile of crap they have.

It also seems to be getting worse, not better, re-same-y-ness (though the "mistake level" is going down in some ways - not the shading issue though), and I don't expect that to change it increasingly starts ingesting it's own dung. Only to get more same-y.

With more photographic-style stuff, I haven't used every tool recently, but a couple I did use which were supposed to be "good" according to AI art fans were only really "good" at seemingly completely ignoring prompts and producing really stunningly bland and samey images that all looked like there were from clothing catalogues or clothing websites, which was kind of interesting.
 

Thomas Shey

Legend
I disagree Thomas, not entirely but even if you do use quite elaborate style prompts, AI still strongly tends to do certain things that make it look both bad and same-y -

Then you disagree, but I've got to say I've seen enough good look AI art that I really can't take seriously claims that its all bad. It simply doesn't fit my experience. I can see an argument with some of them about the same-y-ness, but not even all there.
 

Dire Bare

Legend
Perhaps within toxic sub-communities in the RPG space on X or YouTube. But in the broader community? No.

X (Twitter) is a cesspool regardless of what you're there for, D&D or something else. And the YouTube algorithm promotes "controversial" videos, both for D&D YouTubers and other fandoms. It's a social media thing, not a D&D community thing.
Forums overlap with social media sites in function, although we don't normally think of them under that category.

But which forums of which do you speak? Not the ENWorld forums, unless you and I are reading entirely different threads. RPGNet has long been known to lean towards the negative, perhaps there?
 

Just as a test, in the heat of a moment during a game session I asked Bard (at the time) to write me a short dream sequence (3-5 paragraphs) for a player to experience a very positive "blessing" from their deity, and juxtaposed that AI-written scene with a sequence I had developed a while back for another player that was framed sort of as a nightmare sequence (though parts were really happening) I was waiting to spring on them. (Don't worry, both things made sense in the story, even if this explanation doesn't, since it loses effectively all of the context ;-P) Important to note that the AI sequence was about a god specific to the Forgotten Realms, and the metaphors I asked it to work in were very simple (something like "what blessing would Lliira give a devoted follower via a dream who was doubting their path?")...but framing it as a dream sequence seemed to me like a tall order for AI to do. I fully expected the prompt to fail, and to just have to wing something completely else.

Turns out the players universally thought that was the best sequence in the campaign up to this point, as I interweaved the two scenes. Mind you, AI only did half the work, and didn't handle the shifting of the spotlight between the two players' experiences, but the fact is that the AI-written portion of that scene was used almost word-for-word, and no one picked up on any difference from my usual GMing aside from calling it out at the end of the session as a really good, strong sequence.

So...I dunno, maybe my "usual" is actually pretty sucky, but I hope not! ;-P

Joking aside, the learning I took away is that AI actually can handle certain, moderately decent prompts pretty damn well when you give it the right parameters. And it did it so fast that the players didn't notice a transition from me running the game completely normally to adding an entire pre-written short sequence, half of which was written on the spot using AI from a 2-3 sentence prompt.
First of all, that's anecdotal evidence, that means nothing because no one of us has been there to witness it in its original context nor do we know the speciffic people. Second, it sounds to me like you did the majority of work and, quite frankly, we cannot tell if what AI provided wasn't actually dragging down the overall result, but you didn't see it out of lack of belief in yourself. Quite frankly, I think you're selling yourself short in favor of glorifying worthless toy.
I've got to say if you think all current AI art looks "absolutely terrible", your standards are either exceedingly high, or you're going in with such an expectation it ll be you're projecting on it. Art AIs can produce too wide a variety of work for that to otherwise make sense. The most you can say is that versions that don't have style prompts have a tendency to look very similar, probably from having been trained on mostly the same material.
And I've got to say that if you think AIa art is actually good, you know nothing about art or art creation process or have pre-existing bias to see AI as good or for some reason want to cheer at the idea of artists losing jobs to over-rendered, uncanny-valley inducing, same-looking crap, that you flat out lie to convince people it's a good thing, or all three of them to different degrees. I cannot take seriously anyone who claims AI vomit of pixels looks good.
 

MNblockhead

A Title Much Cooler Than Anything on the Old Site
WotC isn't going to develop an LLM, it is going to license one to incorporate into the Beyond experience.
And this is one use I might welcome, depending on how it was implemented. I already use the beta Adobe Acrobate AI assistant with my TTRPG PDFs. It will summarize the PDF and let you type in natural language questions. I find it very useful when the rules for something are spread about in different sections of the book (it will summarize what the overall rules are and give numbered reference links to jump to the relevant portions of the PDF). I also find it useful when I forget the specific term of art used in the rule set. The AI can often determine the gist of what I'm asking and find the relevant rule even if I don't have the terminology right. That avoids the frustration of keyword searching when you don't know what keywords to use.

I still haven't put it through its paces. Once thing I want to do is ask the Adobe AI to give me an example of how a rule would be used in play or how an encounter or situation could be resolved by the rules. I think I'll go test this out now...
 

MNblockhead

A Title Much Cooler Than Anything on the Old Site
And this is one use I might welcome, depending on how it was implemented. I already use the beta Adobe Acrobate AI assistant with my TTRPG PDFs. It will summarize the PDF and let you type in natural language questions. I find it very useful when the rules for something are spread about in different sections of the book (it will summarize what the overall rules are and give numbered reference links to jump to the relevant portions of the PDF). I also find it useful when I forget the specific term of art used in the rule set. The AI can often determine the gist of what I'm asking and find the relevant rule even if I don't have the terminology right. That avoids the frustration of keyword searching when you don't know what keywords to use.

I still haven't put it through its paces. Once thing I want to do is ask the Adobe AI to give me an example of how a rule would be used in play or how an encounter or situation could be resolved by the rules. I think I'll go test this out now...

BOOO!

1712634568589.png
 



Remove ads

Remove ads

Top