Can AI Replace Your DM?

The current hotness is anything to do with artificial intelligence, and along with some interesting experiments comes a lot of mindless hype. The question is, what can it do for us! [Jesse] provides a fun answer by asking ChatGPT to perform as a Dungeons and Dragons dungeon master.

There are many ways to approach a game of D&D, and while some take the whole thing very seriously indeed we prefer to treat it as a lightly inebriated band of intrepid heroes smacking each other and assorted monsters with imaginary swords and war hammers. Would the AI follow the nerdiest cliches to their pedantic conclusions, or would it sense that the point of a game is to have fun?

We follow the AI D&D campaign from the creation of the world, then a band of heroes. The heroes go to an inn where they are approached by a mysterious stranger with a quest to explore a cave. The story evolves to become a pretty good D&D game, that we can imagine our human DM would have a lot of fun embellishing and performing as he led us along it.

Maybe the key to this lies in that last sentence. While this is an acceptable storyline that might appear in a D&D book, it’s just not enough to carry a game. The machine DM lacks the sparkle that a good human DM would add to the game. The names, locations, and characters are also painfully formulaic, the type of thing that teenagers find really cool when they discover D&D, but later figure out isn’t really necessary.

The AI is doing what it’s good at and spinning a D&D yarn from all the D&D yarns it knows, but since it’s never experienced a real game with a human DM, it still gives only a semblance of intelligence. Human DMs can now breathe a sigh of relief, they’re not about to be made redundant.

Having an AI with a D&D angle has been an entertaining story to write. We more often see the game in context of people making dice.

Header: DALL-E prompt “A game of D&D in progress, fantasy art oil painting”.

18 thoughts on “Can AI Replace Your DM?

  1. When I read the Title, I was wondering how Al Williams was going to replace my Digital Meters.

    Anyway if he wanted to replace them, he’d have to first pry them from my cold dead fingers.
    B^)

  2. I’ve been waiting for an AI DM for a long time now, and it does seem to be getting closer. Rules and mechanics are trivial to implement in conventional code, but what I need is an AI capable of making judgement calls.

    During my group’s last session, the halfling barbarian asked if he could climb onto the turkey hydra. There’s rules for climbing static things and rules for grappling, but I don’t know any rules for a halfling climbing a large creature, so I made him do an Agility check (DC 20) for climbing the bird and as long as he was on top of it he had to roll a Strength check (DC 15) every turn to hold on as the turkey hydra was trying to toss him off. After hacking off a few heads with his ax for a few turns, the barbarian then asked if he could shove some of his javelins (thrown weapons, technically) up the creature’s rear end. Again, I don’t know any official rule that covers this, so I let him do it as a normal melee attack but he was rolling with disadvantage.

    Maybe there’s official rules in some systems that cover all of that, but the point is that there doesn’t really need to be because there’s a human DM making judgement calls to decide if a thing is plausible, which stat would govern success, what difficulty class it should be, and most importantly is it so cool that you secretly let the player succeed regardless of the dice so they get the thrill of being awesome or so absurd that you let them fail and unleash consequences to remind them of their mortality, punish them for their hubris, and generate peril to make sure everyone thinks through their choices.

    That’s the sort of stuff an AI DM needs to do. If we could make that happen, I’d pay money for it so I can finally get to play instead of always being the DM.

    1. An AI with the flexibility to adlib in the inevitable grey areas, and allow for the collaborative story the players and DM produce in the way you describe I don’t think is even close to happening yet.

      For myself I think you are entirely correct in core ideals. Though with a bit of variation depending on the group and exact issue at hand. Which is another thing I don’t think AI will be doing any time soon – making that judgement that this group has 3 rules lawyers who love to argue and derail the session for everyone else, or this group is all a bit passively following the video game logic path so this one idea out of that mould is worth allowing even if it is a bit stupid as encouragement to enjoy the scope of D&D type games.

      For instance I’d not tell them what the DC would be, and perhaps not even which type of check I’d make it be in advance every time – so they can’t know it’s definitely always going to be a check against their best skills/stats with good odds to succeed. Which stops the good number crunching min-maxer from practically KNOWING the outcome – keeps the risk reward factor an unknown.

      But equally I’ll often explain my reasoning as to what ‘rule’ we are going to invent for this situation, doing so both helps them clear up the HOW they want to try something – making sure we are all on exactly the same page and gives them a chance to refine the idea – perhaps in your climb the creature check its bare hand climbing on the spines it happens to have (so perhaps it is a dex or even con based check to make sure the poison on the spines doesn’t numb their hands), or using ice axe or their dagger etc as self made handholds (in which case perhaps its roll to attack twice per action with hits letting you get another handhold towards you goal in, but still with the strength check to resist being thrown). Those ideas can change how long it should take, if it hurts the creature as you climb and so the ‘rule’ used. But also gives them the chance to change their minds a bit if they want to – rule of cool is good fun, but sometimes it has to be rule of stupid Darwin award seekers as well if the game is to have challenge.

    1. Decided to try it as a 1:1. Working pretty well so far, though the AI has a tendency to try and narrate events. It’s profusely apologetic when I correct it though. Also, does a terrible job of staying in character.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.