D&D General The DM Shortage

[Paul Farquhar]

Open ended questions? I don't think so. A pre-determined list of questions the programmers anticipated, sure.

Text based adventures like Zork have existed for decades, but were always limited in that they can't do anything the programmers didn't explicitly code for.

The advantage of an AI is that they can (potentially) handle scenarios that haven't been explicitly coded for. Which makes them more like a real DM, as opposed to a Choose Your Own Adventure book. But they still need a large data set in order to compute a probable correct response.

Otherwise, there's no reason to use an AI. There are plenty of CRPGs that can handle a broad number of (programmer anticipated) player choices. But they're not actually the same as playing with a real DM, who can actually improvise when a player attempts something unanticipated.

An AI could create it's rules as it went along? I don't think that would be terribly demanding for current AI. But would you want it to store it's rules so that if a similar situation arose down the line the AI resolved it the same way? In which case you would rapidly reach a situation where the whole game was AI-generated, and no longer D&D at all. Which might be a better game. And if you used real world data to prep the AI, there are going to be an awful lot of dead adventurers failing to leap across 10 foot pits.

An AI would probably take a different approach to a human DM. It, might, for example, look at your character portrait, and decide how tall they where, how fit, and how heavy they looked, before deciding what the % chance of jumping the pit was. Would you think that was fair if a human did it? Would it be more acceptable for an AI?

 

[Celebrim]

On this question, I point to the importance back in the 1980s of games like Star Fleet Battles and Auto Duel and BattleTech and how much effort and love were poured into those games. Yet hardly any attention is given to those games anymore because the experience they were trying to create is one which is more easily created on a computer. The modern equivalent is something like Star Citizen.

Likewise, a large part of the OSR experience is I think available now on a computer. If all you want from a game is kicking down doors, killing things, and taking their stuff, we can create very compelling computer experiences of that already. The sort of games I was playing at age 12 or 13 have very little interest to me anymore because I can get that combat cycle experience better from a video game. Someone sooner or later is going to come up with a really compelling Rouge-Like that replicates the old mega-dungeon experience really really well, with all the little gritty bits of light sources and traps and searching stuff and improvising solutions out of rope and iron spikes. If Diablo's team had moved more in that direction than toward the arcade like action that has probably more market appeal, we'd already be there.

But we're still 80 years or more from having a DM that can run NPCs and generate plots and backgrounds and stuff like that come close to emulating a human DM. I'm a huge fan of the first Mass Effect game as a story telling masterpiece and I'm a huge fan of Grim Fandango as a story telling master-peice, but neither captures the RPG experience because you can't actually choose anything, and the engine can't actually adapt to your choices. It's all illusionism, and not even as good of illusionism as you can get from skilled illusionist DMs.

The role-playing part of a role-playing game is going to be the hardest to recreate.

 

[Fanaelialae]

An AI could create it's rules as it went along? I don't think that would be terribly demanding for current AI. But would you want it to store it's rules so that if a similar situation arose down the line the AI resolved it the same way? In which case you would rapidly reach a situation where the whole game was AI-generated, and no longer D&D at all. Which might be a better game. And if you used real world data to prep the AI, there are going to be an awful lot of dead adventurers failing to leap across 10 foot pits.

No, an AI can't create it's rules as it goes. Or, at least, those rules aren't likely to be based on anything functional without a good (large) starting set.

Unlike what you see in fiction, current AIs aren't actually capable of though. They just use their data set to calculate the most probable correct response. But the problem with that is that you need a data set that it can calculate that from. For example, let's say that D&D Reddit threads were used for the AI, but someone trolled (pun intended) and got a bunch of their friends to say that knowing that trolls are vulnerable to acid and fire is DC 500. The AI would most likely tell you to make a DC 500 knowledge check to find out what a troll's vulnerabilities are, because the AI doesn't have the "sanity checks" a real DM has to recognize that as an absurd result.

 

[Paul Farquhar]

But we're still 80 years or more from having a DM that can run NPCs and generate plots and backgrounds and stuff like that come close to emulating a human DM.

I think we are very very much closer than that.

 

[Paul Farquhar]

No, an AI can't create it's rules as it goes.

Sure it can. It just compare the situation the player decides to a vast databank (I.e. the internet). Can I jump X feet? Tons of data on people jumping. Can I bend an iron bar? Plenty of data on human muscle strength and iron. That's how facial recognition works. The results might be questionable, but AI has been doing it for years. And humans also produce questionable results, the AI is just imitating them. A computer DM could produce much more realistic judgments, for better or worse. You could limit it's data search to fantasy fiction, of course.

 

[Greg Benage]

Makes no difference. There are rules for NPC reactions, and everything else is either thief skill %, auto-success or auto-fail. It's all easy to code, and has been, right from the gold box games.

That's...not a good description of classic D&D. You at least need something for, "Player describes shenanigans, DM makes a judgment and rolls whatever die they think best."

 

[Celebrim]

I think we are very very much closer than that.

Speaking as a software engineer, no we are not. We've made a ton of progress in the 50 years of my lifetime, but the current technology we have has more limitations than it might first appear.

 

[Paul Farquhar]

That's...not a good description of classic D&D. You at least need something for, "Player describes shenanigans, DM makes a judgment and rolls whatever die they think best."

True. Because the human DM doesn't really know anything about how difficult those things really are. It's even easier to get a computer to judge from ignorance. It's called a random number generator. Hide it behind a DM screen and it would look just like the human DM.

 

[Fanaelialae]

Sure it can. It just compare the situation the player decides to a vast databank (I.e. the internet). Can I jump X feet? Tons of data on people jumping. Can I bend an iron bar? Plenty of data on human muscle strength and iron. That's how facial recognition works. The results might be questionable, but AI has been doing it for years. And humans also produce questionable results, the AI is just imitating them. A computer DM could produce much more realistic judgments, for better or worse. You could limit it's data search to fantasy fiction, of course.

Again, something like jumping or bending a bar are relatively simple and don't require an AI to begin with.

The parts that are hard (even with an AI) are the more open ended things. Like do I know if trolls have vulnerabilities? Do I know what trolls eat? And until an AI can do that with high accuracy, they won't really be able to do a GMs job.

 

[Celebrim]

Sure it can. It just compare the situation the player decides to a vast databank (I.e. the internet). Can I jump X feet? Tons of data on people jumping. Can I bend an iron bar? Plenty of data on human muscle strength and iron. That's how facial recognition works. The results might be questionable, but AI has been doing it for years. And humans also produce questionable results, the AI is just imitating them. A computer DM could produce much more realistic judgments, for better or worse. You could limit it's data search to fantasy fiction, of course.

No, we can't do that. We can't get a computer to work out what is pertinent in a situation. Computers can't really reason. What you could do is the reverse - work out what problems you wanted to have and how difficult you wanted them to be, and then give the computer guidelines on how to fake those situations well enough they'd cause humans in the simulation to suspend disbelief.

We have tons of people working on the task of teaching computers how to reason spatially. We are getting there. The basic tools for doing what you are describing are being worked on right now. But what you are describing is the distance between realizing you have the pieces to make a cell phone, and everyone having a cell phone. It will take decades. And it's not clear that applying that level of processing to a 3D first person shooter (a simpler concept than a full RPG) is necessary to suspend disbelief and make a good game, or that the cost of computing will continue to decline to the point that level of simulation can be done on a desktop.

Think of a rag-doll based physics engine and how often those engines produce nonsense answers that we can tell as humans that live in the real world are wrong. Think about for example how in something like Grand Theft Auto the physics engine can under some circumstances launch objects into space. That's the level we are at right now, and even most of that is faked and simplified.