From my limited DMing experience, much of the challenges is handling when players do something you haven't anticipated. For most encounters (video games do this too) I think about
1. The stealth approach (why yes there is a secret entrance underneath the castle). 2. The direct approach. Like just kicking down the door and heading in. 3. Negotiation: Is there a way to strike a deal with the bad guy?
Beyond that I fall back on trying to provide a realistic response from the game world, but sometimes the players are creative enough that you have to redo the whole thing on the fly.
I like your three-part planning a lot! That’s a great framework.
The trick is to figure out how to put appropriate guardrails and narrative structure on an otherwise too open-ended LLM. If you just leave GPT to its own devices, it allows everything and goes everywhere.
This demo actually has a decision tree of sorts underpinning it, which means the world could feel less ‘GPT just made this up’ and more ‘someone thought this through’ with a bit of polish. I need to do more to nudge the user back towards the narrative when they go too far off on their own, though. And admittedly, it’s a very early demo so there isn’t much of a story locked-in yet.
Games that try to tell an interesting story, the real arty stuff, all that might not be able to use it well.
But Minecraft with good NPC’s, all the open world stuff, that seems like it could be really fun and cool.
How much time did people spend actually engaging with the deep and thoughtful narrative in Skyrim? And how much did they spend just enjoying the world? The latter could be really enhanced with an AI DM, in the medium-term, I bet.
Ah, but the world of Skyrim is itself the deep and thoughtful narrative. It has been painstakingly crafted to give the characters, locations, and history meaning and interest. Only the barest minimum amount of freedom is then sprinkled on top in the form of player agency. The consistency is what makes it so interesting!
For an AI generated story, you'll need to find a way to ensure everything stays coherent. If I convince a random farmer that it's in his best interest to try to kill the king or marry the mayor's daughter, but come back the next week to find him contentedly plowing his field again, my immersion is broken and my day is ruined! I suspect that giving this level of freedom to NPCs will make crafting a stable (playable) world difficult to impossible.
Stumbling into the Underdark by accident is an amazing feeling, because it's obviously an important location from a quest that you don't have yet! Exposition and intrigue, how fun!
I can't imagine a way to do this nicely with generative AI. If I thoroughly explore the king's palace before sending in my farmer-come-assassin friend, how will it know that the gallows are to be an important location later?
Maybe Skyrim was a bad example because it is too designed. But like, Minecraft barely has a plot, and it was really popular. Maybe we can add, like, a couple good characters to open world survival crafting games. We don’t have to skip straight to simulating whole cities.
On a side note, I love this whole thread and I really appreciate both of your thoughts!
If we assume something in the middle:
While we might not be able to predict that you’ll encourage the farmer to (try to) overthrow the king, we might predict that you’ll try something of that sort. Maybe you enlist a guard for an inside job, or perhaps you do it yourself! But if we want to plan for it, we can plant the seeds, and then nudge the GPT to weave the threads of our story together when it happens.
Interestingly, because GPT requires so little prompting for relatively intricate story lines, (yes, really), you can probably add an ungodly number of these semi-scripted moments. I think they would be absolutely magical.
I think all of that would add up to a world that feels alive, with deep world building and many ‘surprising but inevitable’ moments – and the potential for some great emergent storytelling along the way.
I’m very excited for it!
I mean, we have lots of fun in much dumber worlds, where the peak of NPC design is bots that walk from their jobs to their homes at certain times of day. And maybe something slightly more advanced will fall into an uncanny valley (we probably don’t have the technology to simulate an AI farmer following your advice to take part in a grand conspiracy to kill the king).
Forget AI. Just think about simple things. Like plot critical NPCs. Most games either don’t let you attack them or cause you to lose if you attack them. That’s good. It just makes sense. Yeah it’s great that some RPGs proudly demonstrate that the plot can go forward with every character dead but that’s absolute shit for most plots. The constraints are very helpful.
I don’t really enjoy video-game narratives in general anyway. That sort of content is there for people who want it, but I don’t (I know where the books and movies are if I want a well written plot). Skyrim was probably a bad example because it is a game for people who like that kind of stuff. But I still managed to have fun running around, killing bandits, doing dungeons, and avoiding the plot.
I’d rather have something like Mount and Blade with janky LLM run AI. Dwarf Fortress, etc.
Lots of open-world games have this sort of setup where plots seem to 2-3 node long semi-random graphs. Maybe an LLM can connect those nodes haphazardly and produce filler text to justify the connections.
Dwarf Fortress manages to build some really interesting little stories based on a good simulation reacting to pawns with fairly limited actions and internal state (they are wildly deep for procedurally generated video game characters but obviously nothing compared to an actual well-written book character or something like that). Rimworld has even more simplistic pawns and a worse simulation (I’ve even noticed pawns come in with essentially incompatible backstories), but it isn’t a disaster.
The stakes aren’t all that high, people are just really forgiving to the narrator in some genres.
Folks really just aren’t going to engage in it
FWIW, after bouncing the idea back and forth a bit, I’m more hopeful than ever, so even though I wasn’t able to convince you, I hope you enjoyed it as well!
If nothing else, the potential is really interesting. Who knows what will happen. Maybe somebody will come up with an LLM enhanced version of the director AI for Left 4 Dead or something.
I was on this train for a bit but after reading some posts from game developers as to why this is probably a terrible idea to the point of unfeasibility - game design (as I understand it) deals with a finite set of outcomes in a controlled environment. "Integrating" AI with games would involve shattering this constraint in a way that would make games inherently unstable/untestable - at least if I am understanding the argument correctly.