Tobold's Blog
Wednesday, August 07, 2024
 
More AI in games!

Some of my favorite D&D adventures I played more than once, with different groups. And in spite of the adventure being the same, and me as a DM being the same, a different group always ended up playing through the adventure differently. Sometimes markedly so, for example when playing Ravenloft with one group that leaned into the horror aspects, and another group that was more power-gamer oriented. In computer RPGs, I find the possibilities for playing the game again in a different way quite limited. For example I tried Baldur's Gate 3 as the Dark Urge, playing evil and trying to recruit Minthara, but all that still resulted in a game that had only minor differences from the previous "lawful good" run.

Meanwhile in the real world, this week's "correction" at the world's stock markets hit tech companies especially hard. Investors are beginning to realize that AI isn't going to result in billions of profit anytime soon. One of the problems with AI is that it turned out to be unable to replace humans completely in any environment in which the quality of the output matters, for example when writing a legal brief or a medical diagnosis. AI can help produce a first draft, but its tendency to "hallucinate" means that a human needs to proofread and edit the result, to ensure that it doesn't contain any dangerous bullshit. If only there was an environment in which there were no serious consequences to anything that AI produced, which could serve as a test laboratory for future AI development. Also known as "games".

If ChatGPT suggests putting glue on pizza to make the cheese stick, it is a problem, because somebody might follow that advice and then sue the company after getting sick from that pizza. An AI Dungeon Master in a game world might also sometimes say something stupid, but honestly, that also happens to human DMs improvising on the spot. And within the environment of a game, it doesn't really matter. The gain of being able to replay a computer role-playing game and have really different things happen would be far superior to the risk of some of the invented stories not being coherent or logical. Generative AI is perfectly able to tell stories, and would be great to tell different stories to different players, based on slightly different player inputs.

The only problem with that proposal is that it would require a game company willing to try something daring and new. And that isn't where computer game companies of the world are right now. The most creativity we get from game developers these days is them trying to combine feature A which worked so well in game X with feature B which worked so well in game Y. Thus surprise hits like Palworld, being "Valheim with Pokemon".

I do think that generative AI could be the next big step up from the ubiquitous open world games. There are far too many open worlds in which nothing interesting ever happens, and non-essential NPCs repeat the same lines of text over and over. The "open" in open world currently only means that you have some freedom to do the scripted content in different orders, usually with some "gates" preventing you from messing up that order too much. It wouldn't take that much technology to make quest NPCs give different quests every time, even if that was limited to the different types of monsters available in that zone. Our game worlds are so static right now, that even minor technological innovation with generative AI would feel like a revolution. Viva la revolution!

Comments:
You're right. We've had to this to a (very) primitive extent ever since "AI Dungeon" launched quite a few years back now. It just needs a few orders of magnitude of exponential growth in refinement. You can have some pretty impressive text-only adventures using the deep features of an app like "Silly Tavern" and whichever latest LLM tickles your fancy. But hooking it all up into modern graphics is a ways off yet. The early attempts will probably be cheap retro-graphics roguelike and adventures games where the AI starts to replace procedural generation at all stages. Something like the free-yet-great Ultima-like RPG Moonring that I just started playing now could begin to be enhanced that way.
 
AI has been a focus in games for a long time. To me it's just now the technology has gotten to a point where it truly can be transformative to games in a positive way. Not AI mimicking voice actors. AI pathtracing, NPCs, combat ability (hopefully not too good), etc. Those are the areas that they've always been working on a sort of AI so to me those are the logical areas to target with better tools.

I'm sure that we'll see it, it'll just take time for developers to learn the tools and integrate them into their abilities. You mentioned people realizing AI isn't going to lead to immediate profit increases - that's a big problem with people is not seeing the long term view. Transformation doesn't have to happen instantly, but it's going to happen.
 
Post a Comment

<< Home
Newer›  ‹Older

  Powered by Blogger   Free Page Rank Tool