How do you feel about the future of AI in games?

I don't really understand the question. Are you talking about AI asset generation tools? AI within games? Both? It kind of seems like both but in game AI is a totally different issue than the sort of "AI" used for asset generation that we're seeing so much attention on. NPC AI has a much longer way to go before we see a major impact, in part because of the processing limitations of games.

I'm going to use AI the way you seem to be while noting that it's not really "AI" as we see in sci fi or popular culture. These are sophisticated algorithms but have no self-awareness or even abstract thinking.

In terms of asset generation I'm sure it will become a more and more used tool over time, kind of like speed tree but for more and more things. AI tools can do a passable job of creating textures and objects and variations and such. Like Speedtree but for many more things. The AI will generate it and humans will clean it up and iron out the kinks.

But in terms of other stuff while I can see AI being useful for programming tasks (like a super spellcheck/word suggest, which is what it actually is in many ways) the more abstract the task the worse it performs. It's great at making a face that looks like other faces, and probably some animation stuff, but not good at knowing what's true or false, let alone what's fun or good level design.

And of course there will be lots of copyright and labor issues with any "AI" use for assets depending on the training set.

The current "AI" (it's not AI, just larger scale versions of machine learning content generation stuff that's been around for years now) fad is being pushed by the exact same people that claimed 5-8 years ago that blockchain would change the world overnight. Like that situation, the current fad will die out after the c-suites that got suckered into believing the hype realize that the thing they were sold is much more limited in usefulness than the thing they were promised. After the gold rush has ended, actual reasonable uses for the technology will start to show up. We likely will see a lot of useful content generation technologies appear for games but it won't be some all-encompassing thing that can create complex things like entire levels with coherent storylines all at once. Most of these tools are good for very specific things but that's it.

Also, the aforementioned legal disputes that are inevitably going to be showing up because of this stuff in the next while should be interesting. Many of these current tools rely on data scraped from the internet, much of which they didn't have explicit permission to use. The classic example was an image generator that started spewing out images with a messed up version of the Getty Images watermark on it. Getty hadn't approved their content to be used as training data. Of course, now Getty themselves are getting into the image generator game using their own images as the training data so that commercial enterprises can generate images without being worried about legal problems. There are similar issues cropping up in code generation tools that scraped code off GitHub or other places for training data. If they generate code that is largely from a chunk of code that is under a certain license, that generated code is probably not allowed to be used in some instances. It's a mess.

It's going to be used as a PR bullet point for the next little while. Which is kind of like advertising a restaurant by bragging about which brand pots and pans they use.

Overall it's a tool in the toolbox; people that are good at making good games will make good games with it, others won't.

It peaked at F.E.A.R

It's only worth it if we get a Facade 2.

I think this video is worth watching

Basically, this guy plays a demo of a game where all of the NPCs are driven by AI's and they respond to things he says to them.

How do I feel about this? Well, if every NPC in Starfield had some sort of massive LLM hanging around that they could draw character traits from, it sure would make that world a little more lively and interesting. Generally speaking, we've gotten pretty good at procedurally generating worlds and somewhat OK at generating interesting things to fill them with, if we could procedurally generate cities and fill them with characters we could make something pretty interesting. Of course it won't be a replacement for hand-crafted stories and characters, procedurally generated worlds have already proven that they cannot be as good as hand-crafted ones, but it could still be a useful tool.

Another game worth considering is an early access game called Shadows of Doubt, a 3D world that generates crime scenes for you to investigate as a private investigator. I don't know exactly how AI driven it is, but it does randomly generate individual characters that all have roles to play in its world.

https://store.steampowered.com/app/986130/Shadows_of_Doubt/

To clarify, when I say "AI" I essentially mean the use of LLMs or other deep-learning models to be used in real-time. Not counting asset or code generation during dev time. That's what's covered in the first poll option, but not what I'm interested in.

The debate around what does or doesn't count as "AI" is a little meaningless. AI as a field of academic research has existed for several decades under that name. The name doesn't imply that it's as smart as a human or as Data.

@wollywoo: Modern machines do not have the spare computing power to do significant realtime asset generation like that. Maybe for something like building a good "random" world map in Civilization, or making canned NPC responses a little more varied and interesting, but not much beyond that outside weird experimental games. Cloud integration could help but then you run into Internet bandwidth issues.

If you're talking realtime on local machines you're talking pretty minor stuff. Realtime on a server based game could maybe be more interesting but I think the game design would have to catch up to that. You could see stuff like a battle royale game where each session takes place on a brand new AI generated map (or a Trackmania type racing game that did something similar) and that could be neat, but current game design makes learning the maps a key component so you'd have to compensate.

We will be seeing AI assisted level generation relatively soon, but it will just be an incremental improvement over current rule defined level generation like you already get in games like Civ.

sussing out authorial intent and finding the empathy/humanity is like, 90% of the motivation for why i play games. it's also part of the reason why i tend to bounce off games without narrative drive- i need something more than moment-to-moment gameplay to keep me engaged.

so in that context, it could be useful to have LLM-driven NPCs as set dressing- but i already make suspension-of-disbelief allowances for games that i'm invested in. Baldur's Gate 3's titular city already puts me in a time and place- having more unique dialogue for background characters has diminishing returns, i think. i want a story more than a sim, and i think a character disconnected from performance or intentional writing will feel very bland.

another component of game-making i find myself paying more mind to these days is craft. whether it's costume design, character writing, interface art or music- i'm finding i'm considering these things significantly more than i used to. so if AI represents an opportunity to skip a lot of the mundane work so the craftspeople can focus on the more idiosyncratic and artistic aspects, i'm all for it.

but i have a feeling we're just gonna get a lot of stuff that optimizes to boring.

The problem is that it isn't a meaningless discussion because without defining what you mean by "AI" clearly the discussion goes off the rails quickly because what academics consider AI to be is quite different than what tech marketers consider AI to be and further still what writers have labeled as AI in the past. The two examples you gave were quite different and the second one sounded like you meant content generation as a dev tool, which is why most of us responded thinking you were talking about that.

For the question itself, real-time generation of NPC quest dialog or whatever seems like a weird use of LLM text generation because it would require the developer to put a bunch of guard rails up to stop the NPC from going off on a meaningless tangents or would require them to only talk about quest-specific things. It would be a huge pain to train an NPC's LLM to have a certain personality and talk about specific things only. You'd have to do that for every NPC or they'd all talk to you in the same way. At a certain point it would be much less work to just write the dialog instead ahead of time since you'd be able to easily control the scope of the conversation and give each NPC a degree of personality.

Maybe it could be useful for generating dialog for random NPCs that are just in the world so they respond to what's going on? Using LLM-esque tech to generate entire dungeons or quests in real time would suck because it would likely be limited in what could actually be done and none of it would be creatively interesting (remember, LLMs are useless at anything that requires human creativity or remembering of context beyond a surface level. All content LLMs create is inherently derivative. They might "1000 monkeys using typewriters" their way into something interesting once in a while but that's it. Leaving it up to chance whether the thing they create for the player is interesting or not would be awful). You'd end up with a bunch of lego'd together dungeon chunks or generic quest details, which is not really something you need LLM-style tech for since it's been done for decades using other much less computationally expensive methods that would work just as well if not better since the developer would have far more control over potential outputs (roguelikes/lites are already this, for example. You don't need computationally expensive AI methods to make good games that rely on content generation)

@ben_h: I think I mostly agree with you but I do think that LLMs could have uses in improving content generation tools that already exist. For example, imagine having a group of playtesters play a bunch of randomly generated levels in a roguelite and rate them. You could then use them to train an LLM-like program to be able to generate better levels than rules or randomness. Would this vastly improve the quality of the game? Probably not, but you might be able to increase the fun somewhat and reduce the frustration of bad RNG runs.

You could also have NPCs who you are passing by or just talking to randomly provide a little more variety and flavor in their responses. I don't know how much I actually want this, but it's at least intriguing.

But I do think that games can and should incorporate these things in generating assets like faces or maps. Imagine a civ-type game where all the "random" maps are reasonably good, or a Tony Hawk game that can build you endless fun skateparks, or a character creator that can produce limitless high quality faces for you to pick from.

All of these things would be cool and seem doable with the current state of these programs. They're not revolutionary but they could be significant incremental improvements.

I ultimately went with the final choice. I believe it will be used pretty effectively in a few games, but it won't take over how current games are made anytime soon. For example, Hell Blade's gimmick was the full use of spatial sound back when it first became a thing, to have voices happening around the player while they wore a decent pair of headphones to simulate Seuna's mental issues. It's quite the experience, but was not something a lot of other video game developers jumped on. Spatial sound is used a lot more in movies and shows than in video games today. Yes, you can turn on a feature in the current gen consoles and PC to give overall spatial sound, but this is different than a developer actually using it as a key part of a video game's programming. And like HDR can produce mixed results. Kind of like what would happen if AI was used in every video game going forward. But if it does become a mainstream thing, then I see open world games being the genre where it will be most impressive. And maybe sports game where players go against the CPU and it feels like playing against another human being. I guess it can work in fighting games too, Virtual Fighter 4 proved this over a decade ago where players could take on CPU fighters whom are programmed after actual VF pros. One of the things which made that game a classic.

Yeah, I think cloud integration would be a good way to go, for the short term. Internet bandwidth is not that much of an issue since it's only passing text back and forth. There would be a bit of lag but that's not game-breaking. I'd hope in future there will be a lot of optimization that will bring down the size of the models considerably for specific use-cases - e.g. a random orc doesn't have to have all the knowledge of Wikipedia.

Fair enough, I could've been more clear. Regarding NPC dialog staying on-topic, I actually think this is not that hard, at least at surface level. With chatgpt, you can just tell it to act like a certain character and give it rules as to what it can talk about, and it will only talk about those things. If this is all you did, most likely there would be easy ways to jailbreak this and get it to do and say really weird shit, so tuning this properly would definitely be a challenge. But I think it's doable. And I more or less agree with you re: creativity - I'd suggest it'd be used more to augment the authored content with interactivity instead of replacing it. You don't need that much creativity if you're just acting-out a well-defined NPC role.

I think even though current AI's could do chat-bot interactions very well, that there will be a while before any publisher will do anything interesting.

I think the sad part is they could use AI interactions and chat functions to PUSH the concept of "ALWAYS ONLINE" games down our throats. The AI interaction will be the fancy Trojan horse through which they will end all offline functional games. So they won't add AI to make games better, just use it to crack the door open so they can make always always need to be logged in -never offline.

In terms of AI driven NPC conversation i think it's going to be used in very cool ways for very specific kinds of games, like the example given of that investigation game where it's entirely about asking questions, we've tried to do this before in text based adventure games "ask about map" "show lollipop", but with AI you can ask them what they had for breakfast and they'll come up with something and that's extremely cool, how well it integrates into the actual game like getting clues to find things within the game world is up to the developer.

The downsides, i'm not sure if you can make an interesting story relying entirely on this method, that requires likeable characters and high quality performances and if every player is getting different dialogue from text to speech robots that's going to be hard to deliver, you could make a menacing SHODAN or GLADDOS with dialogue specific to you, but the humour or memorable quotes will probably fall flat most of the time, then you're still relying on a writer to force them in and the AI voice might not be able to deliver the lines well, real human characters will be weird they might be believable one day then forget you exist the next, one day their favourite colour is red the next day it's green, even if they fix all the obvious issues you'll still occasionally get "sorry, i do not understand".

Another downside, it requires an online connection, presumably all the AI driven dialogue is generated by servers and hopefully that delay can be minimised but every single one of these games will break one day, maybe the next Elder Scrolls has normal VO but AI generated small talk trained on the actors voices, even if it doesn't break when millions of people send dialogue requests to Bethesda's server they'll shut it down some day, or when they remaster it they disconnect the old version... or they have to keep taking it off sale because we find a way to make it say something racist.

I think we'll see a lot more AI voice trained performances before we see a large adoption of AI generated dialogue, i'm already seeing AI voices in quite a few indie games.

Lastly, in terms of AI art or AI aided coding i assume as many companies as possible will already be trying to implement that in their workflow.

Forget about AI in games. How does everyone feel about the increased permeation of AI on the GB forums? Lol

@cikame: Yeah, the voice thing is definitely an issue. It's getting better - chatgpt's new voice feature seems pretty good - but it's definitely going to be behind human voice actors. Maybe they can do a blend, where some important lines are delivered by humans and then somehow used as hints to the voice synthesizer so that it's not so obvious which are human. An investigation game with AI sounds very cool.

I've thought about the consistency problem too. You might not want every person playing the game to get completely different facts about each character. You could ameliorate this somewhat by having long pre-written bios of basic things about each character. For even more details you could augment this by pre-written AI-generated backgrounds that make up even more detail that don't make much difference to the story. But yeah, at some point it's going to start making up random stuff and it's inevitably going to break immersion when things become too wild or inconsistent.

I'm only worried if they start arguing with each other. Then we're done for.

@shindig: They never argue. They just agree with someone (real or AI) because arguing requires some understanding of the topic at hand (or the appearance of such) while agreeing is much easier.

i think it Can be a neat tool but Will overall hurt ppl instead of help due to greedy corps that only care about profit.

"It's yet another big money maker for corporations that will only hurt human content creators."

On the bright side it will free up time for these people to watch AI outdo them at what they love doing

How an AI Chatbot led a very stupid man to commit treason.

Not sure if this is region-locked but the gist of it is:

Mentally ill man falls in love with an AI chatbot. Chatbot agrees he should take a crossbow to murder Queen Elizabeth II. We can't sentence the chatbot who is ultimately a collaborator but we should have legal avenues to convict AI developers whose software accidentally leads to dumb shit like this.

To circle it back round to games, AI that encourages lootbox spending and other deceitful practices.

There's a prominent user on here that I swear is an AI bot

@storysaver: The irony of this is too much. Just a small glimpse at our AI-dominated future and it's depressing.

I'm pretty optimistic about it! The future of AI in games holds a lot of exciting possibilities. We're talking about more realistic and dynamic non-player characters (NPCs), enhanced storytelling through adaptive narratives, and even AI-generated game content. It's not just about making opponents tougher; it's about creating more immersive and personalized gaming experiences. Of course, ethical considerations and responsible development practices will be crucial as AI continues to play a bigger role in the gaming industry. What are your thoughts on it?

In this thread a bunch of "AI" bots botting in support of AI. Truly Skynet draws ever nearer.

Microsoft announced a big games AI initiative.

I am skeptical that much will come of this. I remember when Azure was the next big thing in gaming. Instead it's just...server architecture that most games use in the same way that games always used online servers.

@bigsocrates: ah, THE CLOUD you say!?