Larian Studios Forums Baldur's Gate III Baldur's Gate III - Suggestions & Feedback Neural network / AI for NPC

Bg3 can be the first AAA game to use AI for NPC reactions.
This could include react to
- what player wears
- what skills player uses
- what condition the party member is (dirty face, wounded, ill)
- class reactions (npc in tavern notices a bard , wounded npc calls for help if a cleric is nearby)
- race reactions
It can make bg3 even more immersive.

somithing like this, but npc's in underground must be trained to know about caves and not too much about upper world events

I’m sure that AI presents some great opportunities (as well as risks) for NPC dialogue in future games, but I don’t see Larian tagging it on at this late stage of BG3 development. I’d be surprised if they divert from the traditionally written/voiced dialogue that we’ve seen in EA.

Just like at my job as an system engineer, AI slowly becomes a standard.It's not there yet.
Same will be for Video Games.

Buckle up, it will ravage our jobs but the perks it has will develop the gaming industry into an absolute beast in the coming years.

Yeah, the only sensible application I could see being made in a short timeframe is something like text-to-speech for minor NPC barks and/or vocal filters used to change voice/vocal timbre for the player character starting from the already existing pre-recorded lines.
And even that seems unlikely as an abrupt introduction at this point in time.

I think the first major use of the recent strides of AI technology will reach RPGs not through "handcrafted" games like BG3, but through procedurally generated RPGs. The type of AI Ardanew is talking about can give unique and realistic voices, responses, behaviors, and context to an environment. Fully expect this kind of tech to be a priority of procedural generation going forward as it improves.

Soon enough all video game stories and dialogues and reactions will be handled by ai. Much like every other aspect of life in the years to come.

I for one will cherish this short amount of time we have left.

I think AI in games will be great. It might be a bit hokey on the first few implementations but once it gets a little experience it will be amazing. The strength of AI is that it can continually improve, and it will do so quickly.

Very unlikely.

People already had a lot of fun with AI Dungeon 3 years ago, arguably more than a lot of generic video games which deliver the most basic plot imaginable. I am certain than AI gaming will at least become its own popular genre in the near future, and probably permeate the rest of the market later.

Also, modern chat AI can simulate a DM very well, they even know about the rules of most tabletop RPGs. I recommand to try it, it can lead to some very creative outputs.

AI's already pass the Turing Test, and given their new expanded use, their training will accelerate. I would be surprised if he don't have Turing Test qualified AI's on our phones (via Siri, Google, Alexa etc) within 5 years. At that point it becomes financially beneficial to do it in games because it will be easier than writing and recording voice scripts.

Passing the Turing Test doesn’t seem to have an obvious connection with the ability to create a good game. I could (hopefully!) do the former, but not the latter.

That said, I do agree it feels as though we’re now teetering on a tipping point for AI and are going to see it crop up in more and more arenas, including more games. It will be interesting, and probably a bit scary, to see what it does well and what it does badly … at least to begin with, before it gets better at that too!

I agree that procedurally generated RPGs seem like an excellent use case to start with, whereas the more handcrafted, story heavy RPGs I like are probably going to be further down the road. Though I suppose you can argue that even few of those are original enough that they couldn’t be generated by an AI that had been provided with lots of info about other such games!

From the, admittedly little, I understand about AI, there’s still going to be a significant role for humans in the immediate future in training AIs, confirming what works and what doesn’t, and preventing them going rogue in some scenarios. But who knows how long that’s going to be the case?

AI will help with the creation of all RPGs, and revolutionize in the same way that the internet has. There are plenty of repetitive tasks that an AI could handle even in "handcrafted" RPGs (which are of course never really handcrafted because they use a variety of automation tools). But the real boon to RPGs will be giving NPCs AI personalities that allows them to roam and react to the actions of the PC and the changes to the world. This can even be entirely voiced because AI voice generation will also continue to improve.

Can't wait for a full 3D game that is 100% made by an AI.

We know AI can code simple programs. We know it can create art. We know it can create audio. Eventually, an AI will be able to code its own game engine, make its own voice acting, and build its own levels. It's going to be terrible the first few times, but I can't wait to see what it's like.

Correction: it can't 'create' art. It mashes together whatever it can find on the internet. Same for voice acting. AI can't made anything without a solid human base, so it will never be able to do a 100% original anything.

That's false.
It's a common misconception, but definitely NOT how stable diffusion works.

I'm no software engineer, so my knowleadge is limited, that I can agree with. I don't know, all my alarms just go off when someone tells me AI can do 'art'.

To my knowleadge AI, at this stage, doesn't know what a cat is when you ask for a drawing of a cat. It can't see or have any associated feelings of a cat. The AI knows what it has been 'fed' by people or just pointed by them where to look for what a cat is. So all the script knows is pictures from a chosen source labeled as 'cat'. Uses actual works of people to do it's own job.

The problem with the claim that "AI just put together existing images" is that is... well, false.
When you are asking Midjourney or Dall-E to draw a cat they aren't browsing their catalog and picking a favorite cat among the millions of existing ones, they are using what they "learned" about the patterns that define the shape and "texture" of a cat observing millions of images and coming up with a new composition of a cat that never existed until minutes before.

Which is also what makes most of the complaints about "violating copyright" dodgy at best.
It will be hard to prove in a courtroom when the "copy" of your work is factually different from the original, given that you can't register/copyright a style.
If these AI companies are going to lose the legal battle, it will be all about the jury questioning their right to access the "training data" and not about the results being something "stolen".

There's also the other side of the coin, which is that even conceding that AI companies will have to compensate artists, it's hard to imagine the "quantification" of that compensation being anything but pitiful.
If your "stolen work" amounts to nothing more than one part in a billion of a dataset used for training the software, what sort of compensation per use should you really expect?
And when, exactly? Every time an image is elaborated? Used? Published?
And paid by whom? The company? The user who prompted the image? The site that hosts it?
Will we start taxing/putting down fan-artists and imitators too? They are doing essentially the same thing without passing through a software, after all.
Will we remove images from search engines, too?

I also find the copyright violation claims of AI to be very misguided - basically a sign that lawyers don't understand what AI is doing. Sure, the AI uses images to learn how to make its own images, but this is in a very similar way to how people learn. One would never question the creativity of a human artist, or claim they violated copyright simply because they looked at other art for inspiration and guidance.