AI researcher here. As I see it, LLMs (large language models) are still at a state too far from making them run believable NPCs, the requirement for it to be connected to an external API even to be able to run decently would make it too problematic to be used in a game in a scalable way. So far, mods can get away with it because nobody will really be mad if the LLM gets jail-broken from. Fine-tuning (training for specific task given a smaller dataset) can get you only so far, game studios don't have currently the expertise for building this kind of datasets. As I see it, I think there are better uses for AI to be considered here: 1. Random maps (or even missions) created by some kind of GenAI (diffusion models probably) 2. Game bots, augmenting them here and there with AI-driven skills. Full self-training style AlphaStar would be probably out of reach for any game studio, maybe outside those who are owned by Microsoft 3. Utilizing low-level generated art assets, the mind-numbing stuff (not character design, etc.)

there's a gulf between 3 visions: 1, game as an arcade/platform for competitions, 2, game as a narrative, 3, game as a simulation of a world or region of course, these intermingle, but i think most people prefer 1 or 2. personally, i was always most interested in 3. i think for 3, all kinds of procedural generation and AI will become extremely useful. they are basically the tools that can unlock virtual reality. however, it seems that 3 is unpopular these days. it was at the peak of its popularity around early 2000s, when MMOs were big. i worry that the vision of creating a single unified virtual world where people can experience limitless things together is disappearing. one symptom of that is the negative reaction most ppl have to the idea of AI NPCs, wanting everything to be static and premade.

@angamaitesangahyando685

3 ай бұрын

Dude, this is exactly my own view when I pondered this question years ago! Plot vs competition (PvE or PvP) vs "the atmosphere", simulation. This is why I find games to be so different, it's just like with movies - compare a WW2 propaganda newsreel with a JAV corn film and with a BBC documentary - all three are technically motion pictures. - Adûnâi

I am a computer scientist in the field of machine learning and I really do not like the term "artificial intelligence" (AI). One should think about these tools as "artificial communication partners". They do not understand but are able to take part in communication and we (humans) do the "understanding" part. Note that I am critical about this technology. There is no intelligence behind, it is all statistics. They produce the most statistically likely output. However, it is not correct to say language models (which they use here) draw from a database. They generate novel sentences that look like as they were drawn from a database. In other words: If you look at a sentence generated by such tool you can not tell (statistically) that it was generated or that it is contained in the database it was "trained" on! Now, how can you control these tools. Well you can give them the information as input specific to your application. Instead of asking "are you alive" behind the scenes you actually ask "suppose you are this character with A, B, C, properties ... pretend that you are alive, answer now the following question: are you alive". Therefore, these things are difficult to control since they need so much data to "understand" language. They probably are trained on data the user does not know (and even the game-developers do not know because these systems are not open). Thus there is always the danger to "break" the language model. Developers can filter the user's input or the model's output and they can "tell" the model not to "tell" things it shouldn't. The voice thing is a completely different thing. One has to look at the two separately. AI is a huge buzzword that currently sells. For me, "generative AI" is parasitical and fascinating. It is parasitical because it sucks the creativity and intelligence of hard working people out of their brains to play an imitation game. I would go so far to say that even if you put out content that everyone can share, you did so in the past, not knowing this kind of usage. I certainly think that even many software developers regret making their code open source such that their intelligence can now be extracted. On the other hand, artificial communication is fascinating. It is the first time that machines can be really controlled via the medium which we call natural language. Communication is exciting because it is contingent, meaning you do not know what the other will answer but you also know that the other's answer depend on your question / interaction. We humans are drawn to such interactions thus I believe these language models will spark enjoyment and fascination also in games. I am not sure if this is already feasible from a cost / resource perspective, especially for small studios.

@GrubbyTalks

3 ай бұрын

Very interesting read, read it thoughtfully. Thanks for sharing!

@midnightblue3285

3 ай бұрын

The inteligent in the behind the creators of AI is ' you ' we are the programers AI could be very useful too for those ones how to use it

What I would really like to see AI used for is dynamic difficulty control. The idea is for the player to choose difficulty on the scale from "i just want the story" to "probably impossible" while AI adjusts missions/encounters based on player performance to maintain desirable level of challenge.

@GrubbyTalks

3 ай бұрын

Has to be a toggle though - that's a bit similar to level-scaling like in D4, no?

@YurGa1980

3 ай бұрын

​@@GrubbyTalks if I understand it correctly, in D4 enemies scale with your level to keep experience roughly the same, you could even say that despite being called "difficulty scaling" it is in essence the polar opposite. What I would like to see is for the game to scale difficulty based on your performance to provide a desired level of challenge. In D4 example player level alone is not suitable for this purpose as it is gonna increase no matter how good the player is. Similarly, scaling monsters to player level doesn't actually scale GAME difficulty, it just keeps it at the level that devs designed as opposed to what player wants. That being said, the player should always have an option to experience the game the way devs wanted it, so ye, it should be a toggle unless it is the core part of the game (and even then it still could be an option).

Given how fast companies have given their chat AI's restrictions, I find it super cool that game developers are looking into including AI responses in their games, so we can type out own chat messages to them, and they will respond based on their AI personality. I hope we will see this in future RPG games, when the technology has been streamlined enough for developers to slap a personality and knowledge to an NPC with a quest attached.

20 years from now we're all having our favourite streamer cloned as an AI. Imagine, we have sooo much video data of people recording themselves playing games. At this point, it's just a matter of the technique that has to be developed to be able to generate video data that is of the same distribution as real video data. But many people are working on this stuff.. rapid progress in science the public does not even know about

Is going to be cool and fun if the dev take risks and let more loose.

Strange that this is coming out so early given that AI can hallucinate many times, and is a currently being researched on topic in generative AI. Imagine if a character gives "false" information about a game mechanic or something else, and the player is not able to find that particular item/quest leading then to eventually find a guide online. We're coming back full circle!

@kysierkevin

3 ай бұрын

nothing is strange here in terms of AI's drastic rapid growth. I mean, if all of us can access internet information just by typing a few words in google. Just imagine how much an AI is able to learn thru large and curated data. Chat GPT 5 is already rumored to go out this summer of 2024 a much more powerful and knowledgeable AI tool and many of this will come. AI is the 2024's world wide web discovery

@GrubbyTalks

3 ай бұрын

Yeah AI can still make stuff up when you suggest things, it's pretty interesting how player experience is gonna be after you gaslight an AI into saying "Zerg are the good guys, stop saying they're evil!" and then they agree, and then Zerg kills all of us

@badrinath5306

3 ай бұрын

Yeah the court room drama is fun already.

@chucklessss

3 ай бұрын

Well said! Not looking forward to that 🤣🤣@@GrubbyTalks

To me this is alarming. A company who doesn't have fund to full release should not try incredibly high risk moves like this. This isn't the next Baldur's gate 3, it's not an RPG, it's an RTS... Conversation with an IA won't fix their problems. They already had so many problems, financial, Graphics, lack of innovation, monetization, honesty with their community... Now this ? They don't seem to know where to put their resources and the lead of Stormgate seems to be taking the wrong path. I wonder how much they spent on this AI.

@ppgButtercup300

3 ай бұрын

If you actually watched the video and read any of the info on it you’d have known they spent nothing. The company doing the tech was allowed to use StormGate as a platform. But you were probably too busy peddling your nonsense narrative…

@freddesk1693

3 ай бұрын

​@@ppgButtercup300 You think they are gonna give them the tech to Stormgate for free if they use it ? Use your brain, or don't and keep posting stupid comments.

@rictekmusics3143

3 ай бұрын

Yeah and imo this tech is too early to try. This is getting weird.

@rictekmusics3143

3 ай бұрын

@@ppgButtercup300 His point went right over your head uh ?

@blackcook-il9mj

3 ай бұрын

@@ppgButtercup300 Nonsense narrative ? Man Stormgate is getting weirder and weirder over time what are you saying ?

I think it's potentially a good idea if the character you're talking to is an ai in the story.

i thought they gonna have gen AI on the AI gameplay (for example the AI learn how the pro players play so they can play with similar level to pro player and beyond, stockfish for RTS)

Only if it isn't taking people's jobs, like VA's and writers But to be honest Stormgate is giving me very weird vibes, like some tech bro's "next big thing" instead of a new RTS we would want to play. I'm speaking this as a kickstarter supporter, losing my faith as things go on Edit: I wrote Patreon instead of Kickstarter lol

@Eri4Jp

3 ай бұрын

I feel they are more creative and innovative in their marketing and new tech rather than into their game.

@raglock1433

3 ай бұрын

Gotta agree. I was really excited at first. Bought into the kickstarter. Now Im more concerned than anything.

@hamsturinn

3 ай бұрын

All technologies took some people jobs. Isn't that a good thing though?

@ender4101

3 ай бұрын

​@@hamsturinnHow would that be a good thing, exacly?

@hamsturinn

3 ай бұрын

​@@ender4101 It is a good thing for the same reason other technologies were a good thing. It would free up people to do other things.

Hey Grubby. Your previous stance on AI art displayed on your stream had me stop watching your stream and videos for a while. Your stance seems to have changed a bit, at least in terms of writing and acting. I am curious if you are still pushing the use of 2D/3D AI art for game assets?

@zheta42

2 ай бұрын

In the last 2 minutes it seems like you have at least gotten a better understanding of the copyright and plagiarism issues. I am not thrilled about your ambivalence, but I am glad you are at least learning more about the issues affecting individual artists and companies when using scraped data sets. I hope you also look into the massive energy expenditure and threat to earth’s climate and resources LLMs pose.

Easy, this should be done by making the NPC an AI. Sound problem solved? I could imagine talking to a Terran Attendant about the next mission in StarCraft.

I am not that happy with the idea of replacing voice actors cause u know job loss. I am fine with them experimenting for deriving dialogue for each scenario. I am not that happy with it. Rn stirmfate is early and can experiment but I think the opening to the public is far too early for an undeveloped game. I just hope theyre careful.

I don't want to talk to a robot i want to listen to a story being told, in any game, having ai in say legacy of the void would mean I am artanis talking to karax and alarac, instead of watching the character of Artanis growing in confidence as a leader and gaining the trust of his people. Ai would make it shit and story is why people love starcraft, warcraft etc

The problem is to make genAI have a meaningful impact it needs to interact with the games systems and once you require multiple function, coherency and a larger scale well... we're nowhere close. If we're just using genAI to replace writing for mission briefings or throwaway lines it feels like, what is the point? Do I really want shit spread all over the room rather than a single cupcake that's handmade? If mixing AI and written / va'd work is tried that's going to be extremely jarring. With current functions of genAI I honestly do not really see what it can do. In the future maybe it has potential if we can get it to work but as of now I see no evidence that we can get it to be reliable. Of course, that's ignoring AIs actual use which is as specific tools, using AI for dynamic animation modification and mixing, great. AI for dynamic sfx modification, great! AI for foliage/minor terrain differences, great! Imagine we have death animations that change depending on the killing blow, blood and sfx that is based upon where they're struck and what they're struck with and then impacts in the ground where the body lands. Yes, we can do all of this stuff already "ish" but the way we fake it doesn't look great. In the future I can imagine AI npcs that actually populate worlds and do simplistic roles, which again, we can do but the amount of work skyrockets if you want to do anything more than basic schedules. Additionally from my experience with aiart, I really don't think it is "learning" it really just looks like an insanely fast, insanely good photobash. When you try out something like krea, it becomes very apparent that the same photos are being used because you can make changes very gradually and see many different generations you end up seeing the same cloud copy pasted in different ways, the same landscape photo distorted, the same rocks, the same trees, the same textures. I think if genAI wants to beat the "it's just copying" allegations, they should post their training data, actually that'd be a great legal requirement, training data needs to be public.

@GrubbyTalks

3 ай бұрын

That's a good point. It's going to take so much work to make it perform well, and most players may not even scratch the surface of its capabilities - so will it really make sense?

@GrubbyTalks

3 ай бұрын

Also agree on revealing the training data. Though I imagine there's a huge commercial and competitive advantage you're giving away, when you do post that.

Ugh. I already hate what i've seen and i've only seen GenAI dialogue/voice work, let alone genAI npc....problem i have is "less is more"; once a dev studio has essentially infinite writing and free voice acting, all the npcs get LOADS of text and LOADS of awful voice work just because they CAN and it gets annoying skipping 50 pages of dialogue for a quest.

I think it's just stupid to use AI for dialogue. It's going to be used by talentless people who cant write for S**t and then AI isn't going to make it any better. Let good writers write your damn dialogues.

This amazing on so many different ways. Yes and yes

If humanity builds an AI based on ANYTHING other than truth it will be destructive. Let the AI characters be the characters they are as realistic as can be and people can make up their own minds if this is a model character to emulate or one that displays traits that are objectionable. All of art had and has characters you can find offensive. Ok. But that's the point to depict real or realistic characters and learn. Maybe learn something that is painful to learn. Without pain there is no learning. Shield a child from ALL outside influence. It will become an inuhman mess that dies at first contact with bacteria. No bubble is eternal. Only the truth is. Go for truth.

On their own all these features seem cutting edge, but for stormgate as a whole, I’m really starting to get worried about feature creep.

It's gonna be a shit show, 100% if they implement it, no matter how much they want to sugarcoat it. The game itself doesn't feel fun at all for me, and deep down my gut feeling was telling me this isn't it and i have been following that gut feeling since 20+ years ago. Stormgate gets very repetitive after a few games. It has barebone working game at best, but the fun factor, out of the window. All i have seen from beta test on February was a game designed around e-sport scene rather than creating a fun base game, hope they get more time to get in the fun things in early access otherwise im so regretting purchasing the deluxe edition. They have to give us a game that reminds everyone why we would always come back for more, why the word ex-'blizzard' employee carry such weight, why they mattered. I just hope this is not just another empty shell that tries to reminiscence the former glory days like how b4b or callisto protocol did, always marketing how their previous masterpiece is genius but in the end failed to capture the second lightning.

I don't really like the implementation shown here. Just like you mention its not really a substitute for real charming voice work. I would want the campaign briefing to be really immersive and the ai voice here is too stiff and robotic(even if it gets better). I think it could be useful for maybe like an encyclopedia search (ie: "how can I make the flayed dragon?" "build building x then y then z. Then kill 1000 units. Finally press top bar button"). Could actually be way more user friendly than a regular encyclopedia.

There are some things generative AI can do that voice actors simply can't. Imagine an AI which can live-commentate a match in a competitive game, or which can be integrated with Twitch/KZread chat and interact with viewers. There are opportunities for interactivity and responsiveness which just aren't possible when you have a set of prerecorded lines. I think the technology is not at all there yet, but it's definitely going to be a big thing.

ai has added nothing positive to the world

I dont care about AI if the product ends up fun

First

@beridledodopidop4822

3 ай бұрын

Third

if AI can create fun games more than western developer shenannigans im in

@midnightblue3285

3 ай бұрын

We can do experiment, but we shoudn't replace the human creativity also

I like AI, i support AI, i bet on AI. Ai doesnt steal, AI doesnt Take the inteligence and the work of others, it learns from it and create something new. People does the same, in a more complex way, but do it too. I'm an artist and a noob game developer and to learn how i should draw people i look at people, COPY their patterns and learn from them until i can make something that i'm feeling happy with, but not only from other artist, but for real life too, like our ancestors did. and in programming? like you literally "stole" all the time code from literally any source that works in your project. Creativity is just an group of copied stuff, arranged in such a way that we desire to be. In the today laws, you cant arranged those stuff in a way that other creator did before you, but still. Stormgate is just, hey, we liked starcraft... what if we made starcraft but WE are gonna take good care of it, and ah, a bit more bulky like WC, and a better editor, and humans v demons, cause is cool. and boom you gotta stormgate (SG, almost the same acronym of starcraft, SC). My point is, if you dont like AI because it wasn't created by a self conscious being, then thats ok, but, dont say things that aren't true

Kind of lazy depending on AI imo....stormgate seems like a very slow trainwreck

@rictekmusics3143

3 ай бұрын

I agree