With AI taking over the tech industry, its role in game development is slowly shifting from novelty to potential necessity. Host Devin Becker sits down with Francois Courset, a consultant at Naavik with experience on using AI in game development, to explore the current and near-future landscape of AI in game creation. The conversation covers everything from Francois’ high-level perspective on AI's role in the industry to specific tools that are already proving useful, alongside those that show promise but still need refinement.
The episode dives into where AI is making the biggest impact today, such as content generation and workflow automation, as well as less obvious areas that might benefit from AI with more maturity or experimentation. Francois also weighs in on whether studios should build internal AI expertise or lean on external support, and what developers should be doing now to prepare for what’s coming. The episode wraps with a rapid-fire round on the worst, best, and most exciting AI use cases in games. Whether you’re hands-on in production or thinking strategically about the future, this episode provides a grounded look at AI’s growing role in game development.
We’d like to thank Levellr — the Discord community intelligence platform — for making this episode possible. Learn more about unlocking real-time community insights at levellr.com.
We’d also like to thank Neon – a merchant of record with customizable webshops optimized for conversion – for making this episode possible! Neon is trusted by some of the biggest names in gaming and can help you sell direct without the typical overhead. To learn more, visit https://www.neonpay.com/?utm_source=naavik
This transcript is machine-generated, and we apologize for any errors.
Devin: Hello everyone. I'm your host, Devin Becker, and today I'm delighted to be joined by a fellow Naavik member here, Francois Corset, a senior games industry consultant. Francois joined Naavik recently and is providing some additional consulting expertise and experience in using AI for game development.
So today we're gonna dive deep into the current and near future state of using AI for game development, touching on some of the stuff, an article I recently had done as well in the state, and then some stuff from Francois and his experience. So just to start things out, why don't you go over your background and experience around AI by itself as well as with game dev.
Francois: Cool. Thanks Devin, and happy to be to be here. Thanks for, for having me. Yeah, so I've, I've been working in, in gaming across research and strategy roles for the past, the past eight years in companies like Nielsen, Ubisoft, and most recently at Netease as part of the, the, the global strategy team where I was in charge of like monitoring, identifying and assessing any kind of new AI powered tools that could be leveraged to improve our production pipelines or directly serve some of our gameplay propositions with this idea of like sort of open innovation, establishing strategy partnerships and collaborations with external tool providers and understanding, where the value was in all those new kind of, kind of AI related innovations. One of the things we worked on was supporting team Yazi, one of the first studios to implement in world AI's technology in a consumer facing product, which was selling US enterprises.
And so throughout, throughout this time at NetEase, I kind of saw firsthand like the, the, the, the lack of general present focus in the way we approach AI. Kind of looking at, at the hype without necessarily having reliable sources of information of like what AI can actually do right now in, in games with all also the, the, the multiple internal roadblocks to actually adopt and deploy AI solutions within, within kind of gaming, gaming companies.
And even on the two provider side, like struggling to kind of, to kind of enter those gaming companies, find the right point of context to actually engage and, and suggest those, those, those innovations. And so kind of took all those, those learnings and figured that they, they might be some, some, some potential to support the, the broader, the broader gaming industry. And I recently joined Naavik to help studios and any kind, of kind of gaming companies looking into AI and, and, and, and gaming as part of like a, a new AI consulting service that, that we've recently announced at Naavik, and more broadly supporting all consulting needs.
Devin: Well, excellent. I mean, as you've put it, like a lot of people are still figuring things out, so you're not hearing necessarily a lot of even learnings from a lot of these companies. It's definitely, let's, let's figure out what we can and can't do. And we're probably not gonna announce anything unless it works and we're happy with it.
So it's even the, the failure learnings, right, where it's like definitely difficult., But like, just kind of on, on a personal level, based off your experience, based off what you've seen and just based off what you think, what's your kind of high level take on AI in general and, and then specifically in game development.
Francois: Yeah. So in general, I think like any hot topics are, are sectors. There's a need to kind of cut through, through the noise. We tend to kind of underestimate the future potential of it because it's gonna be extremely transformative. But at the same time, the current potential tends to be overestimated, which leads to all sorts of, of, of issues when actually deploying solutions.
There was an interesting piece of research published recently by, by the MIT where there was this flashy headline of, of like 5% of test specific pilot AI projects are, are actually reaching production and generating benefits. But regardless of like, uh, the exact number or methodology, it, it kind of tells you that like there, there's a real struggle in, in kind of getting to those actual benefits of AI, how, how to actually impact the, the business, the efficiency. And so there's still a lot of work to be done there in kind of the AI space in general. And then more specifically, with, with gaming, I think it's, it's safe to say that that's AI is already there in a way. Like I, I think 20% of the, the, the, the games released on Steam this year were, were kind of including some sort of AI we're actually disclosing using some sort of AI, and developers are, are already leveraging all sorts of, of largely available tools like chat GPT or, or coding supporting tools. So, AI is there in a way, but still, a lot of the, the kind of low hanging fruits haven't been fully captured yet. So people are still kind of dabbling with those, those AI tools, and often not in a structured manner kind of integrated with within kind of a, a clear, a clearer strategy.
And even besides this, like the game industry itself has kind of a set of, of specific challenges, compared to, to other industries. The fact that gaming teams tend to be a fragmented group of highly specialized teams, that each use like very specific sets of tools that could all need to be kind of, kind of AI enabled in some way and kind of limits the potential of like large scale benefits of deploying tools within those, those, those teams.
The gaming projects are long as well. And it's, it's often tricky to kind of change course in like a three to five year kind of kind of production pipeline. So adopting new technologies, mid project can, can sometimes be an issue. And often in gaming we also don't see a lot of change management teams that are really dedicated to deploying and, and identifying all those new technology innovations built externally that could be beneficial for, for to game production tips.
And on top of that, you have also all the brand risk and consumer backlash that that kind of comes with, with dabbling, with, with, with AI. So it creates an a, an additional layer of risk on top of everything listed here. So, it makes it challenging for, for the, the gaming industry, but it also means that there are like, like very interesting opportunities to, to, to uncover and when properly executed, that can yield like significant benefits for, for gaming.
Devin: Yeah. I imagine a lot of the problems is just from having to find out the hard way, what doesn't work. Yeah. Especially there's just, you know, there's not even best practices yet, right. Like you said, change management and coming in, like people kind of dipping their toes in. They can't go all in. Right.
Unless they're like a very small company willing to risk the whole thing. And you know, they usually, those are the ones just trying to seek funding off of using ai. You very, very similar in some ways to what we saw during Web3 where it's like, this is a new tech, we're trying to figure out how to integrate it, might integrate it kind of naively.
And we're just hoping that we'll attract investor interest. Obviously investor interest in games is kind of at a bit of a low point in, in some ways, so that could be a, a big boon for them. But I'm kinda interested in like the, the very down earth practical aspects of this. Like what can we really do right now?
And so like on that topic, like are there some favorite tools either that like work well now that you like, that you've tried and used or ones that like, have shown enough potential where you're like, maybe it isn't doing it right now, but we should definitely be trying these and like building towards that.
Francois: Yeah. So obviously there are like tons of, of, of, of tools already available or being built either proprietary or, or kind of external ones, probably like close to, to a hundred startups actively building around, around AI in gaming. So like lots of, of like specific kind of kind of areas are, are being addressed on, on my side, like a few that I've really kind of, kind of found like deeply interesting in terms of like the potential behind it.
One is, is kind of pool day AI, which is an AI video editor centered around like ad creative production. And I like it because it, it kind of has this, this, this principle of like being able to, to do a time 100 on your kind of output capacity. So the idea is that you can create multiple iterations of add creatives by simply inputting content and like gameplay excerpts on and, and, and taglines and things like that. And the, the, the, the tool will produce like hundreds of videos that you can then use for a beta AB. So I think this inter intersection of like AB testing with AI and the, the, the potential of like, yeah, increasing significantly the outputs of like one specific task is really where there are like tons of value to, to, to be found.
So that's one. Another one that I liked, a lot is, is craft bench AI, which is like a, a Minecraft asset production tool using generative AI, which supports like UGC studios in creating custom Minecraft assets that can then be, be used within the, within the, their servers or, or, , and, and, and, and there, , like.
What's particularly interesting is that it really showcases how you can make the most of like a very in-depth dataset in the first place. Like if you start training your AI with a very specific pool of data, uh, that is really geared around your, your business objectives, you can really end up with something that is extremely ready for, for being actually published and shown to, to, to consumers.
And so craft bench is really interesting in that, in that aspect. And then finally one that, that, that I think has like a huge potential is model AI, which does, automated QA testing. They're basically able to deploy bots in game that will explore the game environment, find bugs optimize pathing thing, things like that.
And again, you have here the idea that like you can potentially deploy like thousands of bots to kind of, to kind of go through all the different, the different pieces of your, of your, of your level and, and find all, all the, the, the potential, the potential bugs there. So, once again, it's taking one small piece of the production pipeline, and then finding a way to really kind of, kind of, um, address it in a, in a significant way and scale significantly the amount of testing here that you can, that you can do with, with AR tools.
Devin: Yeah, that seems really useful considering just man hours for testing, especially when it comes to like testing something after a fix and all those kinds of things where you constantly have to test same content on a very tight schedule, usually seems, I mean, especially if you're like running live ops or anything else where you're gonna have to be pushing out updates and you need to be testing those, and I can't count how many updates come out partially broken in some way.
And just the, you know, testing coverage is very limited. So that seems like a really, really useful one. I mean, the part that sounds like it might be more difficult is like the recognizing of bugs, that aren't just straight up errors, you know, like recognizing odd behaviors and stuff like that. I'm curious how it works when it comes to identifying those things.
Francois: Yeah, and it, it's, it's always like so it, it's one of those cases where like the entire kind of game testing pipeline won't be kind of, kind of addressed by, by like a, a single tool, but rather it's about like finding those, those like tests that can be, supported by AI and improved by ai.
And then, kind of, kind of leveraging that to let the team kind of focus on the other areas that, that, that can, can't be kind of fully addressed by AI now, or, , hope oversee, uh, AI kind of, kind of a contribution because there was always an a need for like, for like human checks and, and, and, and like managing errors that can come out of, of, of AI models as well.
So, uh, it's about, yeah, again, finding those, those little kind of areas that you can, that you can, that you can address. And then expanding on that progressively as like the technology evolves, our models are becoming available.
Devin: Right. Kind of that theme of like, it's not about replacing people just yet 'cause it doesn't do the same thing as people.
It's about augmenting, enhancing what you can do. Right? I mean, I have heard, you know, obviously people at the programming level often are like, you know, AI programming is no good, stuff like that. But I have heard good things about it for like unit tests and that sort of thing, right?
Where it's like getting that kind of coverage, something that's kind of a pain to keep up on and set up. Like those kinds of things where, you know, as we hear all the time, repetitive tasks and things that are a bit like manual work, those tend to be sometimes best suited. I mean, it makes sense, right?
Something that logical and repetitive is good for a computer. Now we just have a bit more flexibility in how we execute those with ai. Exactly. I mean, that sounds, that sounds great. I am kinda surprised that the. You know, you mentioned the Minecraft one and it's like Microsoft's been trying to put AI into just about everything.
So it does surprise me that they're not pushing that, you know, considering they own mojang. So, my Minecraft is essentially a Microsoft project for the most part. So it kinda, it kinda surprised me that it'll showcase that sort of thing more, even if they are involved in it, and I'm not, you know, sure who's in charge of that, but it, it does sound like a good use case similar to what we see in like ROBLOX and things like that.
And UGC definitely seems like an interesting use case where it's like, well, if people are pushing back when game developers are using it, well then are people pushing back when average users for UGC are using it? And if not, then that's an angle where it makes sense because it's, you know, as they say, democratizing development as opposed to like.
Firing people, basically. That's the part that people don't wanna see. Right. So I, I mean, speaking of like the different areas, you're talking about some of the different areas you could try and hit. Game development is like this big, long, multi-year thing. Lots of things to it, like, you know, like making a movie, but maybe even more complicated in many ways.
Lots of, lots of different areas of development, lots of different people involved. What are some of the areas that you've seen AI be most useful for in game development? Like, like consistently where, like you mentioned a couple, right? With QA and maybe some EGC stuff. What are some other areas, whether there's, you know, specific tools or not?
Francois: Yeah. So I, I think one is, is asset repurposing. So the idea of like taking existing assets and, and, reusing them in a different, in a different way. For example, like create a Christmas themed version of my game assets so that I can do like a Christmas event, for example. So, here, it's, it's more about like, reusing existing stuff and iterating, rather than actually creating assets from scratch. Which is, uh, really where, where, where I think we see the most potential in this, this kind of re repurposing element. So it's especially helpful in live service content production, but it can be valued for SQLs, DLCs, uh, basically any place where, you can feed existing assets into your model to, to kind of, to kind of iterate on, to come back to this idea of like iteration and, and, and being able to, to kinda multiply things where, where really AI has, has like tons of, of, of potential.
So yeah, uh, I think this area is, is particularly interesting especially in 2D right now, um, because for 3D the tech still needs to, to, to, to kind of mature a little bit, but most likely in the, in the coming months or years, this truly the assets will be, will be similarly impacted by, by kind of this, this idea of asset reversing.
Another one, obviously, I'm, I've mentioned at creatives, I mentioned, um, QA, so I'm not gonna dive into this once again, but localization is very interesting as well. It's one area where AI is being used fairly widely for, for a long time already, but, but it's really one of those, those really low hanging fruits again, where the, the, the, the technologies is very mature.
Obviously, there's still need for, for kind of, kind of overseeing what AI does, because inherently has, has like hallucinations issues, errors, things like that, that you still need to account for. But there, there is a huge potential there. And with companies like AI and, um, kind of building around, around clear, around localization.
Voice also has proven to be extremely mature already with 11 labs obviously, but like all sorts of other kind of, kind of startups helping do text to speech or, or, or speech to speech. Kind of, kind of, uh, generation. So this, this is also a very kind of mature area. We've seen it included in the finals from Embox Studio as well.
So it's already kind of reaching consumers to, to, to an extent. And so they're definitely another, another, interesting area. And then, then should the, the final one, the dialogues, dialogue production, is also one, one place where AI can, can already be fairly seamlessly integrated. You have the, the ghost writer tool from UB sub that allows to produce box or secondary NPC interactions.
Again, taking like a, a very, time consuming, uh, parts of the, of the process and using AI to, to, to, to kind of deliver at scale on that specific task. So, so yeah, that, that could be some of the, some of the obvious ones. And I think the, the idea behind this is like, it, it follows a lot of like the type of, of content that you're willing to produce, like producing text content is already extremely mature. Voice content is very mature. And then as you go into images or videos you make, you run into more challenging with challenges when it comes to how ready the technology is, how easily it can be integrated for actual production pipeline needs, especially in AAA context.
So as you go into those, those, those kind of modalities, it becomes increasingly difficult or there's still a need for, for a few, few more months or years of, of development to reach like full, uh, full technology maturity.
Devin: Yeah, a lot of those sound like things that. It would, it's really important to have oversight on in terms of like quality checks and things like that.
Exactly. I mean, I've seen more than enough bad versions, like, especially like AI subtitling of things and I'm, I'm sure that, you know, goes to localization as well. I mean, how many Japanese imports have we had over the last couple decades where just, you know, the memes from the bad translations. I mean obviously like that takes a certain amount of effort to go through and do all that double checking.
Right. And I, and if I, I had my own experiences where it's like sometimes the amount of work to double check everything AI does, you get to the point where you're like, well, why didn't I just do it myself in the first place? Right. Where sometimes, you know, you wonder what was the, what was the cost savings?
Are you seeing anything where there's some good ways to be efficient about that? Are, is there AI double checking other ais work? Like, hey, go through and make sure that all that dialogue that you just, you know, translated from the language makes sense. Right? Here's like the type of characters, like does that, those translations, are they nonsensical?
Stuff like that. Is there like some amount, amount of automated oversight.
Francois: I, I haven't heard about like actual kind of, kind of, yeah, AI reviewing other AI, but you can, you can easily see how that, that could kind of happen. Like, especially if you, if you start having like QA oriented kind of, kind of, solutions, please, please flag the stuff,
Devin: Right?
Francois: Yeah. Yeah. And it all comes down to like internal governance, and I'm being extremely intentional about what you want to do with, with AI because, uh, as, as you said, there's, there's always this, this kind of, kind of need for supervision review and things like that, and you don't plan ahead properly.
There, you end up with processes that aren't generating any significant value or cost savings just because, because you can't trust the, the AI enough behind it. And so you need to really be conscious of like, model limitations, current limitations and, and how to, to manage that risk internally.
And how to, to get, reduce the potential impact as much as, as, as possible. And again, it comes down to how you structure your, your internal teams and product project management around these new tools to ensure that that can be properly deployed and it can be beneficial in terms of efficiency or, or cost saving for, for, for the studio.
Devin: It sounds like, I mean, the important thing then is to not to rush it, right? Like I, I think of the alters, right? What happened with that? And you know, that was also a transparency issue and things like that. And we're not gonna get too much into that sort of topic. 'cause that's a whole can of worms when it comes to ethical stuff, transparency, business models, you know, customer relations, all that stuff.
It's, it's, it's tricky, right? This is more about the, the potential in the tech, but I, I, you know, I feel like, you know, the number of times I've seen like lo localization or other dialogue attempts where you see like actual, like parts of the prompt in there and other things like that, that's just those things that slip through the cracks where it's giving enough time, not just for the actual process and double checking, but like you said, for maybe setting up the proper governance for it or the proper structure so that there's like proper reviews for it.
That it goes through a pipeline, that if there's changes that, that also goes through the same pipeline. So it's like, just don't just jump into it thinking, oh, we're gonna rush into this and suddenly save our game company. Yeah, there there's no magic.
Francois: There's no magic there. Like it, you, you still need to, to do the work in a way and work alongside AI rather than kind of living everything to, to a, that, that's where you run into, into massive issues, even backlash and those kind of, of yeah. News that we see pop up here and there.
Devin: I mean, on the, on that vein too, like you, there, you, there's a lot of these companies experimenting internally. Right? And I think part of that's just like, they, they wanna figure out what will work for them with their particular workflow and, and their particular data and things like that.
But that's, that takes time away from the working on the game, right? Where it's like they've got a budget, they're on that game's budget. They're not usually, unless it's a studio with a ton of leftover money where, you know, if it's like, hey, maybe like Valve or, or something like Rockstar where they've got money to burn, to some extent they, they kind of have to burn some of the game's budget on that.
Are you seeing like where it's, you know, it's some of the developers just looking to see if they could improve their own personal productivity. Is it like independent teams kind of just experimenting kind of a prototype lab? Like, you know, Ubisoft for example, has their, their individual lab doing those sorts of things, see some of that.
What, what's kind of your overall take on how people are approaching that internally in companies?
Francois: Yeah, so I, I think they're like so the, the idea comes down to being intentional about what you wanna achieve with AI and keeping that, that business focus in mind. You see different initiatives, like you mentioned, AI labs being, being built, all sorts of like internal R&D uh, initiatives.
All of these need to be like tied to actual, uh, business objectives. Like, if you end up kind of experimenting with like a team of AI researchers looking into anything that that's possible through AI, that's a significant issue. And, and, and we've seen kind of projects and, and, and pilots that, that, that are more kind of showcasing what's visible rather than like, what's can, what can be actually valuable for, for businesses.
So I think, what, what studios need to do is really balancing the needs from, for between like extremely specialized, tools that, that will, that will yield like located benefits versus tools that claim to address like a massive part of your, your, um, game production pipeline. I think it, again, there's no magic here.
Like production pipelines in games are extremely complex, involved, like a lot of different, different tools and teams. And so, it's about like identifying those efficiencies, reviewing strategically and systematically like all those innovations to see what can be fitted into, into the, the, the various projects.
So it's, it's really about like saying alert, finding the right teams or external tools that can be, that can be easily leveraged and plugged into to address specific careers. And when relevant and mentioned like it's, it's mostly for like broader kind of kind of organizations like really invest on building an edge in a, in a specific area based on internal data sources, for example, that can't, that, that, that can't be accessed by external company and that allows you to actually polish a tool one step further and, and, and, and really be able to, to kind of, kind of optimize your own internal processes with, with those, those kind of larger IPs or, or, or studios.
Devin: Yeah, it sounds like a, a long process for a lot of these companies trying to figure that out. It sounds especially tricky too, because like, you know, you have these business goals and things like that, and most of those are gonna be financial or times based, and you can't really show off the fruits of those results to the customers generally.
Right. Because the players are not gonna like to see that. You're not gonna be showing off like, Hey, look what we did with AI. We saved ourselves money, or we like generated something that you guys thought was real. Those kinds of things you just, you can't brag about. And so it becomes kind of this weird thing where, you know, you could kinda show it off eternally a little bit to be like, look, we found some things that worked, and kind of have the communication and some of that governance and sort of infrastructure inside.
I imagine that a company culture then matters for that too. 'cause you know, you get the internal revolts of developers like, Hey, are you trying to get rid of me? Are you trying to take my job? I don't wanna see KPIs improved there because that means the threat to me or, to the creativity of the game.
Things like that. That seems like a really tricky process to navigate.
Francois: Yeah, and, and, and really the, the teams that really kind of, kind of manage to progress. There are, are companies that, that have those established kind of strategy game take teams in a way that, that really look at, at those, those tools, um, at that understand like all innovations coming up from the startup world, but also look at the broader kind of AI innovations, kind of the, the, the broader business landscape and trying to, to really tie that to, okay, what are our internal processes and, and where can we match those, those kind of two pieces together.
Devin: Right. Especially now too with, with the KPI stuff and the spend that it takes to do that stuff. You have that whole, like people talking about, oh, the returns just aren't there for ai. That sort of thing where there's a little bit of like that sort of. Trying to bubble pop, I guess at this point. Or deflate.
The excitement around AI is definitely like trying to set that realism in there of like, this is not gonna be like a magic thing. We need to actually figure out what's gonna work for us from a business perspective, or at least something that's going to like, benefit us long term if we can have afford to do that.
And, and like those are some of the obvious things you talked about. Like when it comes to things you could generate, things, you can actually translate, things like that. What are some of the more unobvious uses or some of the, like, the really interesting things that you see potential in that? Maybe like outside of the 3D right?
You mentioned 3D is one where it's like obviously like not quite there yet, but I think it's almost a given, like what you said. Yep. What are some of the, like the, the, an obvious ones are the ones that people should be putting more time into or thinking about or just don't let it blindside. I'm like investigate now, kind of thing.
Francois: So I think, one that is, that is extremely interesting. More kind of on the, the gameplay side of things is, is really this, this, this idea of like, adaptive difficulty, and like having a, a, a game world or an environment that, that's constantly kind of adapt, based on the players' input can be like adding more enemies, based on, on what's, what, like you, you, you, you kind of like what the player does, like if, if it's a good player, football enemies to keep it interesting and engaging.
So I think that's, that's, that's really an area where having tools that are context aware and that can allow to tweak, parameters, for the gameplay experience would be extremely, extremely relevant. Been done to, to, to, to a certain extent all the, all the games like Left 4 Dead and things like that.
But pushing that further with the latest AI innovation is definitely an interesting area to, to, to look at for. I think there's also a, an interesting angle for like brand engagement or IP engagement and, and kind of a consumer facing way. We see a lot of engagement, like, not necessarily AI related, but with clip productions, like with Roblox for example, Roblox Clip it. And so we see consumers kind of going in between different content formats like videos, gameplay, texts and so on. And AI has really the potential to help blur those lines between content types. And, and could be used to, to, to really engage consumers in a different way to allow them to create viral clips, viral images, viral texts, things like that to really support the community buzz around specific IPs or aims.
There are like specific initiatives there that kind of point in that directions, like kinetics allows you to kind of create a custom emote by, by scanning your body. Which is, which is kind of one way to do that. You have Dorian AI that, is, is a platform where you, you can, you can write, aided by AI about your favorite, characters or, or, or things like that.
So really kind of pushing this, this text user generated content area as well. So this idea of like, yeah, using AI not only to strengthen ability to create content in UGC, but also to enable consumers to, to double with different sorts of, of content, I think is a really interesting area to look out for as well.
Devin: You make me think of the, some of the stuff that, that we've seen before where people try to experiment with trying to bring characters outta the game. Like, even, even like Left 4 Dead, you bring up, I remember when there was like Twitter accounts for the, the different characters. Like, I don't even remember if those were official, but just like, you know, talking as them and like bantering and sort of thing.
And it's, and it brings up that idea of like control of the experience when it comes to using AI that's reactive rather than like in the production process, right? Like there's this big separation between, is this something we're using to produce content in some way or produce a game? Yep. Or is this something that is real time to interacting with players in some way?
And we have seen some experiments with that, like, , the Darth Vader and Fortnite was a good example, or any of the chat bots that've been online. And, and the big theme that I noticed with those is when they go outta control, it becomes its own form of publicity. Good or bad, right? Like, and you could say maybe all press is good press and maybe, a lot of people might not have heard about the Darth Vader.
Character being in there that you could talk to it and wanted to go talk to it after it actually played Fortnite because they heard about that because of the bad press around that. It could have been that, but that was like two brands having to deal with the backlash that way. You've got the Star Wars side of things under Disney and then you've got the Fortnite side of things, having to deal with that as well and maybe even the AI tech side of things, having to deal with where that might be loose and, and it just makes me think like you, you know, you talk about like the left for dead style directors or any of this stuff that would normally be a bit more procedural or algorithmic suddenly being like, can we use these models to sort of like take a little bit of the rails off and allow for a lot more situations that we then we could code for, but also doing so the rails are off, right?
And you end up in these situations where now, I mean games are somewhat about controlling the experience, right? Where you're building around an experience, but you don't necessarily control. The way you do, like with the movie or something like that. Right? But do you think it's possible for us to kind of find that middle ground where, where we allow players to really just kind of do whatever they want, maybe in a single player environment, maybe not as much as a multiplayer or, or even around that content creation stuff where they're not gonna go, Hey, I'm gonna make Darth Vader, drop the N word and then clip that, and there's my social media content I created with it.
Francois: No. Yeah. It, it, it all comes down to, to this idea of like where's the fund first in what you, what you do, with AI, you need to ensure that that what, whatever you're building, it's gonna add value to, to the end player. Like, is there part of my player base that is actually interested in, in playing what I'm, I'm, I'm, I'm building like the new gameplay, proposition that I'm building using, using ai.
Because you can really see like AI. Easily allowing to do endless things, right? But is endless fun, really. And how does that serve your game to have infinite conversations with anyones that anyone that you can meet in the game. So I think it, it, game studios will need to balance like what's editorial and what's kind of artificial or what's generated by AI within, within those games.
So, as you were mentioning there, there's a lot of, of controlling to do on, on, on, on ensuring that AI is used like in places where it makes sense and in places where it can't be broken too easily by players. So, so that's the first thing. There's another thing that will likely be kind of cultural evolutions around the way we look at, at AI generated content.
You can see it, um, uh, already in this, in like consumers are, are already playing with AI tools and generating images or content for fandoms and online communities in general. So it's already there and consumers are already kind of playing with your brands in ways that you haven't expected them to do.
So there, there will need to be some, some, some wiggle room in the way to, to, to deal with this, this kind of additional freedom. And, and avoid like all the risks and, and, and backlash and navigating that will be, will be tricky. But again, it comes down to. How intentional you are with your strategy in the first place.
What exactly, where exactly do you wanna implement AI and how do you, do you kind of consider all the risks and limitation and mitigate that as you progress with the implement implementation process?
Devin: I'm kind of curious too about even like that freedom skewing the gameplay, not just from a breaking the game in terms of what you could do with it perspective, but it, in terms of players then feeling the desire to purposely break it.
So I think of like, you know, when you have any kind of UGC element, whether it be like the Minecraft level or whatever, Legos, anything like that, you have the old time to a certain phallic symbol that that tends to happen, right? Where people have this innate desire to see where the limits are as soon as they're presented with something that is supposed to not have same limits, right?
So you, let's say you throw an NPC in and it's like, hey. Now you can say whatever you want to it. Chances are they're going to completely break immersion themselves and not roleplay until they've thoroughly seen how far they could push it. And then maybe once they're bored of that, start roleplaying and actually playing to the character and seeing what it does.
I can see that being problematic to some extent versus like a more, you know, this dialect tree system or these other ones where you might be able to allow a lot of flexibility, but you don't necessarily have the same problem. I mean, I even think back to like the old text adventure games with like noun and verb sort of thing, right?
Where everyone's just trying everything that you could possibly do, even if it's just kind of a joke. But that was more in like in the, the days when you didn't have viral social media or something to share those sorts of interactions. And they were all kind of hard coded to some extent. And so, it's, it's a little bit of a different, different world in that sense.
But, but I digress. It's, it's, it's gonna be interesting to see how that, that problem solved. And maybe we just have like sort of an immature. Time period where people kind of have to get used to it before they actually take it seriously.
Francois: So I guess what, see, and we do have also like emerging kind of, kind of moderation tools that also use AI.
We're talking earlier about like AI kind of monitoring AI. Yeah. Moderation is, is also benefiting heavily from, from, progressing in generative AI. And so, this can be like the missing block that, that kind of allows to control a little bit more what's being produced or done around like, all kind of AI applications deployed in games right now.
Devin: Well it's, it's definitely gonna be interesting to see like how that goes. Just because like there's a lot in motion right now, which is a good thing and a bad thing, right? It's a good thing in that there's, the possibilities haven't been set yet and there's lots of things that can be done.
Some of it's dependent on the technology and the timing and money and the market and a whole bunch of other factors, which also kinda created situation we're in right now when it comes to game developers who are considering using this stuff. Obviously, like, you know, we've talked about specific tools, things you've seen work and things like that.
Should game companies in general be experimenting internally or should they look to outside expertise or, you know, tools or things that kind of solve problems already or thought about those things? Figured out those things so they're not burning internal time 'cause especially now that, you know, running into like a lot of crunch time for games, whether that be on time schedule or a money issue or any of those things, or just a, you know, manpower in general.
Francois: Yeah, I think so. There, there's no one way answer there. I think it's, it's often a case by case analysis, but, what this reveals is just the need for having teams, uh, that, that kind of systematically review and assess cost opportunities associated with, uh, all due these new AI innovations across, across the entire, the entire ecosystem or the entire production, production pipeline.
Based on that, like, there are, there are a couple things we can, we can say there, like, the first thing is that AI tool providers like often benefit from significant platform effects, especially when they're consumer facing. They can like reinforce learning on their AI models and improve the quality of the output and even sometime kind of access like private training that data sets or, or leverage some of their clients' data sets to further reinforce their model. So there are like clear benefits on the kind of AI tool provider sites when working across like a, a large variety of clients. Then I think the other angle is, is really assessing like the moats from those companies.
Like, is it just a wrapper around a, a model that is, that is widely available and that I could probably replicate internally? Do I have like unique data sources internally with, we mentioned that previously, but that can allow me to, to build something that others won't be able to build to the same level of kind of refinement or that would be specifically tailored to my own kind of, kind of internal need within, within that studio. So that's, that's kind of another angle. Like if you, if you do have this, this kind of capability to have that edge or you have those kind of, yeah, data sources internally, that can be huge, huge benefits. Then AI is, AI talent is extremely expensive as well, especially for lower size teams.
Not every studio can afford like an AI lab with, with like a few researchers kind of experimenting with AI. So, for like the majority of gaming companies and studios, publishers, like there's a need on reliance on external partners to some extent. Uh, in any case, it can help move faster because you can have leveraged tools that have already been produced, but we need to be conscious of the fact that we're still quite early in those tools kind of maturity. And there are a lot of things that still need to be narrowed out in terms of like, uh, production pipelines, processes and how it really fits within, actual kind of, kind of game companies. So any kind of AI initiatives need to expect some level of customization or kind of working hand in hand with external partners to make sure that what you're deploying kind of really works, um, for you.
And then finally, it's about like keeping focus. Is, is your company building games or building AI models in a way and, and what brings the most value to, to your company? And so, yeah, leveraging external kind of, kind of, providers can, that can definitely help kind of alleviate a lot of the, of the distraction associated with, with kind of looking into, into AI.
And there tons of applications that can already be explored with the solutions that are, that are widely available. So building kind of internal AI solutions or replicating them is really something where it makes sense if you are trying to build a competitive edge in a specific area. If you wanna move faster than competitors and that is gonna justifies the, um, significant investment that you're gonna put behind, like actually developing this, this new tool or this new kind of application area. So yeah, there is value to both. It depends on the company's size and, and priorities. But yeah, generally speaking, you cannot process this. Studios need to, to really look into all those solutions and, and really have this case by case, by case approach of is it mature enough, is it bringing valuable kind of efficiency to, to our company? And if so, moving forward with, with this is our internal oxide.
Devin: Hey, you bring up interesting stuff about the, the data, right? Being a big part of that decision. You know, are we sharing data with partners? Does do partners have data that we don't do?
We need to like, try and build this moat ourselves to have a, a competitive moat against competitors, especially like a publisher or big studio level where you're like, Hey, we're doing this because we wanna be ahead for our, you know, games under our big umbrella. Those sorts of things. So, we can help partners do that sort of thing.
You know, it makes me think of some of the, the studios that will help like, uh, voodoo and stuff like that, sort of groom and, and help these, these studios underneath them, build up what they're doing and having expertise like that could be. Helpful. But it also makes me think of like stuff where Roblox, for example, was touting, Hey, we're the only ones who could make this sort of 3D model generative stuff because we have this massive data set of games using like 3D modeling within our sort of a sandbox.
And we can use that data set to then build a tool that would be very difficult for anyone else to replicate internally or, or otherwise. So it, it is an interesting consideration there, like that might be like on a topic by topic, case by case basis, as you said. It makes a lot of sense to kind of evaluate that carefully before making that decision.
Francois: Yeah. And its also at like broader changes that will like to happen within, within organizations around like how to systematically collect the data that will be helpful to fine tune models or build models because AI models are just as good as the data that you feed them, with initially.
So, companies will also need to, to kind of like, I think, like before even deploying AI, like how can we properly collect data that's gonna be, that's gonna be helpful. Like if I did like thousands of concepts over the past few years, like how can I take the learnings from this and, and actually make something that's valuable.
How to take value and extract value out of all those, those learnings and unstructured learnings and data that studios can, can collect over time. So that's another part where, again, being intentional about the way the company is structured, the teams are structured, the team can work with each other's, can, can really help enable longer term gains and, and, and opportunities for studios when it comes to AI.
Devin: Definitely, I mean, it makes me think like that studios, especially that do a lot of games in the same genre or like have a lot of like historical data from their own games and they're making those kinds of games going forward can especially benefit, because they can train off that data. It reminds me of what you were saying about the live ops sort of thing where you're like doing the seasonal content, like the remixing kind of thing.
If you were making games within, like whether those be sequels or games within a certain genre or theme or whatever, that you could utilize that sort of thing to train the models. Obviously training models is kind of a weird thing that's not always accessible and, and depends on the, the models and stuff like that, and that's a whole tactical thing we won't even get into.
But I think the big question, like how, having discussed a lot of this stuff, like sort of the, the more nitty gritty with the, the details of how studios think about this. The big question is, is should most game studios and, and you feel free to break this down into individual silos and, and categories or whatever, be experimenting, looking into toying with or considering AI right now.
Or in the near future? Or should it be a, most of you should just be like, Hey, let's wait till this stuff's mature. We're already like taking enough risk, even just making a game right now. That sort of thing. What do you think overall on that?
Francois: Yeah. So I think, I think there we, we need to, to come back to the broader picture of gaming right now with like this need for production cost optimization, especially when it comes to AAA kind of projects that have like, that are reaching new heights, budget sizes, as well as also the competition from emerging kind of kind of markets that are delivering similar level of quality at, at a fraction of the cost as well.
So, they are like strong external incentives and, and pressure for studios to optimize production costs for games in general. So that's, that's the first thing. The second thing is, as I mentioned, like AI is already there. Your competitors are using AI already. So, you should probably look into it in some ways to see how that could translate to your own needs.
And we've mentioned that previously as well. There are already like several careers that are mature enough to, to yield like efficiency gains within studios already. So, it's not that much about the maturity anymore, it's about the ability to identify those opportunities and then manage change internally as you kind of integrate those solutions.
And then also your developers are already using AI. So there needs to be some level of intentionality there as well in the way you allow AI tools to be used and where, and, tracking the potential issues that might come with, with using those, those AI tools. So in a lot of ways, like you can't really, like studio, studios will need to adapt to this, this new wave of like AI innovations that is coming. Again, it's, it's kind of already there. The, the, the one part where it's, it's less, less of an obvious kind of, kind of area to focus on would be adding gameplay benefits like some studios are kind of experimenting with finding new gameplay propositions, using AI directly.
And there are a lot of kind of issues to be narrowed out. There are like, uh, what can be enabled by ai? How fun is it? How to navigate consumer adoption? Who is actually interested in AI within my, my, my consumer base? And, and most importantly, even like how do I monetize this because token usage can become like a huge cost center as well.
So how do I deal with that? So, all of this really depends on whether you are trying to find the next new shiny thing or, or the new kind of genre that's, that's gonna be a breakthrough, which is definitely something that as an industry we need to experiment with.
And some companies will find success getting there, but it doesn't have to be a business priority for all studio or all publishers. There are no like obvious indications right now that in three years consumers will only play games with AI mechanics. We still haven't really proven that AI mechanics can actually be fun and bring like significant added value to players.
So, I think on the tool side of things, definitely they're lack. Pretty much all studios and publishers need to stay on top of what's going on there, and understand those innovation and where they fit within their production pipeline. On the content side, it really depends on your company strategy. But it's far from being like a must have to kind of build AI features in.
Devin: Yeah. I mean, there's a lot to consider there, but I knew it wouldn't be an easy yes or no answer, which is, which is good here. So, it, it depends, is always the answer. Right. But, but you brought up an interesting point in there of even if the answer is no, that doesn't necessarily mean that employees within your company.
Or external partners, contractors, things like that won't necessarily be doing that. And I think of like, this comes up a lot in Marvel Snap, for example, where they contract out a lot of artwork and it's not a hidden thing, right? Like they have the indicator of who the studio is that does the artwork or the particular artists or things like that.
And there's this sort of witch hunt accusation sometimes around is that AI generated art? And you see that sort of thing. And, and that's obviously gonna get crazy and crazier, right? Where everyone's gonna be accusing everything potentially of being AI art. But it does bring down that sort of idea of like governance that you mentioned before, before, but not as, not just in the organizational level of where you're like, Hey, here's how we're using ai.
It's also, are you using AI not telling us? Because that can be a huge brand risk if you're, if you're unaware, right? If you just, oh, we have no idea that someone did that. Like, so, the, the con, the controversy that came up with marathon for example, wasn't around AI, but it was around plagiarism, right?
Where it was like, oh, someone that worked at the studio but no longer works here plagiarized from that. Potentially, and therefore, like we, we, we have to do this huge purge now, but you could see the same thing happening with the AI generated art, where it's like, oh, that particular person within our company was using a whole bunch of generative art and our customer base is not interested in that.
Now we've gotta go through purge and redo all that. So, it seems like in itself, regardless of whether you'd like to bury your head in the sand and be like we're not interested in ai, we're focused on what we're doing, you're still gonna have to be like, but is anyone within the company doing that that we need to know about?
Or our contractors who. Maybe these, and I could see that even more in contractors looking to cut corners, looking to shave costs or keep costs down in bids, things like that. Just being like, eh, I mean, localization in general. I feel like if you're, if they're not super clear with you, there's a good chance they're using AI without telling you, like, just because they're gonna be looking for an edge and able to keep costs down.
So definitely something to consider. But I think you mentioned three years ahead, so I, I'm kinda curious the next two years, what do you think that looks like? Put on your little nest hat here. You, you can, you can feel free to be a little bit, you know, optimistic, but this is the, the no hype version of what do you think the next maybe.
Six months to two years looks like in game development in regards to AI and sort of the industry in general and, and games.
Francois: Yeah. So, yeah. Like as, as I mentioned, I think we're overestimating a lot of like what, what AI can do right now. And, and in the next two years, we're gonna need to kind of face, uh, like the harsh realities that comes with AI, which is that you still need to do part of the work.
You need to deal with governance or you need to deal with how to properly implement, manage change, measure success, train your teams on using those AI tools. And this will naturally take time. So, I think in the next two years, we are really gonna get into that phase where AI companies not only double with AI, but adopts AI widely. So AI tools will yeah, become fairly standard practice across multiple processes with others that will likely be unlocked as technology gains and purity, like 3D assets and animation, even coding. So, these arrays will, will kind of increasingly be addressed as well on top of the ones that are, that we've already kind of, kind of, covered.
So yeah, that's kind of the first, the first thing, there, there are tons of needs that, that come with fully adopting AI, especially when it comes to your own teams and talents. And regardless of like, specific processes, there will be like a need to progressive adapt to also how those, how to think about those tools as well. Like we still haven't, like, um, conceptualized fully how like a prompt based tool can be like fully leveraged in a way. So, there will need to be like lots of experimentations and training around how to, to deal with those, with those, changes. I think one of the most concrete change that we're gonna see in the next two years kind of aligns with broader dynamics within the industry of like increasingly relying on, external teams to support like a very narrowed kind of, kind of core development teams.
I think we're gonna see this more and more like small teams that are leveraging external partners or tools to really kind of support their production capacity and, and help them kind of reach, a higher level of ambition, but Clear Obscure.
Devin: Right. Where that was kind of an example of them leveraging a lot of outsource partners to be able to do what they could at that, that budget they had.
Francois: Yes. Like, and, and yeah, kind of keeping like a very kind of focused team. And so yes, we're gonna see this more and more. Small teams that are somehow kind of AI enabled, and yeah, working externally more, more and more. So it won't be like a radical transformation.
Like studios will most likely work in the fairly similar way to each years than they are they are right now. Like, it's not like a fundamental reshuffle of everything we're doing, but rather it'll be a range of like multiple located pipeline optimizations that will yield like significant costs and agility benefits.
So, so yeah, I think that's, that's really kind of the angle people should kind of stick to is this, this idea of like, yeah, it won't like change everything and like it, it, it won't be magical and, and kind of address, address all my needs. There are like a lot of kind of implementation work to be done.
But at the same time, like besides the, the kind of shiny things that AI could potentially do, there are like lots of less sexy areas that can actually be addressed by AI right now. And those should be the ones that companies kind of focus on. And then in the next two years as well, we're likely gonna see the first generation of like fully AI enabled games with new gameplay proposition coming to markets, with these kind of radical kind of gameplay innovations. And that will inform whether gen AI can actually provide significant gameplay value and benefits to, to players.
And so, I'll likely gonna see the first kind of confirmation there and potentially like the first few games that, that reach like a significant scale while relying heavily on AI gameplay, AI enabled gameplay.
Devin: Well, it sounds like at the very least you think AI in some form or another is probably here to stay in that.
Definitely. It'll deliver something, and it's just a question of where we can find where it sticks, which I think is the most pragmatic way to look at it, right? Just, we'll, we'll do something with it. We spend enough money developing all these tools and figuring stuff out and experimenting that we'll, we'll find some use for it.
At the very least, it's just a question of how much we could integrate it and whether or not someone could find a, again, I imagine if someone comes up with a big, huge hit that uses ai, then there'll be a million clones all clamoring to do that, and investment money and things like that. That's tends to be how it happens.
As soon as there's big success, you'll see more of it, but we'll see. Right. But now's the time for some, some quick rapid fire ones. So this is short answers only potential hot takes if, if you got 'em, but just firing off the hip. So just, just three quick questions before we go. So first off, the, the worst AI decision you've seen?
Francois: Well, not necessarily like naming names, but like we've seen multiple cases of AI being used for key arts. And, and I think that's, that's taking AI the wrong way. Like you shouldn't start with your most important assets. You should start with what can be easily automated without like being too visible to your, your consumers and audiences.
So that just shows kind of, kind of a lack of internal governance, as with, as I've mentioned, and, and, and in the way you kind of look at AI use.
Devin: Don't use it for stuff. You're just kicking out the door, I guess. Well, what, what is the coolest or most exciting use of AI you've seen, even if it didn't always work out the way you'd hoped?
Francois: Yeah. Yeah, I think so. I haven't like, talked too much about like very forward looking, uh, areas, but one that I like a lot are wall models like seeing, applications like Deckard, uh, Ossis, where you could basically play, generate Minecraft gameplay in, in real time. There are like multiple issues or roadblocks or like, it doesn't have like concrete kind of business applications right now, but it's one of those areas where you get a glimpse of something that is radically different and then really change the way you even think about how to build games, what building games mean.
So that's one of those areas where, where it kind of opens up like a, a really long-term perspective of like how transformative AI can be in the next five to 10 years.
Devin: Cool. Well then lastly, the favorite game using AI. And that could be in the production process, but even better if it's in a more obvious use.
Francois: Well, I think at the area where, where we see like AI being very, very interesting is when AI is put at the center of, of fun, as I was saying. And one game like that is best by AI which had some, some, some really big success, um, as well. And it really kind of perfectly leverages AI chaos to support, uh, the gameplay.
So you, you kind of basically engage with a, with an AI that kind of judges your, your answers. And it's, it's really fun. It's, it's really easy to adopt as well for consumer and one way to really engage with, with AI from a, from a consumer perspective and, and yeah, to figured out the fun. So, so really interesting one.
Devin: Yeah, it sounds like keeping it loose might be a good way to integrate that way. This doesn't have to be a serious game 'cause it probably won't work that way. Yeah. Makes a lot of sense. Well, cool. Well, I wanted to thank you Francois, for coming on. I really appreciate it. A lot of expertise, a lot of actual practical hands-on use you've had, uh, as opposed to, I think, a lot of people out there that have mostly just kind of talked about this.
So really appreciate your insights and I look forward to a lot of the interesting stuff to go forward both to some of the stuff you're talking about as well as stuff hopefully you'll be involved in. But, again, thanks for coming on. Hopefully you out there, got some good useful tips from this and we'll definitely be, you know, talking about AI where it makes sense in the future.
And obviously if you wanna talk to us about AI, make sure to hit us up as well as we are definitely, as Francois mentioned, we'll be looking into providing some services around that as well, to help people out so that way they're not maybe de-risk a little bit of their internal side of things. But anyways, thanks everyone for listening. Hopefully you enjoyed the show, and we'll catch you guys next time!
If you enjoyed today's episode, whether on YouTube or your favorite podcast app, make sure to like, subscribe, comment, or give a five-star review. And if you wanna reach out or provide feedback, shoot us a note at [email protected] or find us on Twitter and LinkedIn. Plus, if you wanna learn more about what Naavik has to offer, make sure to check out our website www.naavik.co there. You can sign up for the number one games industry newsletter, Naavik Digest, or contact us to learn about our wide-ranging consulting and advisory services.
Again, that is www.naavik.co. Thanks for listening and we'll catch you in the next episode.
