Value judgements of AI games aside, I don't believe him. That is a massive number that I think he's just pulling out of his ass and I suspect it's very incorrect. AI, even the reasoning models...
Value judgements of AI games aside, I don't believe him. That is a massive number that I think he's just pulling out of his ass and I suspect it's very incorrect. AI, even the reasoning models like o3, are just not very good at building large cohesive projects. At best you'll get a function here or there that actually took less time to fix the AI's mistakes than it would've taken to write. I'm actively working on a project where I'm somewhat heavily using AI to help me out with the big APIs that have a lot of boilerplate (Vulkan mainly). It does kind of okay on generic boilerplate, but almost any time I try to get it to write an "interesting" function based on things unique to my project it fails.
The code output feeling about the quality level I'd expect of a junior engineer that got their instructions via the game telephone doesn't really impact it's usability in other domains though. It'll probably just about completely kill ambient music and VO roles as well as taking a big bite out of graphical creation asset roles. I just don't believe we currently live in a world where 80%+ of video game code is coming from AI because it frankly isn't good enough to build the entirety of any but the simplest games.
Anything not fully self-contained may as well be a mystery to any coding AI assistants that I have used. They are pretty good at "make a function that takes X and Y primitives and manipulates...
I just don't believe we currently live in a world where 80%+ of video game code is coming from AI because it frankly isn't good enough to build the entirety of any but the simplest games.
Anything not fully self-contained may as well be a mystery to any coding AI assistants that I have used. They are pretty good at "make a function that takes X and Y primitives and manipulates them" or "given this static function, generate some starting test cases".
As soon as it involves a call out to anywhere else, it is a roll of the dice. Maybe it can come up with something if the package is popular enough on SO to have some answers, and then only if the package hasn't updated recently.
I am with you here. I think they want it to be 80-90% AI made but I think they are lying or some.project managers are lying to them.
It can be super super useful though. For example, these test cases for this file were 90% generated by AI with no revisions. That means the LLM essentially simulated the 1300 line file when...
It can be super super useful though. For example, these test cases for this file were 90% generated by AI with no revisions. That means the LLM essentially simulated the 1300 line file when generating input/output pairs and it did it perfectly.
But I agree they are not a general solution to programming problems. Part of the problem with them is learning when they're a time saver and when they're a time waster.
Yeah this number feels more like "80-90% of games have someone using AI somewhere." Meaning an admin assistant might use AI to help her draft emails, a writer might be using it to break up writers...
Yeah this number feels more like "80-90% of games have someone using AI somewhere." Meaning an admin assistant might use AI to help her draft emails, a writer might be using it to break up writers block, a localization team might be using it as a first pass.
My thoughts as well. Even if someone uses AI to generate a skeleton for a code module with custom names, that's far from having AI do the "meat and potatoes" work of game dev. (And project...
Yeah this number feels more like "80-90% of games have someone using AI somewhere."
My thoughts as well. Even if someone uses AI to generate a skeleton for a code module with custom names, that's far from having AI do the "meat and potatoes" work of game dev.
(And project templates / skeletons have been a thing before AI, but I could just see AI streamlining it a bit more.)
AI is only as good as the operator, and output quality is directly related to input quality. I'm positive that your final assertion is false, speaking from experience. Complicated games can be...
AI is only as good as the operator, and output quality is directly related to input quality. I'm positive that your final assertion is false, speaking from experience.
Complicated games can be built by one person with no assistance. So it makes sense that the same person can create a complicated game more easily using AI.
Garbage in, garbage out also applies to the training models. If we hold the Sturgeon's law to be true-ish, 90% of all of human creation is crap. Thus, the AI will spit out 90% crap. An LLM has no...
Garbage in, garbage out also applies to the training models.
If we hold the Sturgeon's law to be true-ish, 90% of all of human creation is crap. Thus, the AI will spit out 90% crap. An LLM has no way of discerning the wheat from the chaff. It will give the most popular thing, regardless of its correctness.
Just look at stack overflow. Rarely is the chosen answer the correct answer, and you'll find a more-upvoted answer that is better, and a 0-vote answer that is now correct in the 10 years that passed since the question was asked.
Are the majority of Gitlab repos putting out genius code? Or are they putting out amateurish/hobbiest code which is functional but sub-optimal?
I'm not quite sure what your point is? That the models are inherently garbage and there's little to no chance for the output to be good? That would be easy to prove incorrect. Of course there'd...
I'm not quite sure what your point is? That the models are inherently garbage and there's little to no chance for the output to be good? That would be easy to prove incorrect. Of course there'd have to be some qualitative way to judge things. But from what I've seen is quite easy to get good results. The better the prompt/spec the better the results. A mediocre, average, simplistic prompt will naturally get the matching result.
The skill involved in extracting the best output by forming the best input (prompt) is the same as that involved in doing a google search (witness that some people are better at it than others) or even finding something in a reference book. Like hunting for a needle in a haystack.
Key quote from CEO Hino Akihiro: Here's a longer Famitsu article in Japanese with more info about his Top Game Creators Academy initiative.
Key quote from CEO Hino Akihiro:
“Currently, around 80~90% of codes are written by AI and then fixed up and finalized by human programmers. In other words, it means that right now, around 80~90% of games are made by AI.
And it does not just stop at programming – AI technology is encroaching on art, music and even game design on a large scale. Not to mention that it’s already becoming common sense in the creative world to rely on the power of AI to boost work efficiency.
That is why I believe that “aesthetic sense” is a necessary skill for game developers.”
The only this tells us is that he does not understand how programming works. Or basic math. This equation only holds if you assume that this "fix up and finalizing" brings exactly zero value to...
“Currently, around 80~90% of codes are written by AI and then fixed up and finalized by human programmers. In other words, it means that right now, around 80~90% of games are made by AI.
The only this tells us is that he does not understand how programming works. Or basic math. This equation only holds if you assume that this "fix up and finalizing" brings exactly zero value to the result. It is, of course, the other way around. The fixing up and finalizing is the important part and where the value lies.
Besides that, we have the common issue that the statement does not specify what defines as AI-generated code. Does autocomplete count? Then I could easily see how one gets to numbers like that, with a bit of exaggeration. And it really means nothing.
My feelings on this are mixed. I think developing a sense of aesthetics only matters if you consider the thing you're making to be art. Not every game has to be art, that's fine, but being...
My feelings on this are mixed. I think developing a sense of aesthetics only matters if you consider the thing you're making to be art. Not every game has to be art, that's fine, but being sensitive to subtlety and trying to communicate a delicate idea means that idea has to permeate as much of your design as possible.
That's simply not possible with AI that can only produce the output most likely to satisfy the prompt. It doesn't have an opinion about the idea you're trying to convey, let alone an interpretation, so can't augment it with ideas of it's own in the way that a conscious collaborator might.
I think it depends what your goal is when making a game. If it's making money by producing a product with one or two unique bits, heavy use of AI is fine. But it's hard for me to imagine something with strong authorial vision like Earthbound, Braid, or Papers Please being developed in the same way.
Maybe the key phrase is "on a large scale". Big teams have to worry more about staying viable so making unusual decisions is seen as risky. AI is good at being "vanilla" so it won't to rock the boat, which may explain why it's being used so much.
Side note: this sentiment has been beaten to death twice, but it really gets to me that we're automating fun stuff like games, music and art.
I get your overall sentiment and I think you're correct in some cases but quite wrong in others. I'll try to explain. I have two issues with this: AI can be quite creative and it is possible to...
I get your overall sentiment and I think you're correct in some cases but quite wrong in others. I'll try to explain.
Not every game has to be art, that's fine, but being sensitive to subtlety and trying to communicate a delicate idea means that idea has to permeate as much of your design as possible. That's simply not possible with AI that can only produce the output most likely to satisfy the prompt.
I have two issues with this:
AI can be quite creative and it is possible to use it to create a body of work with a consistent individual style. I think the more modern models like what ChatGPT uses go further and further away from creativity as a side effect of aiming for realism (and maybe safety), but Midjourney still seems to be quite well balanced, and certainly their previous models were. Some old models give you much lower image quality but are very creative when used well (even Stable Diffusion 1.5), you can then refine their output using modern models.
I don't remember any off the top of my head, but there are definitely some for example Instagram accounts that focus on creating AI images (or even videos) that are genuinely interesting and have personality and a consistent style. This takes effort to learn to do well, but it's probably still considerably faster than creating everything by hand, though it's only going to work with some graphic styles.
Games are a specific medium in that some of them can be broken down into very small pieces of assets. Take look at this screenshot from Albion. In my opinion that game is art, I remember the experience of landing on an alien planet and roaming through the town full of strange alien flora, listening to jungle-like but strange sounds in the background and slowly finding out things about the alien society around me. Great story, great setting and great gameplay created a work still celebrated 30 years later.
Yet this game, especially in the 3D view (it's partially pseudo isometric 2D and partially 3D) is just made of individual sprites that are well made, but not really artistic or very special on their own. They only create art when put together in the context of the game. They could no doubt be made using AI by someone capable of some of the things mentioned in my first point, who also has enough aesthetic sense to put them into a functional whole - which is what the citation from OP says. Not every game is like this, but many are.
Side note: this sentiment has been beaten to death twice, but it really gets to me that we're automating fun stuff like games, music and art.
Let me give you a different perspective. I've been thinking about making a game for a long time. I have a somewhat coherent idea of what I want to do, but even though it's quite conservative and made in a retro style (visually not exactly dissimilar from that Albion screenshot above) and I theoretically have all the skills necessary one some basic level, it would still be a huge undertaking for one person and I do not have the time and energy to do it, neither the money to hire somebody else, because in todays market it would be entirely a gamble whether it would break even financially. Most likely it wouldn't.
These days I'm seriously thinking about actualy doing it because various AI models would help me create many of the parts that aren't really artistic on their own, but when used well can be assembled into a coherent whole that is special.
So for me this is not "we're automating away the fun stuff", this is also "individuals now have a realistic path to create valuable works they wouldn't be able to before".
Is it being automated? I'd say a new tool is making iteration quicker, reducing the most time consuming part of game making (the programming). I don't have an opinion on art, or music, but I'd be...
Is it being automated? I'd say a new tool is making iteration quicker, reducing the most time consuming part of game making (the programming). I don't have an opinion on art, or music, but I'd be shocked if they were different.
I have no doubt that a game with strong "authorial vision" (previously: "sense of aesthetic"), like Papers Please, could be made using AI with the right direction and intent.
So, I think I get what the Level-5 guy is saying. Aesthetics will become an even more important and defining factor than the are right now because the ability to code (which some think is a secret art, but it's really not) will have been leveled.
Papers Please doesn't just have "a strong sense of aesthetic," though -- it was lauded by critics because it did something new and interesting that hadn't really been done before, and it's also a...
I have no doubt that a game with strong sense of aesthetic, like Papers Please, could be made using AI with the right direction and intent.
Papers Please doesn't just have "a strong sense of aesthetic," though -- it was lauded by critics because it did something new and interesting that hadn't really been done before, and it's also a very tight experience as a result of being made by a solo developer (and a solo developer who's particularly good even for a solo dev at executing on a strong authorial vision). The aesthetics are probably the least important part of Papers Please when it comes to determining why it's been so widely praised.
AI can definitely make something that looks like Papers Please, and now that Papers Please is out there in the world existing and influencing other works and inspiring copycats, it could probably make something that plays like Papers Please (although probably not without a lot of human wrangling at its current capabilities). But AI is not able to make the next Papers Please. A quite large proportion of humans can't execute a strong authorial vision that doesn't fit well within existing genres and trends, and AI is not remotely close to that level.
Of course! I only referred to its aesthetics because that's what had been done earlier in the thread. Of course that phrase has now been edited to say "authorial vision" which is less catchy but...
Of course! I only referred to its aesthetics because that's what had been done earlier in the thread. Of course that phrase has now been edited to say "authorial vision" which is less catchy but welcome as it encompasses more than looks. I'll be sure to use a quote in future. But aesthetics refers to the original interview.
What I mean, and what I posted/alluded to in another reply, is that if one dev can create a game to a certain standard (aesthetics, gameplay, vision, etc) with a certain set of tools then the same developer can surely create a game to the same standard and have a slightly easier time of it with use of a new tool (could be ai, or a new image editor, workflow script, build phase, etc). And another developer of the same calibre could create a game of the same standard using the same tools.
At the end of the day the developer can choose to achieve the goal with whatever tools they see fit and the result will speak for itself. Lucas Pope is big into tools and I'm sure his games would be worse off and take much longer to develop if he didn't automate and tool up during their creation. I'm not implying he uses ai, I don't care either way, but he does use procedural generation, etc.
Saying this as an artist and solo game dev. I've tried using AI as a tool for making art and was genuinely surprised that it did not actually make my art-making process easier or faster. My...
Saying this as an artist and solo game dev. I've tried using AI as a tool for making art and was genuinely surprised that it did not actually make my art-making process easier or faster. My process is slow and tedious-- I draw in an intricate style-- and honestly expected AI to speed up the process somehow because it's actually kinda good at generating intricate-looking visuals. I was able to use it to do that but it also took a lot of time and effort to get to results that just weren't quite as good as what I can do without it. Even without the ethical issues (which are a dealbreaker for me), it just wasn't a worthwhile tool for me.
I haven't tried it for game dev specifically but I think it'll be the same. I'm the type that wants to make games with a "strong authorial vision" so I don't think generative AI will be a very helpful tool for me there. I actually think AI is less helpful overall in a medium like game dev because you need to make so many different parts come together to create a cohesive experience and narrative. Theoretically you could use AI to generate a lot of assets or bits of dialogue but IMO the difficult part in game dev is actually refining, editing, and cutting away things to make the overall game very cohesive. (As opposed to, let's say, using AI to generate an illustration in a use case where all the illustration needs to do is be pretty.)
Overall, I think a lot of people assume that using generative AI automatically makes creative labor easier and faster. My own assessment is that it can do so if you just need to produce a large volume of generic content... and for some, that is the goal. If the goal is to produce work with a "strong authorial vision", generative AI can be used as a tool but it's not going to make things easier or faster because it's not going to be making all those small and big decisions that go into creating that type of work.
Using AI to write the code seems to be a better usage than making it generating the art assets, IMHO. Art is subjective and has meanings and interpretations that AI doesn’t really understand and...
Using AI to write the code seems to be a better usage than making it generating the art assets, IMHO. Art is subjective and has meanings and interpretations that AI doesn’t really understand and it usually lacks consistency. But nobody is going to care if your code uses switch instead of a bunch of else ifs, or how well you built the underlying architecture.
I haven't tried using AI for coding yet. But I enjoy coding even if I'm not great at it. As you said no one will care about the underlying code as long as my game works well, so it's ok even if I...
I haven't tried using AI for coding yet. But I enjoy coding even if I'm not great at it. As you said no one will care about the underlying code as long as my game works well, so it's ok even if I write spaghetti code. I don't yet feel the need to try AI for it at this time, but open to trying it in the future. I agree that's different from using AI to generate art.
I've not tried it for game art, but I don't find you're experience surprising. Maybe if/when ai art can be more consistent and purposeful it might need to be reassessed. For code (my experience is...
I've not tried it for game art, but I don't find you're experience surprising. Maybe if/when ai art can be more consistent and purposeful it might need to be reassessed.
For code (my experience is mostly creating tools rather than games, even though I'm a GOTY winning game dev) I find that if you have good project organisation and ability to manage changes, it really does unlock the 10x that I previously thought must be exaggeration.
The number are probably greatly exaggerated to jump into the AI bandwagon, otherwise it is fairly terrible. It is saying the value his company provides over the tools they use (assistants but also...
The number are probably greatly exaggerated to jump into the AI bandwagon, otherwise it is fairly terrible. It is saying the value his company provides over the tools they use (assistants but also engines, middleware, etc) is about 10~20%, otherwise I guess they are a marketing firm for those AI generated games.
The entertainment market is fairly saturated already. If games standardize to whatever is cheaper to produce using AI, very many game companies are about to dissapear / get absorved by AI companies (like ms or tencent)
I mean, how could he possibly know about all games? He can't. Maybe he's referring to his own company games, then I'd have to take his word. I find it easy enough to believe.
I mean, how could he possibly know about all games? He can't. Maybe he's referring to his own company games, then I'd have to take his word. I find it easy enough to believe.
GenAI would be a good explanation for how bad Layton's Mystery Journey was, if it hadn't come out back in 2017. This isn't exactly filling me with optimism about the new Professor Layton game though.
GenAI would be a good explanation for how bad Layton's Mystery Journey was, if it hadn't come out back in 2017. This isn't exactly filling me with optimism about the new Professor Layton game though.
Times like these just make me tempted to jump on the AI bandwagon and become one of those hacks who masks their lack of artistic or technical talent with generative models.
Times like these just make me tempted to jump on the AI bandwagon and become one of those hacks who masks their lack of artistic or technical talent with generative models.
Value judgements of AI games aside, I don't believe him. That is a massive number that I think he's just pulling out of his ass and I suspect it's very incorrect. AI, even the reasoning models like o3, are just not very good at building large cohesive projects. At best you'll get a function here or there that actually took less time to fix the AI's mistakes than it would've taken to write. I'm actively working on a project where I'm somewhat heavily using AI to help me out with the big APIs that have a lot of boilerplate (Vulkan mainly). It does kind of okay on generic boilerplate, but almost any time I try to get it to write an "interesting" function based on things unique to my project it fails.
The code output feeling about the quality level I'd expect of a junior engineer that got their instructions via the game telephone doesn't really impact it's usability in other domains though. It'll probably just about completely kill ambient music and VO roles as well as taking a big bite out of graphical creation asset roles. I just don't believe we currently live in a world where 80%+ of video game code is coming from AI because it frankly isn't good enough to build the entirety of any but the simplest games.
Anything not fully self-contained may as well be a mystery to any coding AI assistants that I have used. They are pretty good at "make a function that takes X and Y primitives and manipulates them" or "given this static function, generate some starting test cases".
As soon as it involves a call out to anywhere else, it is a roll of the dice. Maybe it can come up with something if the package is popular enough on SO to have some answers, and then only if the package hasn't updated recently.
I am with you here. I think they want it to be 80-90% AI made but I think they are lying or some.project managers are lying to them.
It can be super super useful though. For example, these test cases for this file were 90% generated by AI with no revisions. That means the LLM essentially simulated the 1300 line file when generating input/output pairs and it did it perfectly.
But I agree they are not a general solution to programming problems. Part of the problem with them is learning when they're a time saver and when they're a time waster.
Yeah this number feels more like "80-90% of games have someone using AI somewhere." Meaning an admin assistant might use AI to help her draft emails, a writer might be using it to break up writers block, a localization team might be using it as a first pass.
My thoughts as well. Even if someone uses AI to generate a skeleton for a code module with custom names, that's far from having AI do the "meat and potatoes" work of game dev.
(And project templates / skeletons have been a thing before AI, but I could just see AI streamlining it a bit more.)
AI is only as good as the operator, and output quality is directly related to input quality. I'm positive that your final assertion is false, speaking from experience.
Complicated games can be built by one person with no assistance. So it makes sense that the same person can create a complicated game more easily using AI.
Garbage in, garbage out also applies to the training models.
If we hold the Sturgeon's law to be true-ish, 90% of all of human creation is crap. Thus, the AI will spit out 90% crap. An LLM has no way of discerning the wheat from the chaff. It will give the most popular thing, regardless of its correctness.
Just look at stack overflow. Rarely is the chosen answer the correct answer, and you'll find a more-upvoted answer that is better, and a 0-vote answer that is now correct in the 10 years that passed since the question was asked.
Are the majority of Gitlab repos putting out genius code? Or are they putting out amateurish/hobbiest code which is functional but sub-optimal?
I'm not quite sure what your point is? That the models are inherently garbage and there's little to no chance for the output to be good? That would be easy to prove incorrect. Of course there'd have to be some qualitative way to judge things. But from what I've seen is quite easy to get good results. The better the prompt/spec the better the results. A mediocre, average, simplistic prompt will naturally get the matching result.
The skill involved in extracting the best output by forming the best input (prompt) is the same as that involved in doing a google search (witness that some people are better at it than others) or even finding something in a reference book. Like hunting for a needle in a haystack.
Key quote from CEO Hino Akihiro:
Here's a longer Famitsu article in Japanese with more info about his Top Game Creators Academy initiative.
The only this tells us is that he does not understand how programming works. Or basic math. This equation only holds if you assume that this "fix up and finalizing" brings exactly zero value to the result. It is, of course, the other way around. The fixing up and finalizing is the important part and where the value lies.
Besides that, we have the common issue that the statement does not specify what defines as AI-generated code. Does autocomplete count? Then I could easily see how one gets to numbers like that, with a bit of exaggeration. And it really means nothing.
My feelings on this are mixed. I think developing a sense of aesthetics only matters if you consider the thing you're making to be art. Not every game has to be art, that's fine, but being sensitive to subtlety and trying to communicate a delicate idea means that idea has to permeate as much of your design as possible.
That's simply not possible with AI that can only produce the output most likely to satisfy the prompt. It doesn't have an opinion about the idea you're trying to convey, let alone an interpretation, so can't augment it with ideas of it's own in the way that a conscious collaborator might.
I think it depends what your goal is when making a game. If it's making money by producing a product with one or two unique bits, heavy use of AI is fine. But it's hard for me to imagine something with strong authorial vision like Earthbound, Braid, or Papers Please being developed in the same way.
Maybe the key phrase is "on a large scale". Big teams have to worry more about staying viable so making unusual decisions is seen as risky. AI is good at being "vanilla" so it won't to rock the boat, which may explain why it's being used so much.
Side note: this sentiment has been beaten to death twice, but it really gets to me that we're automating fun stuff like games, music and art.
I get your overall sentiment and I think you're correct in some cases but quite wrong in others. I'll try to explain.
I have two issues with this:
I don't remember any off the top of my head, but there are definitely some for example Instagram accounts that focus on creating AI images (or even videos) that are genuinely interesting and have personality and a consistent style. This takes effort to learn to do well, but it's probably still considerably faster than creating everything by hand, though it's only going to work with some graphic styles.
Yet this game, especially in the 3D view (it's partially pseudo isometric 2D and partially 3D) is just made of individual sprites that are well made, but not really artistic or very special on their own. They only create art when put together in the context of the game. They could no doubt be made using AI by someone capable of some of the things mentioned in my first point, who also has enough aesthetic sense to put them into a functional whole - which is what the citation from OP says. Not every game is like this, but many are.
Let me give you a different perspective. I've been thinking about making a game for a long time. I have a somewhat coherent idea of what I want to do, but even though it's quite conservative and made in a retro style (visually not exactly dissimilar from that Albion screenshot above) and I theoretically have all the skills necessary one some basic level, it would still be a huge undertaking for one person and I do not have the time and energy to do it, neither the money to hire somebody else, because in todays market it would be entirely a gamble whether it would break even financially. Most likely it wouldn't.
These days I'm seriously thinking about actualy doing it because various AI models would help me create many of the parts that aren't really artistic on their own, but when used well can be assembled into a coherent whole that is special.
So for me this is not "we're automating away the fun stuff", this is also "individuals now have a realistic path to create valuable works they wouldn't be able to before".
Is it being automated? I'd say a new tool is making iteration quicker, reducing the most time consuming part of game making (the programming). I don't have an opinion on art, or music, but I'd be shocked if they were different.
I have no doubt that a game with strong "authorial vision" (previously: "sense of aesthetic"), like Papers Please, could be made using AI with the right direction and intent.
So, I think I get what the Level-5 guy is saying. Aesthetics will become an even more important and defining factor than the are right now because the ability to code (which some think is a secret art, but it's really not) will have been leveled.
Papers Please doesn't just have "a strong sense of aesthetic," though -- it was lauded by critics because it did something new and interesting that hadn't really been done before, and it's also a very tight experience as a result of being made by a solo developer (and a solo developer who's particularly good even for a solo dev at executing on a strong authorial vision). The aesthetics are probably the least important part of Papers Please when it comes to determining why it's been so widely praised.
AI can definitely make something that looks like Papers Please, and now that Papers Please is out there in the world existing and influencing other works and inspiring copycats, it could probably make something that plays like Papers Please (although probably not without a lot of human wrangling at its current capabilities). But AI is not able to make the next Papers Please. A quite large proportion of humans can't execute a strong authorial vision that doesn't fit well within existing genres and trends, and AI is not remotely close to that level.
Of course! I only referred to its aesthetics because that's what had been done earlier in the thread. Of course that phrase has now been edited to say "authorial vision" which is less catchy but welcome as it encompasses more than looks. I'll be sure to use a quote in future. But aesthetics refers to the original interview.
What I mean, and what I posted/alluded to in another reply, is that if one dev can create a game to a certain standard (aesthetics, gameplay, vision, etc) with a certain set of tools then the same developer can surely create a game to the same standard and have a slightly easier time of it with use of a new tool (could be ai, or a new image editor, workflow script, build phase, etc). And another developer of the same calibre could create a game of the same standard using the same tools.
At the end of the day the developer can choose to achieve the goal with whatever tools they see fit and the result will speak for itself. Lucas Pope is big into tools and I'm sure his games would be worse off and take much longer to develop if he didn't automate and tool up during their creation. I'm not implying he uses ai, I don't care either way, but he does use procedural generation, etc.
Saying this as an artist and solo game dev. I've tried using AI as a tool for making art and was genuinely surprised that it did not actually make my art-making process easier or faster. My process is slow and tedious-- I draw in an intricate style-- and honestly expected AI to speed up the process somehow because it's actually kinda good at generating intricate-looking visuals. I was able to use it to do that but it also took a lot of time and effort to get to results that just weren't quite as good as what I can do without it. Even without the ethical issues (which are a dealbreaker for me), it just wasn't a worthwhile tool for me.
I haven't tried it for game dev specifically but I think it'll be the same. I'm the type that wants to make games with a "strong authorial vision" so I don't think generative AI will be a very helpful tool for me there. I actually think AI is less helpful overall in a medium like game dev because you need to make so many different parts come together to create a cohesive experience and narrative. Theoretically you could use AI to generate a lot of assets or bits of dialogue but IMO the difficult part in game dev is actually refining, editing, and cutting away things to make the overall game very cohesive. (As opposed to, let's say, using AI to generate an illustration in a use case where all the illustration needs to do is be pretty.)
Overall, I think a lot of people assume that using generative AI automatically makes creative labor easier and faster. My own assessment is that it can do so if you just need to produce a large volume of generic content... and for some, that is the goal. If the goal is to produce work with a "strong authorial vision", generative AI can be used as a tool but it's not going to make things easier or faster because it's not going to be making all those small and big decisions that go into creating that type of work.
Using AI to write the code seems to be a better usage than making it generating the art assets, IMHO. Art is subjective and has meanings and interpretations that AI doesn’t really understand and it usually lacks consistency. But nobody is going to care if your code uses
switchinstead of a bunch ofelse ifs, or how well you built the underlying architecture.I haven't tried using AI for coding yet. But I enjoy coding even if I'm not great at it. As you said no one will care about the underlying code as long as my game works well, so it's ok even if I write spaghetti code. I don't yet feel the need to try AI for it at this time, but open to trying it in the future. I agree that's different from using AI to generate art.
I've not tried it for game art, but I don't find you're experience surprising. Maybe if/when ai art can be more consistent and purposeful it might need to be reassessed.
For code (my experience is mostly creating tools rather than games, even though I'm a GOTY winning game dev) I find that if you have good project organisation and ability to manage changes, it really does unlock the 10x that I previously thought must be exaggeration.
I see, thanks for sharing your experience!
Ah okay, I think I interpreted a much stronger level of "made using AI" than you intended in your original message, then!
No worries at all!
The number are probably greatly exaggerated to jump into the AI bandwagon, otherwise it is fairly terrible. It is saying the value his company provides over the tools they use (assistants but also engines, middleware, etc) is about 10~20%, otherwise I guess they are a marketing firm for those AI generated games.
The entertainment market is fairly saturated already. If games standardize to whatever is cheaper to produce using AI, very many game companies are about to dissapear / get absorved by AI companies (like ms or tencent)
I mean, how could he possibly know about all games? He can't. Maybe he's referring to his own company games, then I'd have to take his word. I find it easy enough to believe.
GenAI would be a good explanation for how bad Layton's Mystery Journey was, if it hadn't come out back in 2017. This isn't exactly filling me with optimism about the new Professor Layton game though.
Haha! I feel for the Prof Layton fans.
to be fair, we haven't gotten a mainline game in over 10 years. So who knows, even a bad one might be better than nothing... hopefully....
Times like these just make me tempted to jump on the AI bandwagon and become one of those hacks who masks their lack of artistic or technical talent with generative models.