John Carmack on AI in game programming
twitter.com
twitter.com
Not having no value, but being of less worth to investors and companies to invest in. This is simple fundamental economics, since game prices are not growing as fast as their input costs. For example, I spent $30 for Atari video games in the 1980s and it was a lot less expensive to produce. That game would cost $90 today with inflation.
For a comprehensive breakdown, see https://www.gamesindustry.biz/are-video-games-really-more-ex...
If your costs are increasing and you can't raise your price then your industry is being commoditized, or at least in a real quandary about how to move forward. AI could be a way to slow the huge, up-front costs that go into AAA games and help limit the risk to making new ones.
If this subject interests you, there is a great long-form interview with Matthew Ball on Stratechery: https://stratechery.com/2025/an-interview-with-matthew-ball-...
Anyway, Carmack is right on the money on this one.
Speaking of $90 - Nintendo recently announced the Switch 2 and physical copies of the games will be just that, $90.
And the Nintendo Switch has sold 5x as many consoles as did Atari. Likely a similar scale for games sold. Nintendo very likely makes more in total than Atari did, even with lower prices.
The main surprise is that they can bs surprisingly lean with their core development teams to this day. Apparently Super Mario Wonder had a core team of around 20 devs.
Economics that are not that old and that have been reinforced via a 'tailor made' customer culture. Marketing and business culture fucked up consumers beyond any recognition, then quantified them and keep optimizing via peer-flagellation and sociopathic feedback loops.
Economists and leadership's ways are ugly.
A single person can make an game. In fact, there's a chance a single person can make a good game. The problem is that the correlation between good and successful is very weak. You basically need a massive marketing machine to break a new IP. Games cost a lot to make but that's by design.
The strategy that AAA studios apply is to go as big as possible and cast a wide net because you can't put a dozen smaller games on every Mountain Dew can. AI is not going to change that.
Making games a bit cheaper is always a plus but the game will grow to fill the budget. It will not mean more games at the top and it will probably not even mean less cost to the studio.
AI tooling will be used but I doubt it will change the way blockbuster games (or movies)_operate beyond the usual progression of industry tech.
What those games, games like Stardew Valley, Undertale, or Animal Well can't do is create an endless income stream in the billions, and right now that's the goal of some big "AAA" publishers. They don't want to make good games, they don't want to make any sort of game really, they want to create an addictive platform for further transactions. They want GTA Online, F2P Gachas, and E-Sports.
They don't want to make Stardew Valley no matter how many units it sells, they want to make the next Fortnite. As a result they keep dumping absurd amounts of money into really sketchy projects (Concorde, Anthem, etc) and they can't seem to figure out why people aren't biting. When in doubt if you're an MBA and you see money flying out of the door and the expected return isn't being generated, it must be almost instinctive to raise prices and lay people off.
Capcom and Bandai Namco had their struggles, but overall had a similar trajectory. But many american companies don't want to think long-term portfolio anymore. They want pump and dumps. The only potential solace of such crashes is that those pump and dumps might stagnate as well.
I think in the context of parent comment, it might be true that games just isn't a great business. Gamers don't like low-risk AAA slop. Adjectives for games such as good, successful, praised, lucrative, don't align each others well. Deltarune(Undertale 2) is to launch on Switch 2 for $10 or thereabouts, literally built by one guy, most likely without any use of AI, and it'll undoubtedly sell containers after containers of Switch 2 units totally irrespective of incoming tariffs. That almost make me feel sorry for AAA title publishers such as EA and Epic. I can only assume that they resemble insurance companies than any others.
Of course there are some exceptions. Like Balatro selling 5 million copies (1 developer + 1 composer), the Touhou Project series (solo dev), Slay the Spire (two devs, ), and Terraria (11-person studio, 9'th best selling game of all time with 60,700,000 copies sold as of November 2024).
The big studios seem to like spending a ton of money on games, but there continue to be new IPs created without a massive budget or marketing campaign.
And while I don't have anything against copy machines per se, that's not how it's being sold to the public. The public is being told this copy machine is a really good power tool that can do lots of things. So what creatives are hearing is "your work is interchangeable with a slightly smarter copy machine, so stop paying creatives and just rip them off".
Anyone who argues that LLMs are "just stochastic parrots" fundamentally doesn't understand what neural networks do. The power does not come from sampling from a distributions over words but from the multi-dimensional representation. And it is that that enables LLMs to be more than just mechanisms that produce copies of material previously seen in training.
You're still just saying that while stealing a lot from one person is wrong, stealing only a little but from many people would be OK. And the people who are stolen from don't agree with you.
Which seems to be exactly what these models are doing
Satisfactory the game didn't "steal" from Factorio - they took the idea of an automated factory game from them, brought it into a 3d world instead of 2d, and pulled in some other ideas, perhaps from other games. This is a hell of a lot closer to stealing than an LLM taking 0.001% each from 100000 images to create something new.. and yet no-one would ever call it such
It's disturbing because it's automated on an insane scale, not because of a solid legal ground
The only valid text makers are humans.
These aren't novel situations. We have centuries of case studies, precedent, and cultural osmosis giving us legal and de facto means of what we feel and say is "stealing intellectual property".
It's not like tribes and families also didn't keep secrets to maintain some power or as matter of trade. We call them "trade secrets" now but those go back before recorded history.
In addition, consider thar back the we had actual communities, unlike now. Bring those back and we can talk about tearing down copyright.
That game has already been made, and even variants of it.
Put an iPhone on the dashboard of a car and simultaneously record steering wheel, pedal, and gear positions, and you've got a photorealistic driving simulator of wherever you were driving that car.
Strap it to a bodycam and pair with annotated recordings the wearer's motion as if it were control inputs, and you have a photrealistic first-person simulation of whatever they were doing, be that paintball, skating, US Army urban combat training, or birdwatching.
Feed it sports TV annotated with player motion, golf, tenis, cricket, soccer, etc.
> I think you are misunderstanding what this tech demo actually is, but I will engage with what I think your gripe is — AI tooling trivializing the skillsets of programmers, artists, and designers.
By the time he got to "power tool" he was talking about something else entirely.
Separately, whether AI models ought to owe credit or compensation to the data used to train them is an interesting and nuanced debate.
As we've seen, poke an LLM enough and it may simply just spit out a near identical recreation. As of now, it definitely proves that this data isn't just "seen". they very much store it on their databases, and we should treat it as such.
>whether AI models ought to owe credit or compensation to the data used to train them is an interesting and nuanced debate
There's nothing nuanced about it in my eyes. Especially if some artists explicitly do not want their art trained upon. Not even 3 years ago Microsoft won a court case via LinkedIn regarding the scaping of their website data. How is this phenomenon any different?
Unless we want to destroy Copyright as we know it, it's pretty cut and dry copyright infringement. But we need to tear it down first and just say we don't like current laws.
“I think you are misunderstanding what this tech demo actually is, but I will engage with what I think your gripe is — AI tooling trivializing the skillsets of programmers, artists, and designers.
My first games involved hand assembling machine code and turning graph paper characters into hex digits. Software progress has made that work as irrelevant as chariot wheel maintenance.
Building power tools is central to all the progress in computers.
Game engines have radically expanded the range of people involved in game dev, even as they deemphasized the importance of much of my beloved system engineering.
AI tools will allow the best to reach even greater heights, while enabling smaller teams to accomplish more, and bring in some completely new creator demographics.
Yes, we will get to a world where you can get an interactive game (or novel, or movie) out of a prompt, but there will be far better exemplars of the medium still created by dedicated teams of passionate developers.
The world will be vastly wealthier in terms of the content available at any given cost.
Will there be more or less game developer jobs? That is an open question. It could go the way of farming, where labor saving technology allow a tiny fraction of the previous workforce to satisfy everyone, or it could be like social media, where creative entrepreneurship has flourished at many different scales. Regardless, “don’t use power tools because they take people’s jobs” is not a winning strategy.” - John Carmack
- Technological advance
- Political and economical fallout once AI starts replacing a meaningful amount of jobs, and quickly
Me? I'm scared. That's it.
Which is what we have, hence the problem.
Yes, AI has the potential to screw things up royally. But do not mistake its' exacerbation of symptoms as the true illness.
What do you mean, is the pool really shrinking, worldwide? I must have missed some statistics?
When you click a button in Unity or Roblox or whatever to generate a new texture, the thing that gets generated comes from a model that could not have been built without using IP. But because it all got chucked into a blender and turned into an anonymous slurry -- and because AI is a politically important growth industry -- the people whose work went into the slurry will not benefit, at all. They'll never see a dime, while the companies selling the slurry will get billions. A lot of those people are the exact ones whose job will be replaced, which is extra painful when you know it was your own work that was used to replace you.
Although in a sense it's pointless to bring up because that milk is already spilt, and it ain't gonna get back into the container.
And actually, the problem is not that my neighbor who’s passionate about video game design but makes bad games. I’m glad social media is full of that. It’s the highly capitalized content farms that flood the zone.
There was similar criticism to yours when the printing press was invented.
That was the relationship we had for some 200 years. The machine automates and accelerates the predictable labor, and the creative heads composes and conducts the tasks.
As of late, we seem to want to flip the script under the guise that "we can make more content" without considering the impact on the creative process. Not how to drive the output, but just how to make more and more. Creativity be damned. That's the issue.
"AI tools will allow the best to reach even greater heights, while enabling smaller teams to accomplish more, and bring in some completely new creator demographics"
>You can indeed let the AI output stuff automatically but it's as interesting as stock photograp
Yes. And stock photography had agreements with the photographer. Much of the art scraped explicity said not to. It's not interesting and potentially illegal. The worst of both worlds.
>Do you disagree with Carmack's assessment?
I left a top level response somewhere, but overall, c. 2025 : not really. It feels like more hype generation, and this isn't the first time he's had to be a hype man. Maybe one day, maybe before or after we get convinent VR/AR hardware. It'll be a while.
And no small team is touching this stuff whike the courts battle about this. Come back to me in 5 years when the ethics is settled and the tech might be more ready. It's pretty hard to even look 5 weeks ahead in these times, let alone 5 years.
Sturgeon's law. 90% of everything is, and always will be, drek.
There's no incentive to fix it because they control the information and profit off the situation. The system is working as intended.
I can see games being similar, maybe a few creative people invent new genres never seen before, or mix elements in creative ways with AI tools, but I'm guessing it will more likely mean a lot of slight variations of games that bring in money churned out quickly.
Pair this with the fact that most people settle their musical tastes to be in line with when they are experiencing the most emotionally significant time in their early lives (high school for some, college for others, etc) and the result is an assumption that
A) What they encounter forms an overall opinion of "all" new music despite being the tip of the iceberg and
B) It's not as good as what they grew up on
Another little cut on why people are trusting big tech less ad less.
I'm sorry, but this seems exactly like what every single generation says. I'm pretty sure that I read some similar quotes aimed at the Beatles back in the 60s.
https://www.theatlantic.com/ideas/archive/2022/01/old-music-...
A lot of what's happening is imo is just there's so much more competition for attention, and music has lost value as a cultural force beyond the music itself for younger people than it used to have, when young people would develop parasocial relationships or "crushes" or build their self-identity on pop stars and their fandom. A lot of this power culturally has been balkanized.
Speaking of the music itself, once the economic incentive is removed to create great songs (we are there already) we will get fewer great songs. That force is countering the fact that we have more tools and distribution channels than ever to create great songs, and those two forces are in opposition and to me, it seems like the fewer great songs force is winning (subjective).
You both listen to so much music and have a palette that is so wide-ranging that you can somewhat objectively judge it as not having gotten better?
It's always been that way, especially before mass-produced recordings were available.
The "Beatles" period was partly because you had to choose wisely what you were going to spend your limited dollars on when you went to the record shop.
The difference today is that recording technology is cheap, and, with streaming, you don't have to "choose wisely" when you go to the record store. Now all of the mediocre artists, who were generally excluded from the "Beatles" period, can get on streaming platforms.
BTW: There's awesome music coming out today. It just takes time for word to spread.
Yes, I think I would make that argument.
But also I think I’m making a different argument. The game market already seems to be pretty flooded. I think many people see it as “we get more great indie games”, but I wonder if we don’t get less. That it creates more of a bipolar distribution. Triple-A relatively unaffected, and more vaporware games but it hollows out the middle.
I also think that “we objected to earlier progress and it turned out ok, therefor all objections to progress are bad” is a logical fallacy.
Often seen as job simulator games because they're so formulaic.
Sometimes you see unimaginative remixes like Palworld.
When you think of Youtube, you might think of pretty decent stuff but you have to remember the absolute slop like Spiderman/Elsa videos, crude ripoffs and near porn gets many, many more views than the good stuff.
I'm not saying strict gatekeeping is necessary, but open systems are an absolute minefield for humanity. That's clear to me after the last 10 years or so.
That said, we've already decided that truly open systems are not viable. That was settled years ago. Now even 4chan has rules.
Right now we have algorithms that promote terrible content based on automated metrics that don't make the good stuff any easier to find. Especially considering the consolidation of platforms into a few hegemonic sites.
In that sense we already have gatekeeping, it's just a terrible approach to it.
And for the record I mind the slop because of how easily it drowns out and eventually overtakes the good stuff. Even good, educational content feels the need to cater to the algorithm to survive.
We're not at a tipping point yet, but it certainly feels we're headed in that direction.
Algorithmically promoting tastemakers might be a solution. Pushing the good content to the top and burying the slop.
Sometimes the new thing is also unpolished and needs work, perhaps even a second artist to pick it up and make it good...
I very much mind the theft. Even Youtube is offering ways to just "generate new videos" which is done by outright stealing from other youtubers.
With hindsight, one of silent assumptions very common of both pro- and anti-AI arguments up to this point was as follows: because AI is potentially superhuman in arts, the existing selective pressures could inversely punish desired creativity if AI outputs were not unfairly treated. Pro-AI arguments assumed that warping the system is wrong, anti-AI assumed allowing the system go down is worse. IMO, superhuman AI didn't happen anyway, and so now AI slop problem is just spam control problem.
No one was ever against punishing spams and unethical actors. It's fine and safe. We're at a postmortem phase during which we'd patch the bugs in laws and law enforcement that allowed AI companies do outlandish hacks like torrenting books and redistributing it as lossy compressed 500GB GGUF.
A similar thing was discussed in the past, checkmate.
We have this commonsense whig-history where everything became a little bit better century by century. But one could argue that the printing press makes propaganda possible. From that you get thought control by the elite. Now you ironically get an elitist outcome, or an elite-benefiting outcome.
Was the printing press a net good? Maybe it’s more complicated than “more books good”.
Beyond that I don’t know what there is to say about our so-called democratizing technologies when the discussion gets reduced to—and never goes beyond—those very slogans about them being democratizing. Yes. We are now free to repeat the same slogans that everyone else repeats.
I get the feeling that Balatro is not a well coded game. It was coded by one person. Yet it's a great game. The world will be better with more Balatro's.
This is a question I’ve been pondering post-2000, where are all the new Kevin Smiths?
Where are all the great counter-culture indie movies?
That Market has certainly not exploded post YouTube the way you might think.
This could be the golden age of the reviewer. Yahtzee could make millions!
But older entertainment doesn't instantly make it better. Will there be more crap? Sure, but that's always been the case. But that also means more stuff of quality.
That could mean it's hard to find the quality stuff, but that's a different issue entirely, and one mostly solved with old school stuff (reviewers, just find reviewers you mostly agree with).
But in the end, good games are still released every year. And many/most of these good games wouldn't exist if the barrier to entry wasn't lowered.
Actually the opposite of what you think might come to happen. If you lower the barrier to entry for creating a movie, we might see a lot more creative ideas coming from people that would never have the chance before
I used to type in program source code from magazines and had no idea what I was doing until something broke then I had to fix it. If I am honest, that was how I learned how to code.
AI will either teach that kind of thing to the new generation, or coding will become irrelevant. Either way, I think that’s good.
But I still don’t want my bank or airplane guidance software using it.
But when project becomes too big, it can also become discouraging.
Not much otherwise.
Given the importance even today of understanding assembly compilation towards low level game performance, I'm surprised he'd say this. It wasn't rendered obsolete, it was abstracted away from most of the stack. Meanwhile you need to understand assuembly even more intimately to look under the hood of a modern day game or game engine. esoteric does not mean outdated.
>AI tools will allow the best to reach even greater heights, while enabling smaller teams to accomplish more, and bring in some completely new creator demographics.
Okay, I'll believe it when I see it. You said the same about Oculus. I'm not even doubting that VR will evolve to a revolution one day. But technology's march can be slow at times.
And that's one of my top 3 problems; I think like VR's hardware barrier, AI is hitting barriers on how iterations with LLMs work. It seems industry's been brute forcing it and we clearly hit a wall already. But we haven't rethought the approach yet. We're just promising and prpmising.
>there will be far better exemplars of the medium still created by dedicated teams of passionate developers.
Depends on how legal proceedings go. At least Quake is Open Source and kinda free ( I think). The vast majority of games probbaly won't let you train that easily. They spent decades making it as hard as possible to back them up, after all.
>“don’t use power tools because they take people’s jobs” is not a winning strategy.
If industry is going to fire you anyway, it's the only move. If industry worked on fostering workers instead of replacing them, they wouldn't be worried. Instead it's finally starting to unionize to protect itself.
I doubt you'd get the easy moddability into an engine with AI, one of the huge points for the longevity.
Recently I (Gemini 2.5?) created an android app that calculates options prices given different sets of parameters set by the user. The hardest part of doing that is doing all the android studio leg work to get the pieces in place to eventually compile the code into an app and put it on your phone.
No real reason that cannot be automated to "Paste the text the LLM gave you here, and the app icon will show up on your phone in a minute" right now. I'm almost positive it is already being worked on somewhere.
Except that the quality of mass-farmed, labor-saving tech produced 'stuff' is approaching a level of literal shit and the methods have poisoned air, water and soil to a horribly dumb degree.
Will the same happen to the analog/digital soil, water, air in SWE, game dev and content creation? Likely. It started a while ago, before the big AI boom and that's what young creators and devs see in their youth and get inspired, stimulated and motivated by: toxic, low quality shit that they have to shove down their throats and into their minds. "I can do better" is not something we see a lot anywhere; not in cinemas, not on the news, not in SaaS and sure as hell not in VC culture or portfolio capitalism.
Next generations brains are wired and minds nourished by the current environment. And we've been fucking up for a while, even if we leave out politics, news, culture and how we systemically perceive, portrait, never defuse and always escalate conflicts in a slow burn fashion.
All that wires brains, reinforces behaviors and thought patterns and perception and nourishes minds.
I like the comparison to farming. The end game is displayed in movies and games alike and it's always dire, dry and satisfies no one.
There's no case against AI, though, but a momentous one against experienced, educated people who have witnessed the shit show long enough and don't need predictive algorithms to know what's coming.
How silly it is to 'let it happen', to 'let it be'.