Column: All A.I. Projects Are Digital Asbestos

The views expressed in this article do not necessarily represent the views of all SHIRO! Media Group contributors.

The cost of modern day video game consoles and personal computers have skyrocketed in the wake of A.I. data center construction. Big language learning model companies essentially preordered a majority of globally produced RAM chips for the next couple of years or so. It is sending the games industry into chaos, causing the Steam Machine to cost more than $1,000 USD, and making the sheer idea of a PlayStation 6 and new Xbox seem impractical. Forget current generation game consoles increasing in price since MSRP. We might be at the end of traditional video game console generations, and A.I. is largely to blame.

So I find it incredibly ironic when people prop up A.I. generated retro game projects. Paying £300 in Claude tokens to generate a slop port contributes to the destruction of its own medium. The fact that so many people in the retro scene fail to see the irony makes me feel like I am going crazy.

In the wise words of Mike Muir, I’m not crazy, you’re the one that’s crazy.

The data centers

The technology powering LLM software like ChatGPT and Claude require massive data centers. A large data center can consume about 5 million gallons of water per day, equivalent to a small town. Just 3 percent of the water on Earth is considered fresh water and only a sixth of that is accessible to humans. Droughts and other environmental conditions continue to diminish water accessibility.

A.I. data centers consume a massive amount of power. It is estimated that throughout 2025, these data centers used about 1.5 percent of the electricity produced in the world — roughly equivalent to all electricity generated in Germany.

These data centers are often built near big and small cities, driving up local energy costs and putting a strain on nearby water use. Leaders of these localities will often push forward with the construction of these data centers despite a negative public outcry under the lofty expectation of getting an economic boost.

Not only do these data centers put a strain on resources, they strain the very purses of those responsible for their construction. They are not turning a profit. Leaked financial info from OpenAI shows it lost more than $20 billion in 2025. The cost to produce and operate LLM software outweighs the money they make. They are simply unsustainable.

Virtually all LLM software has been accused of stealing content and infringing upon the work of others to build their foundations.

The generated results

Retro video game content is the focus here. Though I would be foolish if I did not tell you that images, videos and music produced by generative A.I. make Liberty Mutual Insurance ads look like fine art. Countless easily researchable examples show how often A.I. gets stuff wrong. It can tell you things that are demonstrably incorrect, convince people to commit self harm, and even send innocent people to jail.

Still, some praise the benefits of this software, using it to apply for jobs, carry out research and create programs. Claude in particular is often used to generate software. Someone can simply ask it to write a program, often at the expense of “tokens” if it becomes complex enough.

This seemingly rose to prominence right when retro video game homebrew projects really hit a good stride. We have seen coders (without the use of A.I.) port games to other consoles, reverse engineer source codes through “decompilation” projects, and create translation patches for a staggering variety of games. Others have created online lobbies for multiplayer games that do not natively support online.

So… can I just, like, ask the machine to make these instead?

Retro slop

Yes, you can ask the machine to make these. But the results kind of suck.

Some have told me that I do not have the right to critique the use of A.I. in programming since I am not a programmer. As it turns out, some of the people making these A.I. generated retro projects are not programmers either. For that reason, along with the extreme cost to society of this tech, I feel like I can, in fact, critique it.

When I talk about this with people who actually know how to code, they tell me a program that is almost entirely vibe coded is hard to update and debug. Sure, you can get the slop machine to make something that boots on a Sega Saturn and it might even be playable. But unlike human coders, non-coders using Claude cannot effectively fix the issues scattering the vibe code. Say this vibe coder convinces a human coder to try fixing it. That human has to sift through a mess of disorganized and inefficient code, making the process impractical and stupid.

I can understand what it must be like to be a vibe coder. Suddenly, your hands have birthed a program into the world without any coding experience. It must feel like anyone can be a programmer, even you! Well guess what, champ? You can be a coder without A.I. All you have to do is learn how to actually code — just like everyone else.

In the retro coding scene, there are legitimately talented and skilled programmers who see A.I. as another tool. Maybe they write the code out themselves, but have Claude take care of monotonous tasks or debugging. This may result in competent software. But the aforementioned resource problems with data centers persist. Perhaps if this type of coding assistance can be done on local hardware, it would maybe be more ethical. This would come back around full circle to being something that is not really “A.I.” in my opinion. Regardless, if the coding assistance software relies on resource-intensive data centers, then it has no good use case.

Earlier this year, a “decompilation” of GoldenEye released, bootable on modern PC hardware. But it was packed with problems. Turns out it was vibe coded, and many of the problems present in this project seem to imply it was not even properly decomped. Graslu00 wrote, “…so this won’t help modding in any way other than making new code injections easier, but without knowing what part of the code does what (decompilation), there’s not much use for this.”

Famously, a Legend of Zelda: Ocarina of Time PC port hit the internet at around the dawn of the decomp scene. It was recently discovered that the team behind it has been using A.I. to program for years. In their attempts to cover it up, it is now unclear how much of this project was vibe coded and how much of it is not.

Donkey Kong 64 is going through a similar spat. Wes Fenlon wrote a lovely article on PC Gamer about how a human-made decompilation of the game is essentially competing for attention with a separate vibe-coded decomp.

These vibe-coded porting projects are infecting the decomp scene as a whole. The software is unreliable, broken and hard to fix. It is becoming increasingly difficult to navigate these decomps as a video game enjoyer. Time and time again, I will see an announcement for such a project, get excited, then temper my anticipation as I carefully vet it to see if the slop machine was used. If so, how much of it was slop? A little? All of it? Sometimes I will not find any self-reporting of A.I. usage, play the project, then later come to find out it was slop the whole time. Certainly explains the glitches and such.

We have reached a point where I can no longer get excited about any retro project. It feels dystopian. This is no longer fun.

Saturn slop

The Sega Saturn is no stranger to the clankers.

This is my favorite console. The scene around its modern-day happenings is pretty tight knit. Sure, drama happens, like anywhere else. But the hobby always felt like a safe haven from the troublesome issues of everyday life. It got me through COVID, it gave me lifelong friends and opened my mind to creative passions no other hobby has provided. But no, even we are not safe from the onslaught of slop.

I am not going to name specific projects because I do not wish to give them the pleasure of attention.

In my perspective, there are two main categories of Sega Saturn A.I. homebrew projects. One is Slop Saturn Software.

This year alone, there have been numerous Sega Saturn homebrew games that were nearly entirely prompted into existence with Claude, or other such LLM software. Some are ports of games that did not come out on the Saturn, some are modifications of existing games that were nearly entirely vibe coded, and others are original pieces of software that either credit A.I. or show strong hints of A.I. usage.

During the 1993 U.S. senate hearings on video game violence, Tom Zito asked Senator Joe Lieberman if he actually played any of the games they were criticizing. Lieberman reportedly replied with something to the effect of, “No.” In an effort to not be like this, I went ahead and tried out some of these projects and watched gameplay videos for those that were not published.

They are a huge mess. None of them run well.

My baseline for comparison to other homebrew projects is the Unreal port tech demo from XL2, which was not vibe coded at all. This brought the first two levels of a 1998 first person shooter to the 1994 Sega Saturn in a way that took advantage of its unique hardware. It is creative, it runs super well and looks amazing. This work rivals commercially released 3D games on the console.

One of the more recent slop ports brought a PlayStation classic to the Saturn, a port that was previously promised by its developers back in the 1990s. The port was canceled in 1998, citing hardware difficulty. It also happened to be during the height of the console’s financial decline.

This recent slop port, largely constructed with generative A.I., is littered with problems and is prone to crashing. But it is praised as proof that it was possible to run this game on Saturn hardware all along.

Of course it could run on Saturn hardware. Duh! The developers just chose to stop optimizing it on Sega’s failing console, putting it on the upcoming console instead. Unreal, a 1998 FPS benchmark, can run on Saturn without the RAM cart. So no sh!t, Sherlock, you can absolutely get an adventure game with static backgrounds to function on this thing. In fact, some other person already started working on a port of that game years ago without the slop machines. Open your eyes.

The person who posted the aforementioned Saturn port, largely built with A.I., does admit that it has issues. The real test of whether A.I. is useful will be how well this person can fix those game-breaking bugs. Will they have to spend hundreds more dollars in Claude tokens? Can they feasibly fix it by hand? Will the end result meet our expectations? If other retro slop projects are setting an historical precedent, then you already know the answer.

Another unnamed example are controller mods for pre-existing commercially released games. Using the slop machine is unnecessary here, since similar peripheral mods were able get injected to other games without A.I. The scene already did this multiple times without slop, we do not need it.

NetLink mods for other games using A.I. also have cropped up. But if Knight0fDragon can get online multiplayer to work in Shining Force III without A.I., then guess what — we do not need A.I. for this either.

Some of you disagree with me. A lot of you think these are great use cases for programs like Claude, and will tell me that I am wrong. You are going to play and enjoy these projects despite me preaching to you about why they are kind of bad.

To you, I ask, is it worth it?

Is it worth allowing these projects to set the bar lower than the high standards set by non-vibe coding hackers? Is it worth the buggy mess that will be prohibitively hard to fix? Is it worth the expense of “tokens” to get these things put together? Is it worth the immense cost to society brought on by these inefficient and monstrously unsustainable data centers?

This brings me to the second category of Sega Saturn A.I. homebrew projects.

Translation patches that just use Google Lens for you

To me, these projects are by far and away the most offensive.

Have you ever watched Pink Floyd’s Live at Pompeii? It was filmed in 1971, right in that pre-Dark Side post-Barrett era. As its title suggests, the band played a live concert at the ancient Roman amphitheater in Pompeii. Those on the film crew were the only audience members. Shots of volcanic mud and statues excavated from the city ruins were interspersed with the performance. Wide views of the four playing with a full set of stage gear to an audience of practically no one were striking to say the least. The song Echoes, which usually lasts more than 20 minutes, was played at the amphitheater. Throughout this journey, the song calls upon the concept of loneliness while clashing directly with how interconnected the human condition is throughout time, melodically ascending from a single piano note into utter chaos, managing to bring itself from the ashes and back into order as it closes. To me, this performance illustrates what art does and what art is. A natural creation, be it intentional or otherwise, will serve to communicate with you even if it is no longer organically alive. When someone says, “man this art really speaks to me,” it is pretty much speaking to them. Can something provide this same human connection if it was never alive in the first place?

Each video game is a work of art, but if you made it this far into this article, you probably did not need me to tell you this. Within these interactive creations on the Sega Saturn, it is far from uncommon to find a literary artwork woven into the game. These often tell stories, make characters come to life and convey a dynamic range of emotions to you, the video game enjoyer.

When these games get translated, mechanically converting them to another language is not enough. They must convey the same artistic impression to the player. Its characters and story need to make you feel the same way as they do to those who understand its source language. Working Designs was often criticized for warping dialogue with western pop culture references and unrelated jokes in its own English translations. For decades, the scene has remained sensitive to the issue of localization. Be it fan-made or official, hardcore retro fans are quick to scrutinize such work that may skew the original artistic direction in any way.

Machines cannot effectively convey the human condition, and certainly cannot convert the artistic direction of an RPG to another language.

Even before A.I., plenty of people have used auto translators to play a Japanese game in another language, fully aware of potential accuracy issues. These tools are abundant. At the same time, many fans have created translation patches for games, doing their best to faithfully bring the whole of the experience in a way Google Translate cannot.

Tom of dateemups wrote a very good perspective on this issue in the fallout of the Segagaga debacle, talking about attempts to “represent the work and mindset of somebody else with your own words.” That article, written by a professional localizer, masterfully lays out the danger of using machine translation when converting a video game to another language. Even if the machine output is reviewed and edited, its use inherently warps the original vision of the text.

Despite a cunning ability to target ads after spying on us with mics inside televisions, the machines will never fully understand us and they will never understand what we create. Not in a trillion years.

A patch that uses machine translation is insulting to the player and the original artists. My open letter to machine translation patchers: Congratulations. You just spent an inordinate amount of time doing something I can do by holding Google Lens up to the monitor. On top of this, you have also managed to do the equivalent of an Ecce Homo restoration to a video game. Good job.

Even games with a small amount of text are prone to being botched by machine translation. Take the Sega Saturn game Steamgear Mash, for example. Its main character does not speak much, but uses a lisp when dialogue appears. It was recently shown that machine translation failed to pick up on this detail.

I witnessed a heartbreaking moment in the SHIRO! Discord server not too long ago. The discussion of A.I. translation patches came up, as it so often does. Someone said, “Meh as long as the translation is satisfactory and accurate I don’t see much of a problem with it.” A prominent longtime Saturn translation patcher replied, “High quality localization efforts are a core part of this community. This makes me sad to hear.”

That person is right. High-quality localization has been a major part of the recent Sega Saturn renaissance. Machine translations managing to garner the same amount of attention only serve to end this renaissance, showing that we as a scene have surpassed our peak. I would like to think that massive translation undertakings like the Sakura Wars games, the Bulk Slash dub, the Shining Force III scenarios and the re-translations of Working Design games help define the Sega Saturn scene. Nowadays, controversial and incorrect machine translations are giving us a terrible reputation.

So how do we move forward?

You do not have to support or play games that are built with A.I. Ignoring them is probably the best thing we can do. But it is clear that as long as A.I. software is made accessible to the average person, these projects will continue to be created.

I wrote this knowing full well it would make a lot of people in the scene somewhat mad at me. Might even burn a few bridges. That’s fine. But I am tired of pretending that this sort of work is acceptable. It has profoundly changed the retro scene and turned the Sega Saturn community into a cesspool.

The SegaXtreme Saturn Homebrew Showcase, something I am usually excited for, had A.I.-driven entries. They gave me a sickening feeling with something that usually makes me endlessly happy. My contest roundup video in 2027 will exclude all A.I. projects.

Fun fact: I actually hate talking about A.I. Several people have asked me for my thoughts on this, so here they are, all in a cohesive article. Now I never have to talk about this ever again. Total sidebar here: Jason Graves made a video about people using A.I. to make retro game YouTube videos and it is pretty hilarious.

Our reporting on these things has been put into the spotlight by both sides of the debate. Internally, I have brought up this topic many times with the SHIRO! team. Personally, I do not think A.I.-driven retro projects are newsworthy at all, and you will never see my name in the byline of those stories. But we pride ourselves in not having a single boss. We are a team of people archiving the goings on of the Saturn scene. To the credit of my colleagues, there are a number of slop Saturn projects they have opted to not report on.

Our own privateye has a running list of the Saturn translation projects that are out there, and any that use A.I. are marked.

If I were to sit here and say that I am upset with everyone who plays these things, I would be lying, since I just do not have the energy or bandwidth to get mad at that many people. The people who create them certainly make me less than happy. There are a handful of projects that may have an honest human text translation, an honest coder inject the text, but maybe use a little A.I. to help clean up their own programming. Again, this calls back to the previous issue of any software using these data centers being harmful to society. It is crucial that such coding assistance be operable on local hardware.

I might lose a few of you to the following thought. I do think there is a path forward to redemption for anyone who has created retro slop. Deep down, I believe these people do have a genuine drive to add something of value to the retro gaming scene. This drive can be used to learn how to code properly, or maybe learn a new language with their brains. If you are a vibe coder or vibe translator, I implore you to learn how to do this the correct way. You are better than this.

The people I am most upset with by far are the architects of these big generative A.I. programs. How is something so resource-intensive allowed to run rampant around the world when they have proven to be ineffective in many ways, harmful and unprofitable? It is mindboggling to me. It has affected every aspect of my life, causing loved ones and colleagues to lose their jobs to hyped-up promises, infecting every piece of software I use at work and at home, clogging my own hardware with bloat and even managing to worm its way into my niche little hobby that was always “safe.” I am absolutely sick of the tech currently being marketed as generative A.I. and I want it to go away immediately. My life was better without it.

There are people in elected positions and people running for elected positions who are advocating for more guardrails and restrictions against these programs. Be aware of what is going on in your local area and make your voice heard if you can.

Ever since this technology hit the mainstream in 2022, the cost to run A.I. data centers continues to be greater than the amount of money they make. Despite charging for extended use, it is a measurably unsustainable business model. So at some point in the not-so-distant future, the A.I. industry as we know it will collapse. Its software will likely stick around in some fragmented way, though using it may become prohibitively expensive to the average consumer. When that happens, we may finally see a slowdown of slop projects in the retro video game scene. But its impact and harm will remain. In its wake, we will be forever left to sift through digital asbestos.