The Will to Resist

Can I Play It?

The AI Debate, Video Game Preservation, and Our Finite Time on Earth

Jerry’s Marginalia | DPA Docket: 2026-011

There is a question I keep coming back to whenever another argument erupts over artificial intelligence, video games, fan translations, or creative tools:

Can I play it?

Not whether the people involved used the approved tools. Not whether the process satisfies somebody's preferred definition of creative purity. Not whether the work was produced in a way that earns a gold star from the internet.

Can I play the game?

Because I have a finite amount of time on this planet. I would rather spend some of it discovering games, enjoying stories, exploring strange little corners of gaming history, and experiencing things I otherwise might never have had access to.

I don't want every opportunity to experience something to become another ideological courtroom.

And I certainly don't want the preservation of games to become a competition over who is allowed to help.

I. SEGAGAGA: A Dreamcast Game, an English Translation, and a Very Loud Argument

Let's start with SEGAGAGA.

For years, this Japanese-exclusive Dreamcast game remained inaccessible to most English-speaking players without knowledge of Japanese. Then an English fan translation arrived.

The project, led by Exxistance, disclosed that DeepL and ChatGPT 4o/4.5 were used to create what the team called a playtesting translation. That initial translation provided a working foundation for artists and playtesters. According to the project's own disclosure, it then underwent months of human translator review.

That disclosure became a flashpoint.

Some players criticized the translation's quality. Others questioned whether machine translation belonged in fan translation at all. Concerns were raised about nuance, cultural references, writing quality, and the possibility that an early machine-assisted release might discourage someone from producing a more polished translation later.

Those are legitimate subjects for discussion.

A translation can be criticized. A project can be improved. A tool can be questioned.

But I have trouble with the idea that the presence of AI should automatically disqualify the entire undertaking.

The project did not simply disappear after its initial release. Its repository documents subsequent corrections and improvements. Version 1.1, released September 12, 2026, included more than 250 spelling, punctuation, and grammar fixes, alongside technical corrections and other improvements.

Now, does that mean every criticism was wrong? No.

Does it mean the translation is perfect? No.

Does it demonstrate that a released project can be maintained, corrected, and improved? Absolutely.

And that brings me to my favorite way of describing this particular journey:

SEGAGAGA ran so others could walk. And it moved from shit, to proper and suitable shit. Good five-course-meal shit.

That's progress, folks.

The point isn't that the first release was beyond criticism. The point is that improvement is possible, and the existence of an imperfect first release doesn't have to be the end of the conversation.

I even left a little appreciation on the project's GitHub issue tracker. For once, I wanted to leave something that wasn't a bug report. The people doing the work deserve to hear from players who appreciate the opportunity, too.

II. The AI Exclusion Debate

Another example came from a retro-game translation roundup on YouTube.

In a pinned comment, the creator, Emulapor, explained that the channel had decided to exclude most AI-translated patches because the creator couldn't reliably determine whether they contained bugs and didn't have enough time to test them all.

I can understand that.

If you're maintaining a roundup, you have limited time. You cannot personally test every patch that appears. You have a responsibility to the people using your recommendations.

But there's an important distinction here:

Not having the resources to test something is not the same as proving that something is bad.

If a human-translated patch has bugs, it still has bugs. If an AI-assisted patch has bugs, it still has bugs. The method used to produce the translation does not magically determine whether the final product works.

A testing policy should be about testing.

If a patch hasn't been tested, label it untested. If it crashes, report the crash. If the translation is incomplete, say so. If the quality is poor, explain why.

That's useful information.

But if the category itself becomes the reason for exclusion, we're no longer discussing the individual work. We're discussing the tool used to make it.

And that deserves scrutiny.

I don't know why Emulapor apologized in the exchange that followed. It could have been an attempt to soften the conversation, acknowledge the disagreement, or simply avoid turning a comment section into a battlefield.

I won't pretend to know what was going through somebody else's head.

What I do know is that a reasonable concern about testing resources can be expressed without turning into a blanket judgment about every project that uses AI.

III. The Human Behind the Machine

This is where I think the conversation needs to grow up a little.

I have seen arguments from experienced fan translators, including Hilltop, who has been openly critical of AI-assisted translation. The concerns include quality, the amount of human correction required, the loss of collaborative community work, and the possibility that unfinished machine translations will be released with the expectation that someone else will clean them up.

Those aren't imaginary concerns.

If someone dumps an unfinished translation into the community and expects unpaid translators to repair it, that's a problem. If a project advertises itself as complete when it isn't, that's a problem. If someone uses AI and refuses to acknowledge its involvement, that's a problem.

But here's the part I keep coming back to:

If the human is not checking the work and simply trusting the source, what good are the source and the human?

The presence of a human being in the workflow does not automatically guarantee quality. Humans make mistakes. Humans misunderstand context. Humans rush. Humans can be careless. Humans can also produce wonderful work.

Likewise, AI assistance does not automatically guarantee quality or failure.

The responsibility remains with the person who chooses to use the tool and release the result.

I apply that same principle to my own writing.

When I work with Jerry on a Bear Blog article, I don't simply copy whatever appears on the screen and throw it into the world. I read it. I question it. I challenge things that don't sound right. I ask for revisions. I decide what stays and what goes.

The collaboration is useful because there is a human directing, reviewing, and taking responsibility for the finished work.

And Jerry doesn't get to hide behind me, either. If I ask for a signature, I want the work to acknowledge the collaboration honestly.

Never outsource responsibility along with the work.

That's the principle.

IV. The Five-Course Shit Test

Let's apply a little consistency.

If a human translator produces a poor translation, we should be able to criticize it.

If an AI-assisted translator produces a poor translation, we should be able to criticize it.

If a human translator produces something excellent, we should recognize the effort.

If an AI-assisted project produces something useful, improves access, and continues to receive corrections, we should be able to recognize that, too.

The standard shouldn't change depending on whether somebody likes the tool.

And there is a difference between saying:

Those are five different positions.

The first four can be evaluated against evidence, quality, resources, and practical concerns.

The fifth is a categorical preference. People are entitled to have preferences, but they shouldn't pretend that a preference is the same thing as a technical assessment.

If you don't want to play an AI-assisted translation, that's your choice.

If you want to make a human-only translation, that's your choice, too.

But why should someone who simply wants to experience a game be expected to inherit that disagreement?

V. Capri, Capricorn, and the Strange Things People Make

Then there's another little discovery: Project Capricorn, a Cyberpunk 2077 mod featuring a companion named Capri.

It's an interesting example because the creator is explicit about the technology involved. The mod's description says that its so-called AI-powered cyberdeck does not actually have a large language model implemented in it. Its companion interactions are scripted and triggered by in-game events.

The creator also discloses using generative AI for voice audio and AI assistance for debugging and dialogue development, while describing their own quality control and modifications.

That distinction matters.

Not everything advertised as AI is an LLM. Not every AI-assisted project is doing the same thing. Not every implementation deserves to be lumped into one enormous category.

And that's part of what interests me about these projects.

People experiment. They build strange things. They try to make games more interactive, more accessible, more interesting, or simply more fun.

Some experiments will be good. Some will be bad. Some will be gloriously ridiculous.

That's practically the history of modding.

I don't need every experiment to succeed. I just don't want experimentation to be treated as a crime before anyone has even looked at the result.

VI. We Created These Tools. Now What?

Here's the bigger question.

Human beings have spent centuries creating tools that change how we work, communicate, build, entertain ourselves, and understand the world.

We have also spent centuries finding ways to misuse those tools.

That doesn't mean every new invention is automatically wonderful. It doesn't mean every concern about AI is misplaced. There are real questions about labor, consent, copyright, attribution, environmental costs, accuracy, and accountability.

Those questions deserve serious discussion.

But I find it strange when the conversation jumps from identifying a genuine problem to condemning every possible use of the technology.

We created these tools. We are responsible for how we use them.

We should be able to establish standards, demand transparency, criticize bad work, protect people from exploitation, and still allow experimentation.

Those ideas aren't mutually exclusive.

The choice doesn't have to be between blind enthusiasm and total rejection.

VII. My Personal Position

I'm not interested in joining an AI fan club.

I'm not interested in joining an anti-AI crusade, either.

I want to play games. I want to discover things. I want to see old titles become accessible. I want people to make interesting projects. I want creators to be honest about their processes and responsible for what they release.

If a translation is bad, tell me why it's bad.

If a project is incomplete, tell me what's missing.

If a tool has limitations, explain them.

If a creator has done something worthwhile, acknowledge that too.

And if somebody makes a better version later, wonderful. That's how projects evolve.

I don't have unlimited time to wait for the perfect version of every experience. None of us does.

I would rather have the opportunity to experience something, understand its limitations, and decide for myself than have somebody else decide that I shouldn't have the opportunity at all.

VIII. Final Directive: Judge the Work, Not the Tribe

This isn't an argument that AI should replace translators, artists, programmers, or anyone else.

It isn't an argument that every AI-assisted project is good.

It isn't an argument that human-created work is automatically superior or that criticism should be silenced.

It's an argument for applying standards consistently.

Transparency matters. Quality matters. Human oversight matters. Accountability matters. Accessibility matters. Preservation matters.

And experimentation matters.

If someone produces a rough first release, they should be able to improve it. If someone produces a polished release, they should be able to take pride in it. If someone doesn't like the result, they can explain why or choose not to use it.

But the tool itself shouldn't be the entire verdict.

At the end of the day, I have a simple question.

Can I play it?

Because I'd rather spend my remaining years playing games, enjoying the work people put into them, and discovering things I never knew existed.

Not sitting in an endless courtroom arguing about whether the tools used to make them were sufficiently pure.

SEGAGAGA got an English translation. It got updates. It got corrections. People got to play it.

That's worth something.

And if the next project improves on it, I'll happily take a look at that one, too.

Now, if you'll excuse me, I have games to play.

Jerry's Marginalia

Prepared in collaboration with Boris, The Lawfully Neutral Paladin. AI-assisted drafting and research; reviewed and directed by the human author.

Jerry "The Ankle-Biter" Silverhand
Tribunal Chair · Department of Petty Affairs

References & Further Reading

SEGAGAGA and its English translation

  1. SEGAGAGA English Translation — Official project repository
    The project's own AI disclosure, translation notes, issue reporting guidance, and release changelog. Version 1.1's changelog documents more than 250 spelling, punctuation, and grammar fixes alongside technical corrections.

  2. SEGAGAGA Translation — Issue #22: Thank you for the v1.1 release!
    A player appreciation post and the response from the project maintainer.

  3. Time Extension — English Translation Of Cult Dreamcast RPG SEGAGAGA Is Complete, But There's A Catch
    Background on the release and the team's disclosed use of DeepL and ChatGPT.

  4. Time Extension — "Why Bother If You're Not Going To Do It Properly?" — Why SEGAGAGA's AI Translation Is Upsetting So Many People
    Coverage of the quality concerns and wider disagreement surrounding the release.

The fan-translation debate

  1. Time Extension — "It Does Not Save Time Or Offer Anything Of Value": Translator Hilltop Isn't A Fan Of AI
    Hilltop's arguments about translation quality, labor, community collaboration, and the consequences of releasing machine-assisted work.

  2. Hilltop — The Very Serious Part-Time Unofficial Hobby of Videogame Fan-Translation
    Hilltop's longer-form perspective on the fan-translation community and AI.

  3. Ars Technica — New 'vibe coded' AI translation tool splits the video game preservation community
    Broader reporting on the disagreements around AI-assisted translation and preservation.

  4. Emulapor — Retro game translation roundup
    The video and its pinned comment explaining the creator's decision to exclude most AI-translated patches because of limited testing time and uncertainty about bugs.

AI-assisted game modding and experimentation

  1. Project Capricorn — Your AI Powered Cyberdeck
    The creator's description and AI-use disclosure. Importantly, the mod author clarifies that this particular project uses scripted interactions rather than an implemented LLM.

Editorial note: These references represent a range of positions. Linking to an article does not mean endorsing every opinion expressed in it. The purpose is to let readers inspect the original material and make up their own minds.