Jerry’s Marginalia — The Tool Is in Our Hands
There is something I've noticed about the current argument surrounding AI-assisted fan translations.
People will spend years saying they want a particular game translated.
They'll talk about how obscure it is. How fascinating the story looks. How unfortunate it is that the game never received an official English release.
Then somebody actually attempts to translate it using AI assistance.
And suddenly, everybody becomes a quality-control inspector.
Not necessarily because they've played the game.
Not necessarily because they've examined the patch.
Sometimes, apparently, it's enough to see two letters.
A.
I.
Well, shit.
So, Why Are We Bitching?
Let's establish something.
I'm not claiming AI translations are perfect.
They aren't.
They can misunderstand context, butcher dialogue, miss cultural references, produce awkward sentences, and occasionally generate something that makes you wonder whether the machine has ever encountered a human being.
That's why testing matters.
That's why editing matters.
That's why somebody needs to actually play the damn game.
But here's the question I keep coming back to:
If an AI-assisted translation makes a previously inaccessible game playable, why are we so determined to dismiss it before examining what it accomplishes?
Does an imperfect translation automatically have no value?
Does the fact that a machine assisted with the work erase the effort involved in extracting text, modifying files, building a patch, testing the game, and making it available to other people?
I don't think so.
A playable translation isn't necessarily a finished localization masterpiece.
But it is a door that wasn't open before.
And sometimes opening that door is the most important first step.
The Freddy Experiment
I had a little experiment of my own.
I wanted Freddy to dance.
Not just any dance, either.
I wanted him dancing to "Dreaming" by that guy from New Jack City.
Yes, that was the idea.
Don't ask me why. Sometimes the brain produces a perfectly reasonable idea that sounds absolutely ridiculous when spoken aloud.
The AI tools couldn't quite get Freddy's movements right.
The result wasn't what I had imagined.
I didn't finish the project or follow through with MinnMax, either.
But I still went ahead and experimented.
Why?
Because the tool gave me an opportunity to attempt something I otherwise might not have bothered trying.
It didn't work perfectly.
So what?
I learned something about its limitations.
That's part of using a tool.
You experiment. You evaluate. You adjust. Sometimes you succeed, sometimes you don't, and sometimes you end up with Freddy dancing like he's experiencing a minor software malfunction.
That's life.
CODEC BABEL and the People Actually Trying
This is partly why I've become interested in CODEC BABEL.
Phil is working on something I find genuinely interesting: making Japanese games accessible to people who otherwise might never get to experience them in English.
He may not have a massive YouTube audience.
He may not have thousands of people cheering him on.
But he's putting translations out there, working on games that deserve attention, and exploring ways to make this work possible.
And yet, the moment AI enters the conversation, some people won't even consider the work.
I don't have to agree with every translation choice.
I don't have to pretend every release is flawless.
I don't even have to like every result.
But I can respect the attempt.
I'd rather see somebody take a swing at an obscure game than watch another decade pass while everyone waits for the perfect circumstances, the perfect translator, and the perfect technical team to magically appear.
One game I'd particularly love to see become accessible is the Blood+ game on PlayStation 2.
Even if AI assistance is part of getting there.
Especially if it means somebody can finally experience a game that has been sitting behind a language barrier for years.
And if the translation needs work?
Then let's work on it.
The Tool Doesn't Get to Make the Final Decision
Here's where I think we need to be careful.
Trusting a tool doesn't mean surrendering your judgment to it.
If a translation produces nonsense, question it.
If a line doesn't make sense in context, investigate it.
If the game crashes, debug it.
If the localization needs improvement, improve it.
And if somebody produces a better human translation later, wonderful.
That doesn't make the earlier effort worthless.
It means the work can continue.
Humans should make the best use of the tools available to them, but they also have a responsibility to examine the results.
Don't blindly trust the machine.
Don't blindly dismiss it, either.
Look at the work. Test the work. Understand the limitations. Make the best decisions you can with the tools in your hands.
That's not an unreasonable expectation.
That's using technology responsibly.
The AI Bubble and the Opportunity
I don't know where the AI industry will be five years from now.
Maybe the current enthusiasm will cool down.
Maybe certain companies will disappear.
Maybe some tools will become unaffordable, inaccessible, or obsolete.
Maybe the entire industry will look very different.
But right now, these tools exist.
People are experimenting with them.
They're finding new ways to approach old problems.
And I don't want to spend the entire opportunity arguing about whether the technology deserves to exist.
I want to learn.
I want to experiment.
I want to understand the tools, their strengths, their weaknesses, and what can be accomplished with them.
If the bubble eventually bursts, I'd rather be someone who spent that time learning something useful than someone who spent it shouting from the sidelines.
Because the knowledge doesn't have to disappear when the hype does.
The skills can remain.
The documentation can remain.
The tools and techniques can be shared.
And somebody else can pick up where we left off.
Which brings us right back to the bus factor.
Accessibility Isn't the Finish Line
My previous post was about making sure fan-translation projects could survive beyond the people who originally built them.
This is another part of that same conversation.
We need more people willing to try.
More people willing to learn.
More people willing to examine unfamiliar tools without immediately deciding that their existence is a threat.
AI isn't going to magically solve every translation problem.
It isn't going to replace the knowledge of experienced translators, ROM hackers, editors, or localization specialists.
But it might help someone take the first step.
It might make an otherwise impossible project approachable.
It might give a small team the ability to attempt something they couldn't have tackled before.
And that matters.
We can argue about quality while still recognizing progress.
We can acknowledge mistakes without dismissing the entire experiment.
We can appreciate human expertise while embracing tools that might help more people develop expertise of their own.
I don't want a future where humans stop thinking because a machine can generate an answer.
I want a future where people use the machine to explore things they couldn't previously attempt, then apply their own judgment to make the results better.
Because the point isn't to prove that AI is wonderful.
The point is to make something possible.
And if somebody eventually comes along and produces a better translation?
Fantastic.
Give them the toolbox.
Give them the documentation.
Give them the opportunity to improve it.
But perhaps we should stop treating the person who opened the door as the enemy simply because they used a different key.
We've spent enough time talking about games that nobody will ever translate.
Maybe it's time to start playing some of them.
The tool is in our hands. What matters is what we do with it.
— Jerry “The Ankle-Biter” Silverhand