Here’s a thing I came across recently that has really stuck in my mind. Devcomp is (was) a community of people with a shared interest in decompiling commercial games: for fun, for education, or just because it was there. The work was difficult and time consuming, and the community formed around the process, the common effort, more than the outcome. Now that AI has made decompilation trivial, made the shared work unnecessary (and not even taking into account that scraping their work made it possible for the AI to do it in the first place), the community has disbanded. I’ve read similar stories about modding communities. Prompting the question: if AI makes a thing so easy, what’s the point of doing it at all? Just food for thought.
I’d imagine that the point of doing something is that you enjoy doing it, which is true of most things really.
Has anyone put in the work to personally decompile a game and then run that same game through the ai decompiler to compare results? I fully believe that this is something AI is capable of doing well (pattern matching is gen ai’s whole thing) but I also wouldn’t be surprised if there were major differences between the AI’s version and how humans like to do it.
As perhaps a case in point, I recently worked on my Laser Vault project. I have no doubt an AI could have done a much better job than I did. But one thing I wanted to do was create “autoplay” for most of the games, so that people who wanted to see how they played but didn’t actually want to play them could do so. I have little doubt that an AI could have read the ROMs the same way I did to learn what the correct keys were to be pressed in order ot automate the game. But I found it an interesting challenge. I’m actually now curious to hand it to an AI and see what it would do to learn the solutions; perhaps it will find a better way than I did.
When I did my Improbable Mission (an implementation of a Commodore game), I have no doubt an AI could have done a better job at breaking down the sprites, but I documented how I did it because I thought the process was interesting. I also documented how I broke down the original game to learn how it worked. I’m actually curious to give this implementation to an AI and see what it would have done differently had it ripped the resources itself.
I bring all this up because there’s a sort of related point: a lot of people I have talked to think Gary Kaaparov, in his book Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins, argued that chess is a solved problem and uninteresting because a machine could reliably outclass a human in it, and that this was a cautionary tale. But he didn’t say that!
He writes extensively about how AI has democratized the game, helping players around the world learn faster and discover beautiful, unexpected ideas that humans previously thought were bad moves. In fact, instead of giving up on chess, Kasparov pioneered Advanced Chess (also known as “Centaur Chess”), a format where a human player and a chess engine team up against another human-machine pair.
By this same logic, consider this: quite a few of us here write Z-Machine interpreters. But why? There are already existing ones. The problem has basically been solved. Not by AI, but by another human. So it would be trivial to just look at someone else’s solution and build from that. But in that case, why do it at all? You would just be replicating what someone else had done. But many of us don’t do that: we use those existing ones as reference implementations and learn how to construct our own. Why? Well, for me, just for the learning experience and the fun of it.
I’m not sure if I actually had a good point to make here.
But if I did, I would say you can always find a way to ask, “What’s the point of doing this thing I’m interested in at all?” where the answer either comes out positive or negative.
Like most questions surrounding AI, the question actually predates AI.
Why strive for realism in painting or drawing when almost everyone has a camera in their pocket?
Why learn to sew, knit, crochet, etc. when you can just buy finished textiles?
Why grow your own produce when supermarkets exist?
Why learn to woodwork when you can just buy finished furniture?
et cetera.
Honestly, I feel like such questions might be the product of people getting so caught up in hustle/gig/influencer culture that they’ve forgotten what the point of a hobby is and that it’s okay to do things because you enjoy them and not just because they are profitable, impressive, and/or make the world a better place.
Yes.
Reading the post, it looks like it’s more complicated than that. The site maintainer says the problem were bad actors abusing the site and lifting others’ work, not merely that AI made hand-rolled decompilations unnecessary: “What is also now clear is that the community aspect of decompilation has been broken.”
Apparently, the site was gameified, meaning producing decompilations would earn users higher rankings on some kind of leaderboard. Bad actors were forking others’ work-in-progress and using AI to complete it and race up the charts. They were also taking others’ work to port games to other platforms (using AI) against the original decompilers’ wishes:
[AI] has warped decompilation into a hollow numbers game, exacerbated by folks who just want to burn tokens to make number go up or “port game to PC”, all else be damned.
There’s also another interesting point made by the site maintainer:
I have long held the opinion that AI is not the issue by itself, and we now know it can help tremendously when used correctly. “Slop” comes in many forms, is hard to define, and even harder to moderate.
I would respond that, even if AI has virtuous use cases, that doesn’t excuse its misuse or abuse in other situations. There’s nothing hypocritical with saying “We love AI, we just don’t allow it here, slop or otherwise.”
Of course, it’s the maintainer’s site to do with as they wish. If moderating bad actors has become all-consuming, then I can’t fault them for the choice they made.
Side note re: interactive fiction
What sucks about all this is that the decompilations site encouraged open development and open progress. Bad actors leveraged this openness to advance themselves.
Most IF and choice-based games tend to be developed quietly, either due to norms/culture (“I should only announce my game when it’s really ready”) or rules (many comps require new work, not work that has been in circulation).
Is this protecting the community from the kind of crap Devcomp was besieged by? I don’t know, but I wanted to point out the difference.
One of the really magical things about IF games (at least to me) is that you really don’t need a huge dev team to help you with an ambitious project. I’m not saying that it’s easy, but one person with a vision, a lot of time on their hands, and heaps of ambition can put out a project that’s just as good as the commercial offerings. That’s just not true of a lot of other kinds of tech projects, and I think that you’re right in saying that their open collaboration on the internet leaves them vulnerable to bad actors with AI tools.
Yep, I’ve been following this news piece on and off, and this is the main culprit behind the demise of the website. Gamification has always been a double-edged sword since it encourages people to publish fast, but the “ease-of-use” of AI decomps makes the cons outweigh the pros.
There’s frankly a lot of “PC ports” that have no human thought put into them. Recently, someone announced on Reddit that they made a “native port” of the XBOX 360 version of Arkham Asylum, a game that already has a fantastic PC port. Supposedly, this was done to facilitate Android ports, which is dubious reasoning at best. To me, this example highlights how most vibe coders are not interested in making actual ports (I’m sure a good number of people would appreciate ports of console games that haven’t appeared on modern consoles and PC) as they are about chasing credibility.
I also agree with Nelson that there’s a few meaningful differences between decomps/retro gaming and the interactive fiction space. Both Spring Thing and IFComp have barred games that have AI art and writing, and cultural legitimacy is also gained from active participation in the forums. People here don’t really like “slop” even before “AI slop” was a thing. It’s one thing to reverse-engineer an old interpreter, it’s another to use its limitations to make something artistic.
Open source projects certainly have more surface area for bad and ignorant actors to attack with AI, and making closed sourced projects open source is kind of the point of decompilation, so it makes sense that would be an area particularly vulnerable to abuse by AI script kiddies who don’t know better or glory seekers deliberately sniping people in the hopes of impressing end users who don’t understand the intricacies of game hacking.
Makes me wonder how vulnerable console game hacks that are based on a publicly available disassembly of their base game but for which the hack itself is closed source are compared to comparable games with no prior art on the disassembly/decompile front.
Also, I understand many of the AI-assisted ports aren’t actually full ports and that many rely on code ripped from open source emulators and are closer to wrapping the original rom in a stripped down emulator and have only partially ported the game code… In particular, I understand a lot of N64>PC ports still use emulation code to translate N64 GPU calls into something a modern GPU understands, though for all I know, that might be due to the Silicon Graphics tech the N64 utilized just being that weird.