Hacker News

points by CookieCrisp 1 day ago | 0 comments

I think this is a bit disingenuous. The stance they’re taking seems consistent to me. I’m Interpreting their view as “Using AI to create something new isn’t theft, using it to recreate or access an existing product is copyright infringement” - makes sense to me

[–] garyfirestorm 1 day ago

You can’t have the first one without the second one. AI doesn’t create something out of thin air, it is only recreating from training data. Which is infringing portion of the argument.

[–] archontes 1 day ago

How, to you, does training an AI violate copyright?

[–] alerighi 1 day ago

If the material that you train the AI on is copyright protected you are by definition violating it.

Otherwise AI becomes a glorious copyright-laundering machine, use something that you want to copy to train an AI, then ask that AI to replicate the stuff it was trained on. Of course it will make a perfect copy, but who is the creator of that intellectual work? The AI itself not for sure, but the original author of the stuff that the AI copied.

An no, it doesn't matter if the AI doesn't copy exactly but does a bit of shuffling of stuff, if not you are arguing that it's not cheating if you copy the homework of your classmate but you rearrange some words to make it not identical...

[–] archontes 1 day ago

How does that follow from the definition of copyright or the protections it offers? This is law we're talking about. Cite the black and white.

From copyright.gov "Copyright does not protect facts, ideas, systems, or methods of operation, although it may protect the way these things are expressed"

What rights does copyright provide?

Reproduce the work in copies or phonorecords.

Prepare derivative works based upon the work.

Distribute copies or phonorecords of the work to the public by sale or other transfer of ownership or by rental, lease, or lending.

Perform the work publicly if it is a literary, musical, dramatic, or choreographic work; a pantomime; or a motion picture or other audiovisual work.

Display the work publicly if it is a literary, musical, dramatic, or choreographic work; a pantomime; or a pictorial, graphic, or sculptural work. This right also applies to the individual images of a motion picture or other audiovisual work.

Perform the work publicly by means of a digital audio transmission if the work is a sound recording.

Training an AI isn't on this list. An AI may produce infringing outputs, but training an AI isn't a violation of copyright.

"Copyright laundering" is a nonsense idea. Copyright is not designed to prevent people or machines from producing things in a style or similar things. It protects against exact reproduction or reuse, and that only to the extent that the form is separable from function. If the form is inseparable from function, the expression cannot be copyrighted at all.

Let's assume "shuffling stuff" as you describe it would solidly place a work into the "derivative works" category.

Fine, the AI has produced a derivative work. The original copyright holder has a claim on that work if it were to be distributed, sold, displayed, or performed.

Does the AI hold a copyright? No. Only humans can hold copyrights. Does the prompter hold a copyright? No. A human has to perform the transformative work, not merely arrange for a machine to do it. AI outputs are not copyrightable.

But how you "feel" about it isn't a useful measure. It being like "cheating off someone's homework" isn't useful, either. Hell. Copyright is carefully crafted to allow "cheating off someone's homework" provided that the result doesn't reproduce elements of the original exactly.

[–] Micrococonut 18 hours ago

You make a great argument for expanding the definition of copyright to include protection from use in AI training.

[–] archontes 14 hours ago

Maybe so. Maybe a disruptive technology now exists that renders the modus operandi of many industries moot. It's called disruptive for a reason.

I am also of the opinion that AI weights aren't copyrightable, as they're the result of an automatic process. "Open source" weights are a laughable non-sequitur.

I am not generally of the opinion that we need to expand copyright in order to protect a status quo.

[–] cyanydeez 1 day ago

Ok, so lets say you "train" and "AI" using the binary source code of windows, then ask the AI to recreate Windows.

Does that violate copyright? What if we us the binary source code of MacOS, Linux and Windows, then ask it to recreate Windows?

Does the fact that we added some other ingrediants violate the idea that the LLLM will create a substantially similar product?

I'm not defending windows, but indicting LLMs as anything but "fair use".

[–] withinboredom 1 day ago

I don't think that's how it works. From experience from working on an experimental online learning system... it can't remember verbatim unless it is _really_ in there many many times. It can reconstruct it pretty close though.

source: online learning of codebases in the hundreds of thousands to millions of lines of code

[–] protimewaster 1 day ago

I agree. I'm not sure why anyone thought that these companies would be okay with someone releasing an unofficial title in their flagship franchise, AI created or not.

It seems like the news outlet is running this as "suddenly they object to AI" and ignoring the fact that they're objecting to it because it's clearly an infringing product. It would be infringing whether it was made with AI or not.

This is akin to somebody saying we should be allowed to own knives for cooking, and then calling them a hypocrite when they object to knives being used to stab people in the face.

[–] xyzsparetimexyz 1 day ago

It's not even that. The main thing is _publishing_ the decompilation. Doing the decompilation is fine. If I own a game I'm pretty sure I am (or should) be able to legally decompile it or recompile it and turn it into something that runs in a browser or on my phone or whatever. But the source code (not relevant here unless there's a leak) and binary are copyrighted and decompilation doesn't count as a transformative work.

[–] AlotOfReading 1 day ago

It would be wild to make the argument that decompilation isn't transformative. A classic test is whether the derivative work serves a different purpose than the original. Google thumbnails for example. That seems pretty obviously true for decompilation, since it's the first step in quite a lot of modding, porting, and preservation uses that the original binary doesn't serve.

[–] avaer 1 day ago

> If I own a game I'm pretty sure I am (or should) be able to legally decompile it or recompile it

That's a nice theory (and I morally agree with it); not sure the courts would agree; it hasn't been tested. I expect the money would win in this case if we extrapolate from history.

[–] avaer 1 day ago

Rules for thee, not for me.

These companies would love to slurp up your game (via AI) but replicating theirs (via AI) is no bueno. Whether you've infringed comes down to whether you can afford the legal fees.

Remains to be seen how it plays out but I always liked Sega's approach to Sonic fan games. They are their own rich genre, which have largely symbiosed with corporate. Though, I suspect most studios won't understand the value of that approach.

[–] Ohentis 1 day ago

I don't really think this is the case. The rules feel very consistently applied. Training is fine. Output may still be copyright infringing in some cases (but difficult to prove and none of these tech companies have tried yet). Publishing decomps is not fine and never has been. Piracy is not fine and never has been (and anthropic got in trouble for it). The only case that I know of where an AI company got off free for something an individual would be punished for is the various hacks that they've "accidentally" done.