If GitHub ends up having to tweak their product to avoid ethical/legal concerns, I actually imagine it could still be pretty cool. Right now Copilot is a black box that spits out code with no attributes; what if they worked on instead making it a glass box, where it always brings up snippets of other projects along with their licensing info so that you can decide how to incorporate the ideas fairly yourself? Or they could still output the same code suggestions, but always include attribution and license data along with it. Making the product more transparent would probably make more people comfortable with using it, anyway.
We've filed a lawsuit against GitHub Copilot
821–824 of 824 posts
Re: We've filed a lawsuit against GitHub Copilot
#822Earlier quoted context omitted.
1) This "copilot is great 'cause copyright is evil" argument breaks when you look at the fact that copilot is copyrighted, closed software tool for producing closed, copyrighted software. If you trained copilot on GPL'd software and specified that copilot's output was also GPL'd maybe you'd have some reasonable claim (but even then, the attribution claim would come in). 2) So far, these tools are "better search" sche…
#1 is a pretty dumb argument. Should all of Silicon Valley VC backed startups send Linus Torvalds billion-dollar checks? I don't see a world where Copilot isn't fair use (at least in America).
Linus is richer and more famous than musk, and musk is just another CEO with a reasonable salary.
-Some universe out there, with a god smarter and better than ours-
Re: We've filed a lawsuit against GitHub Copilot
#823Earlier quoted context omitted.
Yet we all use web browsers that copy copyrighted text from buffer to buffer all the time. This doesn't even include all of the copying that ISPs perform. It might be fair to say that the read performed in training has the same character since no human is involved. The real copyright violation would be using a derived work.
A browser isn't a amalgamation of billions of pieces of other works. A browser executes and renders code it's served. Copilot's corpus is quite literally tomes of copyrighted work that are encoded and compressed in its neural network, from which it launders that work to create similar works. Copilot itself, the neutral network, is that corpus of encoded and compressed information, you can't separate the two. Copilot…
I don't think "compressed" is a good way to think about this because the process is lossy. The model can't be "decompressed."
If you think this way, hashing is a copyright violation.
Re: We've filed a lawsuit against GitHub Copilot
#824Earlier quoted context omitted.
No, they got absolute copyright on the API signatures. The court did not even question any copyright, it just assumed the APIs are copyrighted by Oracle. Than it looked for reasons why copying the APIs could possibly be fair use… By the skin of their teeth they found some very involved and case specific reasons why Google's use of the copyrighted APIs was, after all, fair use. https://www.bhfs.com/insights/alerts-art…
The reason why SCOTUS bent over backwards to not talk about copyrightability was not because they assumed it was true for APIs, but because they didn't feel like they had all the facts. They basically said "we don't know if it's copyrightable, but if it is, here's a ruling that makes this case and anything similar to it go away". Oracle only has copyright over APIs in the Federal Circuit, because they were able to ho…
I still don't buy the part that there is not much to worry.
The thing you call "thin copyright" is still copyright. Being protected or not is in the end a binary judgment: If your stuff is "a little bit" protected it is actually fully protected—with all consequences that follow from that.
Also, alone the "assumption" of the highest US court that APIs are protected is a very strong signal. They could just have ruled that there is no protection at all; case closed. But they preferred to go for a weasel solution. This has reasons… They deliberately didn't open up the door for API freedom. (Most likely to still be able wield that weapon against foreign concurrency should they feel like that some day).
The point is: IP law is completely crazy. The smallest brain-farts are routinely protected.
The exceptions to this rule are actually stronger in civil law, but still even in the EU single words or sub-second audio samples are protected by default. (Regarding APIs the situation is better though: It's legal to reverse engineer something for e.g. compatibility, and a few other reasons; but that are explicit exceptions. The default is that almost every expression of even the slightest form of human "creativity" is copyrighted; the bar is extremely low; and gets actually pushed constantly lower and lower by common law influence).
So on both sides of the Atlantic the default is that every single line of code is protected. There is nothing like a lower bound in size. Than, form there, you could try to argue that there should be an exception from this protection in some particular case, e.g. there was no "creativity" at all involved. But you will need to win a—often very hard, expensive, and ridiculously long—fight over that issue, and wining that is nothing like a sure thing; the default is that just everything is protected to the max. (Just have a look at all the craziness around news headlines in the EU; Google lost that case back than; to understand this better, as this may be very surprising to US people: civil law does not recognize anything like "fair use"; there are exceptions of copyright protection that have in the end almost the same effect, like grants for libraries or educational purposes, but those exceptions, and their limitations, are listed explicitly in the law; if no exception is listed there just isn't one, and only the very vague "creativity bar" remains).
Regarding Copilot: It makes not much difference whether this machine spits out some verbatim copies of (clearly copyrighted!) snippets or some "remix" thereof. There is no "novel" code if at best all what this machine does is creating "remixes" of the code it has in its database based on the query given. (Its "knowledge base" is nothing else than a very funky database; technical details regarding the actual implementation of that database or its query system should not matter legally).
Before this comes up again: No, any comparisons to how humans learn are irrelevant in this consideration. That machine is not a human. It's a machine. End of story. So even if you consider also a human brain a kind of "funky database" this makes no difference.