|
|
Log in / Subscribe / Register

Class action against GitHub Copilot

Class action against GitHub Copilot

Posted Nov 11, 2022 9:51 UTC (Fri) by gspr (subscriber, #91542)
In reply to: Class action against GitHub Copilot by bluca
Parent article: Class action against GitHub Copilot

Someone has made a machine that outputs *potentially* copyright-infringing code. Whether or not it really is, is the core of the debate, and not something that's easy to answer. What I fail to understand is how "I and all my peers just use that machine to output non-copyrightable boilerplate" is any sort of excuse.


to post comments

Class action against GitHub Copilot

Posted Nov 11, 2022 10:24 UTC (Fri) by bluca (subscriber, #118303) [Link] (10 responses)

Because it's not an excuse, it's explaining how this works in the real world, outside of clickbaity articles and copyright troll lawsuit fishing for money. Because this matters, a lot.
My smartphone camera can also *potentially" output copyright-infringing pictures. My mp3 player *potentially* plays copyrighted songs. And so on - these are tools, and their main intended and common use matters, and for alleged open source supporters to side with copyright maximalists and trolls looking for a quick payday is missing the point so much that it's not even fun anymore.

Copyright maximalism is bad for us. The only reason this gains any traction is because it's done by Microsoft, if Copilot had been built by Mozilla reactions would be quite different, and that's just sad and short-sighted.

Class action against GitHub Copilot

Posted Nov 11, 2022 11:11 UTC (Fri) by gspr (subscriber, #91542) [Link] (9 responses)

I think these analogies are terrible. Your camera and your music players are indeed tools that you can use to infringe on copyright. They didn't come to you in a state infringing on copyright. The claim about Copilot, however, is that it "contains" (for some value of contains) and produces copyrighted material.

Class action against GitHub Copilot

Posted Nov 11, 2022 11:47 UTC (Fri) by bluca (subscriber, #118303) [Link] (8 responses)

No, those analogies are perfectly adequate. Copilot does not contain any infringing material, it's not a repository. And you have to intentionally make it produce exact copies, with carefully selected inputs that you need to have pre-existing knowledge of, it does not happen randomly, it takes a lot of effort. So it's exactly like a camera or a music player, in that regard. It's the user intent that causes that behaviour, not the tool itself.

Class action against GitHub Copilot

Posted Nov 11, 2022 12:14 UTC (Fri) by gspr (subscriber, #91542) [Link] (7 responses)

> Copilot does not contain any infringing material

That's the question at the heart of the conundrum, and it's a very complex one. Sure, we can probably agree that Copilot does not contain bit-for-bit copies of copyrighted material. But that's not the bar. Distributing a lossily compressed copy of a copyrighted image without permission can still be infringement. On the other hand, distributing the average value of all the pixels in said image certainly is not. The spectrum in-between is where it gets hard, and this is (in my opinion) probably where Copilot and similar fall.

Class action against GitHub Copilot

Posted Nov 11, 2022 15:32 UTC (Fri) by Wol (subscriber, #4433) [Link]

As a completely different example, copying material as part of a lawsuit does not infringe copyright. I can copy a work, present it to court, and copyright cannot touch me.

But if somebody else then takes my work and publishes it in a newspaper, that's not a legal document. Me putting it in a legal document did not strip copyright, it just gave ME immunity. The publisher can still get done for it.

Cheers,
Wol

Class action against GitHub Copilot

Posted Nov 11, 2022 19:55 UTC (Fri) by bluca (subscriber, #118303) [Link] (5 responses)

Of course it is complicated, but the crux is that I don't see how this tool can be defined as something that distributed copies of anything. It is very obviously not built to pick existing snippets and shove them out of the door 1:1 to users. It's not how it works in the vast majority of cases, where it builds something adapted to the surrounding environment - which is what makes it so darn beautiful and useful to use. Then there are users who go out of their way to carefully construct a surrounding environment (=> input query) using their pre-existing knowledge to intentionally cause it to spit out pre-existing snippets in order to write clickbaity articles or, in this case, look for a quick payday.

Class action against GitHub Copilot

Posted Nov 12, 2022 15:24 UTC (Sat) by farnz (subscriber, #17727) [Link] (4 responses)

It can be defined as something that distributes copies of something in exactly the same way as a human engineer can be defined as someone who distributes copies of something.

If I have, in my notebooks, details of how to do something in kernel-dialect C, and I read a snippet of code from those details then adapt it to the codebase I'm working on, then I've distributed a copy of the snippet in my notebook. If the snippet in my notebook is not protected by copyright, then this is not an issue; if it is, then I've potentially infringed copyright by copying out that snippet and adapting it.

The same applies to Copilot - its model takes the place of the engineer's notebooks and knowledge of what they can find in their notebooks, and its output is potentially infringing in exactly the same way as a human engineer's output is potentially infringing, complete with fun questions around "non-literal copying".

Class action against GitHub Copilot

Posted Nov 17, 2022 13:16 UTC (Thu) by esemwy (guest, #83963) [Link] (3 responses)

A human engineer can be inspired by someone else’s code. An AI has no such ability. Computers memorize, and have rules. The fact that the AI is really complicated doesn’t change that. It’s more as if they obfuscated the source and are trying to pass it off as their own.

Class action against GitHub Copilot

Posted Nov 17, 2022 14:08 UTC (Thu) by farnz (subscriber, #17727) [Link] (2 responses)

You're getting into the philosophy of what it means to be human, and missing my point at the same time.

My point is simply that if a human can infringe while doing the same thing that Copilot does, then it's absurd to say that Copilot cannot infringe because it's an AI - rather, it's reasonable to say that Copilot's ability to infringe copyright is bounded on the lower end by the degree to which a human doing the same thing can infringe copyright.

I've also provided a sketch of how a human can infringe copyright, which I can expand upon if it's not clear; unless you can demonstrate that Copilot is incapable of doing what the human does to infringe, however, you can't then claim that Copilot can't infringe where a human can.

Class action against GitHub Copilot

Posted Nov 17, 2022 15:57 UTC (Thu) by esemwy (guest, #83963) [Link] (1 responses)

The distinction does make a difference in this case. If a human takes a snippet of code and copies it, only changing the identifiers, my understanding is that is still a violation of copyright, and a company can’t take code, compile it, and claim it’s not an infringement. Copilot seems more like the latter, where the former example would be the human analog.

Class action against GitHub Copilot

Posted Nov 17, 2022 17:10 UTC (Thu) by farnz (subscriber, #17727) [Link]

I'm not following your reasoning here - in what way should Copilot be permitted to do something that would make Microsoft liable for the infringement if an employee did it?

Remember that I'm setting a lower bound - "if Copilot was just a communication interface to a human being at Microsoft who looked at the context sent to and then responded with a code snippet, would Microsoft be liable?". My claim is that if Microsoft would be liable in this variant on Copilot, then Microsoft are also liable if Copilot is, in fact, an "AI" based around machine learning, but that this is a one-way inference - if Microsoft would not be liable if Copilot was a comms channel, this doesn't tell you anything about whether Microsoft are liable if Copilot is actually an AI.

To summarize: my reasoning is that "an AI did it, not a human" should never be a get-out clause - it can increase your liability beyond that you'd face if a human did it, but it can never decrease it.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds