Hacker News Reader

Comment by trickstra | Hacker News Reader

And now they can also claim it was just autocompleted by Copilot.

Using a tool to violate copyright isn't a valid legal defense.

There is a "data minig" exception from copyright law, saying that you can use any available code to feed your AI and don't have respect the license. And the output of your AI is yours. So TikTok could claim they either use Copilot, which was fed OBS source, or that they created their own private AI similar to Copilot. Microsoft showed the way.

It hasn't been tested at court AFAIK, but if a company can use this defense to protect their profits, for sure they will try.

Is there?

Can you point it out to me? I’ve looked, but haven’t been able to find it. I imagine it’d be somewhere on this page, in section 107 or later [0], but I haven’t actually found a clause which looks like it either has the intent or side-effect of making “data minig” exempt from having to obey copyright requirements.

[0]: https://www.copyright.gov/title17/92chap1.html#107

In the UK there is a data mining exemption to copyright for non-commercial research if you otherwise have "lawful access" to the work [1] but it clearly wouldn't cover something like copilot and just allows you to build the model. It doesn't give you a copyright exemption if the outputs of the model would otherwise be infringing on the copyright of the input material.[2]

[1] https://www.gov.uk/guidance/exceptions-to-copyright#text-and... [2] That is to say it is not mentioned, and to give you such an exemption it would need to be mentioned. Hasn't been tested in court either as far as I am aware

It's literally the first search result for "data mining exception": https://www.copyrightuser.org/understand/exceptions/text-dat...

This is about EU and UK, but think about it, Microsoft Copilot is already doing it, and so far it seems like they will get away with it. So if feeding an open source code into AI and then using the output in a proprietary application is illegal in the US, then Copilot must be illegal. According to wikipedia in the US it falls under fair use because it is "transformative" https://en.wikipedia.org/wiki/Data_mining#Situation_in_the_U... however Copilot has been seen producing verbatim blocks of code matching some open source projects and so far no legal problems. Why couldn't TikTok claim the same?

Local exceptions are a bad bet when dealing with global distribution.

I really don't think Microsoft Copilot is relying on a local exception.

The legal exception is only in a couple jurisdictions, most of the world they'd still be liable.

That's not plausible

Hey, just because it would get laughed out of court, that doesn't mean they can't claim it!

I‘m not so sure about it being laughed out of court unfortunately.

Can concur, courts don't know jack about tech and have to rely on experts that aren't always independent.

Rebutting the argument doesn't rely on tech knowledge.

“But, your honour, I didn't copy this person's book! I used the autocomplete on my phone, and it just so happened to produce their 500 000 word novel!”

You need tech knowledge to think that's even plausible. Sure, they wouldn't dismiss it out of hand (I think “laughed out of court” is a figure of speech), but I wouldn't be surprised if it got a few giggles.

This would be more like GPT-3 completing a specific novel.

ISTR someone used GPT-3 to autocomplete a chapter of Harry Potter.

And doing so is still copyright infringement.

Reply on news.ycombinator.com