If a neural network generates an image, who owns the copyright? The owner of the dataset that the net was trained on? The designer of the network architecture? The person running the code? Or... the AI system itself?
I looked into this a while ago and apparently there's never been a clear precedent and the lawyers are still debating it in the journals. The de facto interpretation everyone seems to use is that the copyright is owned by the person who generated & selected the image.
-
-
What if both generation & selection were done programatically via a decentralised infrastructure (e.g. blockchain)?
-
Based on cases like Naruto (https://en.wikipedia.org/wiki/Monkey_selfie_copyright_dispute …) the reasoning might be that the owner is whoever is most proximate to causally initiating the generation+selection process.
-
I believe, for copyright, we need to reference a particular jurisdiction first. If we are talking about the US, then I agree with the general reasoning You need to trace the causal chain back to a human. Who owns the code that generated the image? <--key inquiry
-
Does the manufacturer of your camera own the images you take with it? Does the writer of the text editor you wrote your blog post about those photographs own the copyright to your blog post? What about the writers of the OS and networking code and hard drive code used to send it?
-
The manufacturer of the camera doesn't own it once I buy it. The owner of the code. Whomever has the copyright to it, would own the output value.
-
Nope, doesn't work. Your don't own the copyright to the text editor, OS, networking, hard drive, or any of the other things involved either.
-
? Irrelevant. There is no copyright to the "hard drive"
-
Of course there is. The hard drive runs itself with a ton of firmware.
Край на разговора
Нов разговор -
-
-
In the case of terrapattern, apparently Google argued that the weights of the network trained on their imagery was an unlicensed derivative work:https://twitter.com/golan/status/976502373521731584 …
-
One can only assume that Google would never dare take that argument to trial as such a precedent would backfire spectacularly on them.
-
Actually this case was because there was an explicit contract that covered this contingency.
-
Ah. Well, that would do it. But even if one agrees to transfer the copyright of any models trained on the data provided under the contract, the general principle probably doesn't hold (and Google and all tech companies should hope it doesn't).
Край на разговора
Нов разговор -
-
-
Do you have a pointer to the journal papers debating this by any chance?
-
No, I dumped them into a Reddit comment years ago, be hard to refind it. But they're not hard to find in Google Scholar, the obvious search terms about derivative copyright and database copyrights work.
-
Ok, thanks for the keywords
Край на разговора
Нов разговор -
Изглежда зареждането отнема известно време.
Twitter може да е претоварен или да изпитва моментно затруднение. Опитай отново или виж Twitter Status за повече информация.