The myth of the seven differences

(AI and Copyright, Issue 4)

We concluded the previous episode with the observation that ChatGPT and Dall-E are technically among the Merriest Multipliers one can think of. After all, they bring their final products into being – without any remorse – using the existing texts and images that were provided to them by their creators for learning. Everything they create is built from those, you might say. And since we noted last time that even a reproduction in modified form is still a reproduction, building that final product from existing copyrighted material should require the permission from the creators of those existing texts and images. Right?

technical copy ≠ relevant copy under copyright law

Well… that depends. Actually, not every work that is technically constructed from elements of another work is automatically a reproduction in a modified form under copyright law. In the case of Chat GPT, this is also almost self evident. A user of that program would really have to make a huge effort when entering the “prompt” (the instruction to the program), to cause a text to be generated that even approximates an existing text.

the girl with the pearl earring

With an image generator like Dall-E, this subject is a bit trickier. Last winter, for example, the Hague museum Mauritshuis lent Vermeer’s “Girl with the Pearl Earring” to another museum. It took the museum little effort to replace this masterpiece with several modern adaptations, cretaed with the use of AI. The place of honour was provided to a work by German artist Julian van Dieken, which had been created using Midjourney. Of course, everyone can immediately recognise the original painting in this work.

Is this really a reproduction in a modified form? It very well could be[1]. And if Vermeer’s work had still been copyrighted, Van Dieken could have been sued for it, since he apparently edited the AI output himself (see section 3).

Incidentally, the same would be the case if the whole AI intervention had not been there, but the arranger himself had gone directly to work with the original (as seems to have happened here, for example). And if we zoom out a bit further: everyone knows of works where it is obvious that the creator “took inspiration” from another artist’s work. Sometimes just a little too much, but usually this does not give rise to copyright conflicts. We may take our inspiration from anywhere, as long as we remain original ourselves.

the seven differences -not

So where do we find the line between being inspired and copying? Funny thing is that in this context, some people give credence to “The myth of the seven differences”¹. Where this myth originates is a mystery, but I still notice its existence myself every year among my students. The myth holds that you could get away with imitating someone else’s work, as long as you make sure there are seven (or in some versions: five) deviations from the original. Nice thought, but not true.

Sometimes a work with seven (or as many as ten) differences will still be a reproduction in modified form. (Go ahead and count, in that aforementioned reproduction by Van Dieken.) Meanwhile it is also conceivable that a new, different work is created, even if there is only one actual difference. Indeed, it very much depends on what exactly is the same, or different.
To examine that further, we must take a sharper look look at what makes a product a “work”, i.e. a copyrighted product of literature, science or art.

Next time we will dive deeper into that.

For those who are interested: the illustration accompanying this blog is again by Dall-E2. This time with the prompt: “Oil painting of the girl with the pearl earring with seven differences.”

[1] Note that this would obviously not have any copyright implications here: the copyright has long since expired, since Vermeer died (much) more than 70 years ago.

1. In English better known as “the 30 percent rule in copyright” ↩︎

Questions?