The Merry Multiplier

(AI and copyright – episode 3)

Anyone who has read the earlier installments in this series knows by now that the Instructor (the person who instructs the AI program) can very easily elevate himself or herself to the status of Maker. And thus obtain copyright on the AI creation. All they have to do is add a personal touch to the output of the AI – a small modification that makes the creation just that little bit different. This creates a reproduction of the AI creation in a modified form, which, according to Section 10 of the Copyright Act (please scroll to paragraph 2), is protected as an independent work.

This answers the opening question of this series. It is not the machine (the program itself: ChatGPT, Dall-E, or my new favourite Perplexity.ai), nor the creators of that program, but under certain circumstances the Instructor who can copyright a creation created (mainly) by AI. Simply by making a reproduction of it. But a special kind of reproduction.

Two types of reproduction

Reproducing is one of the two core rights of the copyright owner. The other is disclosure to the public. Owning the copyright on a work provides, within certain limits, the exclusive right to reproduce or disclose that work. Reproduction comes in two forms. First of all, “ordinary” reproduction: i.e. just plain copying. And secondly: reproduction in a modified form.

A lot can be categorised under that second term. Think of the stage adaptation, film adaptation, or simply translation of a book; a hip-hop adaptation of a Golden Oldie, or a photograph of a sculpture. In general, as Section 13 Copyright Act summarises it, “any complete or partial adaptation or imitation in modified form, which is not to be considered a new, original work“. (As the spelling in Dutch shows, this is one of the few articles that has not been amended since the introduction of the “Copyright Act 1912”).

Such a reproduction in modified form is still a reproduction and can therefore “simply” be prohibited by the copyright owner. This was experienced in the 1980s by the (different) makers of the Suske and Wiske¹ parodies De Keizerkraker and De Glundende Gluurder. One album was an anarchic and the other a pornographic parody of the Suske and Wiske stories. Both creators depicted the well-known cartoon characters and their entourage in completely different circumstances or poses than Suske and Wiske creator Vandersteen had ever done. However, they were still sucessfully sued for copyright infringement.[1]

Protected as independent work

But if that was the case… how then could Instructors of an AI work ever be protected by means of making a reproduction? This has to do with another aspect of the reproduction in modified form. This is not only a reproduction, it is also an independent work! The reproduction side is governed by Section 13 and the work side by Section 10(2) of the Copyright Act. The latter section provides, as we saw earlier, that the reproduction in modified form is protected as an independent work, albeit without prejudice to the copyright on the original work.

So what does this mean? Well, quite simply, he or she who reproduces in a modified form remains “indebted” to the creator of the original work. A translation of a book may not appear on the market without the original author’s permission. But the translation itself is also protected. Permission or not: everyone must also keep their fingers off of the translation. If that would be circulated without the translator’s consent, whoever does so will have to deal with both the translator and the author of the original work.

The merry multiplier

But what is the most happy circumstance for the adapter of an AI creation? That this AI creation itself is not protected! (see parts 1 and 2 of the series). So there is no “original work” to be protected: the AI creation is not a human creation and therefore not a work within the meaning of the Copyright Act. This means that the Instructor who makes an adaptation is not indebted to anyone, but is himself protected under Section 10(2). That is, within the existing system – in other words: as long as legislature or case law do not provide otherwise …

Meanwhile, we should note that ChatGPT and Dall-E are of course (at least technically speaking) the Merriest Multipliers of them all! After all, they rake together their end products from all existing texts and images fed to them by their creators and plucked from the Internet. But wait. Doesn’t this in itself constitute a reproduction in modified form, which the Instructor then builds on?

More on that next time.

(Meanwhile, I was curious to see what Dall-E would do with the instruction “Suske en Wiske playing with a photocopier”. It turned out to be nothing – not a Vandersteen creation in sight. Not even when I repeated the instruction in English (“Willy and Wanda playing with copiers”)-see the image accompanying this blog. Could the creators already have built in safeguards, in view of the Vandersteen case law? )

[1] And their reliance on the so-called “parody exception” did not hold up either – they had stayed too close to the originals for that.

1. US: “Willy and Wanda”; UK: “Bob and Bobette” ↩︎

Questions?