Art, AI and copyright

Posted on: November 9, 2018 by

A big splash was made when a lot sold at Christie’s New York last month for $432,500. That sort of amount is usually small change for the major international auction house, but not when it comes to a particular sort of artwork: one made by artificial intelligence (or AI). In fact, this was reported to be the first AI artwork ever sold at auction and managed to fetch 43 times its initial estimate of $10,000. The piece, called Portrait of Edmond de Belamy, is part of a series created by a French artists’ collective named Obvious.

In all honesty, the piece isn’t much to look at. To some, the whole thing might seem like a hoax. October after all saw another unexpected auction room incident involving a partially shredded Banksy. So what exactly is this work of AI – and why would anyone pay so much for it?

Edmond de Belamy, an AI generated work that sold at Christie’s last month for $432,500 (image c/o Christie’s).

From what I gather, works created by the Obvious collective are made by feeding data into an algorithmic process known as GAN (or ‘generative adversarial network’). The data in this case involved images of 15,000 portraits from the 14th to the 20th centuries. Computer-generated images are then passed through what’s called a ‘Discriminator’ which rejects all images until one is submitted that effectively tricks the Discriminator into thinking it fits within the dataset. And Portrait of Edmond de Belamy is the result. It is therefore a computer-generated image, but one that must have sufficient characteristics shared with portraiture from years past to be accepted by the Discriminator. (I might be wrong about how this operates, and if you think I am, please tweet to @IAL_art_law and let me know how this process actually works!).

The visually less-than-pleasing outcome would not look out-of-place next to certain works of cubism or a Francis Bacon (see comparisons below). And maybe that’s saying something: not that the piece is particularly successful, but rather that artists like Picasso 100 years ago and Bacon 50 years ago were onto something. Without realising it, they might have been creating composites from a long tradition of European painting, something algorithms have now arrived at through a different route. Given that comparison, perhaps we can find value in the present piece. We can see it in the tradition of Picasso, of Bacon, as well as being a statistically accurate amalgam of 15,000 prior works of art.

Pablo Picasso’s Portrait of a Woman (1910, Boston Museum of Fine Arts) and Francis Bacon’s Portrait of Isabel Rawsthorne (1966, Tate), works which can arguably be seen as precursors to the Portrait of Edmond de Belamy. Both of the above works, still in copyright, were reproduced for the purposes of criticism and review (see section 30, Copyright, Designs and Patents Act 1988).

But there are issues here. For one, there are claims that the code underpinning the GAN algorithm was not actually developed by Obvious. In fact, the Obvious members have even admitted to this. The code comes from one Robbie Barrat, a nineteen-year-old coder who shared his algorithm through an open source licence. One professor and AI artist has even used the Barrat code to come up with the following AI works, any of which could pass as the brethren of Edmond de Belamy:

Computer-generated works by Tom White, using the code of Robbie Barrat, the same code used by Obvious to create the Portrait of Edmond de Belamy (c/o The Verge)

The point of this was to show how little Obvious had changed the original code, if at all. Interestingly, in its coverage of this matter prior to the Christie’s sale, The Verge made no reference to any copyright issues underpinning the disagreement. Now, in some ways, this is a moot point: any copyright in the GAN code developed by Barrat has seemingly been waived (or at least widely licenced) through his making it available via open source. Besides, it doesn’t look like this is a battle Barrat wants to fight – as a recent high school graduate, he likely has other fish to fry.

There is also the matter of whether the artistic product of the code can itself be protected by copyright. The work seems to have been created in France, where Obvious are based, and was then sold in New York. I’m unsure exactly how either of those jurisdictions would treat computer-made works (if you know the answer, again do tweet it to me at @IAL_art_law). From what I gather from Eleonora Rosati writing on the always excellent 1709 Blog, at the EU level and internationally, copyright protection is generally reserved to human beings. That said, there is a difference in the approach in the UK, where the copyright legislation (the Copyright, Designs and Patents Act 1988) does provide that in relation to a ‘computer-generated work’ the copyright will be owned by the person who makes the ‘arrangements necessary for the creation of the work’ (section 9(3) of the Act). Were such a situation to arise in the UK, there would inevitably be a debate as to who, between Barrat and the artists in the Obvious collective, would be that lucky person: the original coder or the team that ran the code?

Now, UK law is not applicable here – and in fact, the copyright discussion as a whole may seem rather academic. No one is trying to claim copyright in the piece (at least not yet). Its value is in the original printed version, the physical object that sold for $432,500. Nevertheless, if I were the buyer of such a piece I would want to get the copyright situation nailed down pretty quickly. An assignment to the purchaser of whatever copyright Obvious has in the piece (again, this would likely depend on French and/or New York law and may not exist in the first place) would be wise, or otherwise a warranty from Obvious that no other printings of Edmond de Belamy are to be made. Imagine having spent all that money only to discover, weeks later, that the outfit is mass-producing the image and selling it to anyone with a heartbeat. There is something to be said, after all, about protecting the original, even the original of an AI-created artwork. Perhaps, along with its financial value, a computer-generated work like this can actually possess an ‘aura‘.

Nevertheless, as one considers the process by which the work was created one realises that this is just another form of gathering Big Data, whereby the instigators are effectively doing what Google does, taking enormous amounts of data and using it to produce a result. In this case, the result seems somewhat primitive, perhaps even ugly. But these are early days. Isn’t it a matter of time before someone develops a more advanced code that can allow the Discriminator to be a little more… discriminating? With such a system, the output chosen may look strikingly like the creation of a human hand. Of course it won’t be, but won’t that make it all the more startling? So maybe, taking all that into consideration, there’s a reason why the bidder paid as much as he or she did. Maybe we are witnessing the vanguard of a great new tradition in art.

Either that, or it’s a fad. The jury’s still out.

For those interested in learning more about copyright and art contracts, our forthcoming Diploma in Art Profession Law and Ethics will cover the topic (it begins January 2019), as will our more copyright-specific Diploma in Intellectual Property and Collections (which runs every June).