Artificial-Impressionists: a movement of AI-generated artwork

Remember Naruto, the monkey that took a selfie?  Courts in the United States determined that human authorship was a requirement for copyright protection.  So, if we can’t award monkey copyright, what about AI?

We have already seen AI produce all different types of content: photos, music, articles, poetry and novels.  And just last month, Christie’s auctioned off a piece of art created by AI (pictured below).

Image © Obvious

The “Portrait of Edmond Belamy” looks like something out of the 18th or 19th century, perhaps a Manet or a Renoir. The slightly blurred features of Mr Belamy are a signature element of the impressionists.

What gives away the fact that this painting was not created by a human is the signature in the bottom right of the painting (pictured below).  The signature is a reference to the core component of the algorithm that produced the work.

Image © Obvious

The painting sold for USD$432,500, 45 times higher than its estimated value of USD$7,000 – USD$10,000.[1]

How was the painting created?

The AI was programmed by Obvious, a ‘Paris-based collective of artists, Machine Learning researchers and friends interested in AI for Art’.  The method used by Obvious is known as Generative Adversarial Networks, aka GAN for short.

GAN is a machine learning algorithm that is comprised of two parts, the Generator and the Discriminator.  To create the images, Obvious fed over 15,000 paintings from the 14th to 20th century into the algorithm.  A new image is then created by the Generator and shown to the Discriminator.  The Discriminator tests the image to see if it can tell if the painting was made by a human or by a machine.  If the Discriminator cannot tell that the painting was made by a machine, the painting passes the test.  The two parts of the algorithm compete, essentially training each other.

The algorithm learns from data fed into it by humans, generating new content by making independent creative decisions.  The question then is who created the painting?

If the creator is the one who produces the art, then the algorithm is the painter and has appropriately signed the painting.  Or does the inspiration come from the person who created the algorithm and fed it the data, making this person the true artist? Some may argue that the algorithm is simply the tool the artist used to create the work.  But does this argument hold when these systems are capable of making independent decisions about the artwork without human intervention?

So who owns the intellectual property?

If we accept that the creator of the work is the AI then this poses a problem when establishing intellectual property in the painting.  Most jurisdictions around the world require a human author to warrant copyright protection (remember the monkey selfie).

In the US, the Copyright Office has declared that it will “register an original work of authorship, provided that the work was created by a human being.”[2]  The same position has been taken in Europe by the Court of Justice of the European Union.[3]

The UK has taken a different approach. Section 9(3) of the Copyright, Designs and Patents Act 1998 (UK) states that “[i]n the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.”  This wording opens the door for copyright protection to be extended to works created by AI.

In Australia, section 32 of the Copyright Act 1968 (Australia) (Act) states that copyright subsists in original works of which the author was “a qualified person at the time when the work was made” or “if the work extended over a period – was a qualified person for a substantial part of that period”.  Section 32(4) of the Act clarifies that a “qualified person” means an Australian citizen or person resident in Australia.

This concept has been expanded on through case law in which the courts held that the exertion of independent intellectual effort by a human was necessary for copyright to subsist in work.[4]  In Acohs Pty Ltd v Ucorp Pty Ltd[5] the Full Federal Court of Australia declared that copyright did not subsist in HTML source code of material safety data sheets generated by a computer because the work had no human author.

If the Portrait of Edmond Belamy was created in Australia, no copyright would subsist in the painting.  This fact exposes the gap in the copyright laws of Australia and many other jurisdiction in relation to AI-generated content.

AI and machine learning will only continue to develop and improve, promising great benefits in creative fields, science and technology.  These issues around ownership and protection of copyright need to be resolved soon – so that clever people and cutting edge organisations are incentivised to invest in the creation of AI.


[2] The latest edition of the Compendium of U.S. Copyright Office Practices, Third Edition.

[3] Infopaq International A/S v Danske Dagblades Forening C-5/08.

[4] Telstra Corporation Limited v Phone Directories Company Pty Ltd [2010] FCAFC 149.

[5] [2012] FCAFC 16.