In a world in which machines take over human handling, creativity is what separates us from machines… right? As technology challenges the way art is perceived, Christie’s will become the first to sell an artwork created by an algorithm.
Portrait of Edmond Belamy, 2018, created by GAN (Generative Adversarial Network - © Obvious.
The Portrait of Edmond Belamy, a fictional dark-haired man, dressed up in a black shirt with a simple white collar, has been painted with rough brushstrokes that reminds of the Dutch Golden Age portraits painted by Rembrandt van Rijn. However, the work misses the realism of Rembrandt’s portraits and those of his contemporaries. The painting is even more abstract and the depiction is slightly distorted, creating a vague image. Even though the canvas appears to be unfinished, the artist took the time to sign the work regardless. However, the signature is not a name nor a symbol, but an algorithm:
Close up of the signature of Portrait of Edmond Belamy, 2018, created by GAN (Generative Adversarial Network - © Obvious.
In my previous blog post Masterpieces Remastered: Rembrandt in the age of technical reconstruction I briefly introduced the Rembrandt Research Project (2016) by ING Bank, in which artificial intelligence (AI) was used to generate a ‘new/final’ painting by Rembrandt van Rijn. The project was initiated to see what science and technology can create and how these techniques may or may not challenge art history. However, this painting was an experiment and has therefore never been sold. However, on the 26nd of October 2018, the first artwork created by an algorithm was sold via an auction at the famous auction house Christies in New York.  The estimates for the artwork’s price ranged between $7,000-10,000, but the painting was sold for almost half a million dollars, which is more than works by Andy Warhol and Roy Lichtenstein were sold for at the same auction. 
The sale of an artwork that is in its creative process and execution is non-human or ‘traditional’ for a price higher than most practicing artists will ever get for their paintings prompts various questions: can computer-generated images be considered artworks? Do algorithms generate authentic artworks in any way? What is creativity? This blog post will therefore give a small introduction in understanding what creativity is and how scientific creativity challenges the way human creativity is considered. In this post I want to give an introduction into an era in which computer technology will inevitably challenge the way art is made, the way originality is perceived and how it makes us think about what creativity is. It helps us to define whether or not ‘artificial creativity’ is indeed a contradiction in terms, or if it is more similar to human creativity than initially thought.
The creation of an AI-artwork
In order to explain the technique used to make artworks like the Portrait of Edmond Belamy, (AI that uses Generative Adversarial Networks (GAN), a method that uses existing images to generate a new image) Hugo Caselles-Dupré, PhD student in machine learning, says in an interview with British author Jonathan Bastable the way the French Obvious group created the portrait as follows: “We fed the system with a data set of 15,000 portraits painted between the 14th century to the 20th. The generator makes a new image based on the set […].”. To explain AI in more detail: one machine - the generator – is handed a dataset of an X amount of information, in this case pictures and images, which presents the information from the dataset to a second machine – the discriminator – and simultaneously tries to fool it. The discriminator is taught to detect the fakes the generator has produced or files that do not belong. These machines are tweaked to recognize one or more characteristics or features in the images inserted in the system. The images are selected and ordered accordingly. In this process, the networks which are used to order the images constantly learn from each other and make each other better at their own task. Ultimately, the result comes out in the form of an algorithm, which can than be converted to a unique image. 
Generative Adversarial Network (GAN) scheme – made by author.
In understanding how an AI’s creativity is similar to or different from what makes human creativity, it is important to understand what creativity is. To this question, there is no simple answer (if there even is one) and the discussion about the process of creativity is too elaborate to be covered in this article. However, reading the collection of articles about this discussion in psychologists Robert Sternberg and James Kaufman’s The Nature of Human Creativity (2018) reveals that it is certain that creativity is considered a highly human characteristic that can only be executed and understood by human beings.  Over the course of our lives, all the things we experience and see are saved in our own intelligence network, our brain. Based on our knowledge, we are able to select and combine the images, emotions and experiences of our memory into something new and unique. This is what we often consider to be the essence of creativity. 
Nevertheless, looking at this process in a straight forward way, the way human beings construct images is not that much different from how AI’s generators and discriminators work: AI systems assemble existing information and transform it into something new. However, one must keep in mind that AI systems do not have a lifetime of impressions, images and emotions to recombine. They have a more limited selection of information and this information still has to be handed to them via humans. The parameters, frames and guidelines applied to the algorithm are still in our hands. 
Nude, Robbie Barrat, March 28, 2018, created with GAN.- © Robbie Barrat
Nonetheless, it is undeniable that these algorithms are able to produce something new and original via recombination, just like humans do. Is the creation of something original enough to define something as a product of creativity?  Reading The Nature of Human Creativity and the vast amount of tweets and newspaper articles on the sale of the AI artwork reveals that originality is seen as yet another key element that is part of the creativity process. As explained before, AI is a method which is constructed through sharing and assembling information on the web. A lot of comments on the Belamy artwork are about the fact that it is not original as it used a code from Robbie Barrat, another artist who used AI to create artworks.  The GAN system that created the portrait of Edmond Belamy is therefore seen as a fake, unoriginal and uncreative.
However, this discussion reminds us of a classical phenomenon in art and the understanding of creativity: the idea of translatio-imitation-aemulatio, and of a more temporary example given by Pablo Picasso: “Good artists copy, great artists steal”. Looking at AI systems from this perspective means that the AI system of Obvious would perfectly mimic and exemplify what creativity is: taking existing information (translatio), interpreting and reproducing (imitatio) and to ultimately rearrange it to create something new and unique (aemulatio). Besides, Robbie Barrat - the creator of the nudes – exemplifies that human artists like Sol LeWitt work in a similar mathematical way of reassembly-translation-creation like AI does.  Therefore, one could say, that the algorithm indeed creates a product of creativity and is not at all as artificial as it seems.
Clearly, the creation of an AI artwork causes a lot of commotion. The problem in deciding whether or not a mechanical system is creative or original lies in the fact that we cannot define the terms and conditions of what makes something a product of creativity. Besides, the discussion on the intervention of science in humanities is mostly focused on the things humans can do, but digital and mechanical systems cannot do (yet). What can be concluded however, is that machines are good at mimicking human behavior and that they will inevitably get better at it. What would be interesting to explore in the future, is approaching these techniques in a different way and discovering what technology can do that humans cannot. This way, we will be able to understand that these techniques are useful tools that help us to use our own creativity more efficiently.
 Ifeanyi, K.C., “This AI-generated artwork is about to make history.” Fast Company, https://www.fastcompany.com/90224502/this-ai-generated-artwork-is-about-to-make-history, published 22nd of August, 2018 (accessed October 23, 2018)
 Cohn, G., “AI art at Christie’s sells for $432,500”, The New York Times, October 25, 2018 (accessed October 29, 2018)
 Bastable, J., Is artificial intelligence set to become art’s next medium?, New York: Christies, https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx?sc_lang=en#FID-9332, published (accessed October 23rd, 2018)
 Goodfellow, I., Bengio, Y., Courville, A., “Deep Generative Models”, Deep Learning, Cambridge: MIT, 2016, pp. 646-694.
 Sternberg, R.J., Kaufman, J.C., The Nature of Human Creativity, Cambridge: Cambridge University, 2018.
 Ibidem, pp. i-xxii.
 Brown, P., Husbands, P., “Not Intelligent by Design”, Art Practice in a Digital Culture, Farnham: Ashgate Publishing, 2010, pp. 65-92.
 Runco, M.A., “Authentic Creativity: Mechanisms, Definitions, and Empirical Efforts”, The Nature of Human Creativity, Cambridge: Cambridge University, 2018, pp. 246-260
 Barrat, R., https://twitter.com/DrBeef_, 2018 (accesed October 29, 2018.
Cameron, Fiona, Theorizing digital cultural heritage: a critical discourse, Cambridge (MA): MIT, 2007.
Gardiner, H., Gere, C., Art Practice in a Digital Culture, Farnham: Ashgate Publishing, 2010.
Hoy, Meridith Anne, From Point to Pixel: A genealogy of Digital Aesthetics, Hanover: Dartmouth College, 2017.
Sternberg, R.J., Kaufman, J.C., The Nature of Human Creativity, Cambridge: Cambridge University Press, 2018.
© Liselore Tissen and Leiden Arts in Society Blog, 2018. Unauthorised use and/or duplication of this material without express and written permission from this site’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Liselore Tissen and Leiden Arts in Society Blog with appropriate and specific direction to the original content.