Creativity and the Machine

Updated: Sep 25, 2018

By Liv Gunhild Fallberg.


From a distance you can see a traditional portrait painting of a gentleman, placed in an engraved golden frame. This is Edmond De Belamy, the great grand-son in the Belamy family. The painting looks like a sketch, because the edges of the canvas are left unpainted, and Belamy’s face does not appear finished. You look closer, and notice that the face actually looks quite distorted, and you can see some repetitive marks in the background, as if the artist has dotted the brush in some kind of cloned pattern. “Who can the artist of this remarkable painting be?”, you may wonder, and look for a signature down in the far right corner of the picture. There you will find no artist name, but an algorithm.

This is not the beginning of a Dan Brown mystery. What you are looking at, is the first picture set to be sold at an auction house, made by an algorithm [1]. It is to go up for auction at Christie’s on October 23. The French group Obvious, consisting of Pierre Fautrel, Hugo Caselles-Dupré and Gauthier Vernier, is behind this work of art. They are a collective group working with art made by Artificial Intelligence (AI), that is the manufacture of intelligent systems [2]. This means that Edmond Belamy, and the ten other painted Belamy-family members, are fictional characters, produced by a computer.


It all started a year ago, when the members of Obvious came across the Generative Adversarial Network (GAN). It is a machine learning program, introduced in 2014 by Ian Goodfellow and his fellow researchers [3]. In short, the GAN consists of two networks, the first is called a generator, and the second is called a discriminator [4]. The goal to the generator is to fool the discriminator, and the goal to the discriminator is to spot the fakes made by the generator.


In making a GAN artwork, you give the generator X numbers of pictures, in this case 15 000 portrait paintings from the 14th to the 20th century [5], and the generator produces its own picture based on these. Then this image is sent to the discriminator, together with a number of other similar ‘real’ images. The discriminator marks all these images as either real/human made, or fake. If the discriminator cannot detect the fake image produced by the generator, the generator have succeeded in fooling it. And if it gets detected, the generator makes adjustments to the picture, and the sequence goes back and forth in a corrective feedback loop, until the discriminator is fooled. The networks learn from each other, and make each other better at their own task.


What happens with our understanding of an artwork when it is made by an algorithm? Can AI be creative? The members of Obvious are interested in these questions. Already in 1958, the political scientist and pioneer in development of AI, Herbert Simon [6], claimed that: “There are now in the world machines that think, that learn and that create. Moreover, their ability to do these things is going to increase until - in the visible future - the range of problems they can handle will be coextensive with the range to which the human mind had been applied” [7]. He saw the potential of the machine as a substitute for human functions in organisations[8], and among this lays the notion of creativity.

The understanding of creativity is important in the separation between humans and machine. All the things we humans see and experience around us through our whole life, influence us when we make art. We make something new and creative on the basis of our knowledge. This is how the GAN operates too, only it has a much smaller repertoire of influences and knowledge. That is why it has to be trained, to get better at making art. The programmer set som guidelines and frames, and based on this, the generator creates new art. When this machine makes paintings that have never existed before, is that creativity? The image generated by the GAN is not a merging of the input images, but a completely new image. The result is never the same.


On Christie’s webpage, the Obvious group explains that they chose to make portraits in their first art-series, similar to those paintings we are all familiar with from the museums, to illuminate the question of creativity by algorithms [9]. They show us that an AI can make similar paintings to those in museums, only printed with inkjet on canvas. The machine needs some training, but eventually we may not be able to recognise the difference between a human and GAN artwork.


On the contrary, John Smith, Manager of Multimedia and Vision at IBM research, says that we cannot really learn a machine creativity, because we do not really know what creativity is [10]. The GAN can make something new, the generator can fool the discriminator, but is that creativity? Smith says that the goal should not be to make the AI creative, but that we can use technology as a kind of assistant, that helps us be more efficient in our own creativity [11].


Obvious are not the only ones working with GANs to make art. Among others is Robbie Barrat, who uses AI to make nude paintings [12]. But they are unlike any nudes we have ever seen before. Contrary to the Obvious portrait paintings, these nudes are almost unrecognisable variations of shapes and colours. The nudes look like some kind of deformed hallucinations, or ‘blobs’ [13]. But at the same time, they somehow resemble nude paintings. That is because they have all the key elements, the usual light skin-tone, the reddish ornamented fabric-background. We can sort out some body-shapes too, like a torso, a leg. Even thought humans can see that this is not a ‘real’ nude painting, it fools the discriminator. Barrat’s generator have learned that to fool the discriminator, it does not need to produce perfect human forms, which is much harder than only producing different blobs in the right colour. For the generator, the goal is not to make a perfect nude, but to make something that can fool the discriminator. This opens up for surprising results, and a whole new world of possibilities. When discussing AI, we tend to focus on the things humans can do or make, that machines cannot [14]. A more interesting question may be what machines can make that humans cannot.


Machine learning are more available than ever before, it has become inexpensive, and therefore more people have access to experimenting with it. There are even an annual global robotart-competition [15]. The GANs have only been around for a couple of years, so it will be exciting to see where all of this leads. At the auction at Christie’s in October, it is estimated that the algorithm-signed Belamy-portrait will sell for $7-10 000 [16]. For now, it will be interesting to see what the public thinks of the painting.



Notes

[1] Ifeanyi, “This AI-generated artwork is about to make history.”

[2] Obvious, “A naive yet educated perspective on Art and Artificial Intelligence.”

[3] Goodfellow et al., “Generative Adversarial Networks”.

[4] Skymind, “A Beginner's Guide to Generative Adversarial Networks (GANs).”

[5] Bastable, “Is artificial intelligence set to become art’s next medium?”

[6] Dasgupta, Multidisciplinary creativity: the case of Herbert A. Simon.

[7] Barber, "The Arts of The Natural: Herbert Simon and Artificial Intelligence,” p. 338

[8] Ibid., p. 336

[9] Bastable, “Is artificial intelligence set to become art’s next medium?”

[10] IBM, “The quest for AI creativity.”

[11] Ibid.

[12] Bailey, “AI Art Just Got Awesome.”

[13] Ibid.

[14] Barber, "The Arts of The Natural: Herbert Simon and Artificial Intelligence,” p. 338

[15] Robotart, “The Robot Art Competition and Exhibition!”

[16] Ifeanyi, “This AI-generated artwork is about to make history.”


References


Barber, Walter F. "The Arts of The Natural: Herbert Simon and Artificial Intelligence." Public Administration Quarterly 12, no. 3 (1989): 329-47. http://www.jstor.org/stable/40861426.


Bailey, Jason. “AI Art Just Got Awesome,” Artnome. 04.05.2018. https://www.artnome.com/news/2018/3/29/ai-art-just-got-awesome.


Bastable, Jonathan. “Is artificial intelligence set to become art’s next medium?” Christie’s. 09.21.2018. https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx?sc_lang=en.


Dasgupta, Subrata. Multidisciplinary creativity: the case of Herbert A. Simon. Institute of Cognitive


Science, University of Louisiana at Lafayette. 2003. http://csjarchive.cogsci.rpi.edu/2003v27/i05/p0683p0707/00000130.PDF.


Goodfellow, Ian, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,


Aaron Courville and Yoshua Bengio. “Generative Adversarial Networks”. Cornell University Library. 06.01.2014. https://arxiv.org/abs/1406.2661.


IBM. “The quest for AI creativity.” 09.22.2018. https://www.ibm.com/watson/advantage-reports/future-of-artificial-intelligence/ai-creativity.html.


Ifeanyi, KC. “This AI-generated artwork is about to make history.” Fast Company. 08.22.18. https://www.fastcompany.com/90224502/this-ai-generated-artwork-is-about-to-make-history.


Obvious. “A naive yet educated perspective on Art and Artificial Intelligence.” Medium. 06.26.2018. https://medium.com/@hello.obvious/a-naive-yet-educated-perspective-on-art-and-artificial-intelligence-9e16783e73da.


Obvious. “Home.” 09.22.2018. http://www.obvious-art.com/index.html


Robotart. “The Robot Art Competition and Exhibition!” 09.21.2018. https://robotart.org/.


Skymind. “A Beginner's Guide to Generative Adversarial Networks (GANs).” 09.21.2018. https://skymind.ai/wiki/generative-adversarial-network-gan.

OSLO FORM LAB 2018