Do you recognise these people? The answer, for everyone in the world should be ‘no.’ You shouldn’t be able to recognise them, because they do not exist.
These are computer-generated photographs of people, created by a Style-based Generative Adversarial Network (GAN). Based on work by researchers from Nvidia, the system analyses photographs and is then able to recombine them to generate new faces.
We might argue that this artificial intelligence (AI) process is exercising a form of creativity. It is creating ‘new’ and ‘novel’ things. (Check the Wikipedia page for ‘Creativity’ to see how diverse the definitions have become.)
The artist known as…
On the 25 October 2018, at Christie’s auction house (New York) the ‘Portrait of Edmund Bellamy’ was sold for $432,000; a price more than 40 times higher than it had been expected to reach.
This enigmatic work, on canvas, suggests something of the work of Francis Bacon. It is one of a series of portraits of the Bellamy family. In the common position for the artist’s signature, in the lower right corner, we are presented with:
What does this mean? Who is this? A good deal of the art world was incensed by this ‘signature.’ Well, this is where things get a little complicated.
The work is by a French artist collective called Obvious. Pierre Fautrel, Hugo Caselles-Dupré and Gauthier Vernier are artists and AI researchers who use GANs to create ‘original’ artworks. Their work, beyond the aesthetic character, challenges us with questions about the nature of creativity and the role of the artist.
In ‘signing’ the work with an algorithm, Obvious were trying to suggest that the AI process was ‘the artist’. Both traditional and other AI artists found this disingenuous.
Obvious are not pioneers in the realm of art that is generated through some level of artificial intelligence. One of the first was William Latham, who worked with IBM on the creation of FormGrow, a system that uses biological ‘rules’ to create 3D forms.
Some have suggested that Latham’s work is more ‘evolutionary’ than AI, since there is a process of the ‘artist’ making selections as to which outcomes will be prioritised within the ongoing process of generation. Latham has referred his process as ‘creation’ and ‘gardening.'
The difference between Latham’s work and the GAN process used by Obvious is that in the AI-driven work (Obvious) the computer is left to generate the work without human intervention.
However, AI artist Robbie Barrat, points out that in the use of GAN to generate artwork, there are a number of human ‘interventions’. The Bellamy Family portraits are heavily influenced by Barrat’s earlier work Portrait GAN.
In these works, the AI is provided with a large data set of existing images, to analyse and ‘learn from’. As Barrat points out, there is a good deal of human intervention in this AI work. The human is responsible for:
- selecting the data set
- designing the network
- training the network
- curating the outputs.
This further raises questions about where the ‘artist’ resides within the production of AI art.
Between creative process and the execution
Some years ago, I found myself in a discussion about whether a person without arms could be considered a sculptor if they hired someone to make their ideas real. Is the ability to sculpt the material (clay, wood, stone) a prerequisite for being the artist?
At the time, and today, I argue that the ‘art’ resides in the idea of the work, which is separate from the making of the work. In modern art, it is not uncommon to find that the ‘work’ has been created by a team of ‘assistants’ and not the ‘artist’. Andy Warhol’s studio, for example, was referred to as The Factory, in part because the work was mass-produced by a team, and not Warhol himself.
Even earlier, in 1914, Marcel Duchamp challenged the relationship between the making process and the artistic process in ‘Bottle Rack’, and in 1917 ‘The Fountain’.
In both of these works, referred to as ‘readymade’, Duchamp did nothing to the existing objects, other than to sign them and set them for display. The fountain was signed ‘R Mutt’. It has been written:
Whether Mr Mutt with his own hands made the fountain or not has no importance. He CHOSE it. He took an ordinary article of life, placed it so that its useful significance disappeared under the new title and point of view – created a new thought for that object. (Norton, Louise. The Blind Man, Vol. 2, 1917)
Given this, we may argue that the AI works are indeed works of art, but that the artificial intelligence is not the artist, since those who design and train the GAN are making the creative decisions.
Listening to the data
There is a rumour that some of the most popular playlists on Spotify are of songs written by AI, in order to save the company from paying out millions in royalty fees. Spotify denies this, but they did (in 2017) hire AI researcher François Pachet, who had previously worked at Sony developing an AI that writes music.
DADABOTS, Zack Zukowski and CJ Carr, have developed a neural network that generates ‘black metal’ and ‘math rock.' While the music will likely only appeal to the most ardent followers of ‘black metal’, it remains an interesting experiment in the development of artificial intelligence systems that model themselves on biological learning.
It remains the case, however, that Zukowski and Carr are exercising some form of creative practice in their choice of the samples that they use to ‘teach’ the neural network.
While they cannot predict what the AI will generate, they are also curating the output, selecting what to include on their albums, and what to feed back to the system. To some extent, their process will reinforce those attributes that they find ‘good’ or ‘valuable’. It remains ‘their’ art, with the AI being the production mechanism.
Art in the age of AI production
In 1935, German philosopher Walter Benjamin wrote “The Work of Art in the Age of Mechanical Reproduction." In this essay, Benjamin argued that the use of technologies was causing a perceptual change in art, due to the emphasis on speed and reproducibility.
It was his contention that the value (the ‘aura’) of the work of art existed in its presence in time and space. And, this aura was in decay because technologies made it increasingly difficult to locate the work within time and space.
In simple terms, a painting exhibited greater ‘aura’ because it was singular and unique (in time and space). Whereas a photograph, due to its ability to be infinitely reproduced, becomes denuded in value because we cannot say with any certainty which is the ‘original.' Benjamin’s work has informed art history, architectural theory, cultural and media studies throughout the 20th and 21st centuries.
Where does the ‘aura’ exist in AI-generated work? An extremely prolific artist, using traditional or even digital methods, can only produce during waking hours and is limited by the physical act of ‘making’. An AI can generate millions of individual works as long as the hosting computer is operational. Does this mean these AI artworks are further distant from the unique or can we reclaim the ‘aura’?
Chris Peters, a former software engineer and AI artist, uses a GAN trained to analyse and create 19th-century landscapes. He then re-paints the AI generated landscape using traditional oils. Peters suggests that this process allows him to better understand what the AI is doing, because he spends so much time examining the AI output as he paints.
There is a kind of ouroboros, or infinitely repeating cycle, in Peters’ process, as he trains the AI with paintings, the AI generates the landscape, and then he re-interprets through the act of painting a landscape.
The genie never goes back in the bottle
Art has never been static, unchanging. It is reinvented and redefined almost continuously. Duchamp opened the bottle by making us understand that art need not reside in the physical object, but can be in the idea or concept that we ascribe to the object. AI artists, however, may now have broken the bottle by reconfiguring the relationship between the generation of ideas and the generation of work.
Whatever you may think about the aesthetic qualities of AI art, there is no denying that the work has continued the age old debate about ideas, creativity, art and production. And, $432,000 is probably not the limit to what AI art will soon be fetching in the auction houses.