It was put up for auction at Christie’s along with over 300 other lots, including works by Andy Warhol. It was expected to sell for about R150 000 which would seem like a great return for a print created by a machine using mostly publically available code.
Yet it sold for more than the Andy Warhol piece (which sold for R1,3 million), surprising everyone and earning it its 15 minutes of fame.
Humans have a hard time defining art and machines have no idea what it is, so for a machine to create something that can not only be presented as art but sold for such a large sum warrants an explanation.
This type of AI uses a program called a Generative Adversarial Network (GAN), which are effectively two programs that compete with each other in performing a task.
The point is to use one program to learn what something is, in this case, a portrait by creating a dataset of 15 000 images of portraits from the 14th to 20th century. Once it has identified what makes a portrait it generates a version of its own.
The second program looks to do the same thing but rather than create a new image it tries to spot the image generated from the other program from the actual portraits. If it can, the image is discarded and the other program needs to create a new one.
When the newly created image can’t be distinguished from all the other real portraits the program is considered complete and the image above was the what resulted.
To a human, I think, you would not need long to tell that the portrait is at best not a very good one. It looks more like a failed restoration by someone who did not know what they were doing like the Ecce Homo.
Ecce Homo – Nailed it
While this may appear to be a silly use of clever computer technology, the real use of this technology is how it works with all the other types of machine learning and adding to tools to replicate the functions that our brains perform all the time.
In human terms, when a child is learning to write, it is best that an adult not only show them how to write but also point out when and how their early attempts can be improved until it is sufficiently neat to be read and understood.
Getting a machine to learn in the same way is not easy, but Generative Adversarial Networks (GAN) can and what they produce or what they can detect could either help create something, like a model for a brand that is not an actual person, or spot a fake, like a picture that has been manipulated.
For artists, it could one day create movie actors or dancers that are not real but able to perform and appear as if they are. You may think it will ruin things for real actors or you could argue that it will allow many more stories to be turned into movies and the ones that turn out to be really popular may be remade with real actors.
For now though don’t be surprised if there is a spike in AI artworks sold at auction and maybe even some real artists claiming their works are actually created by a machine.