As of now, I have not mentioned the implications of AI for the art world. According to certain experts, most of the strides that have been made in “AI art” recently, have been due to a type of Artificial Neural Network called a Generative Adversarial Network.
According to certain industry insiders on the subject, a GAN is different from a traditional Artificial Neural Network, beginning with the fact that it depends on an approach that is derived from Game Theory and a traditional ANN does not. In order to easily understand how this plays out, Rohith Gandhi, in a Medium post on the subject, asks us to imagine the AI that is structured as a GAN being in a two player game that never ends. He defines the two players as the “Generator,” and “Discriminator.” Inside of this analogy, Gandhi appears to explain that this particular game is essentially a never-ending battle between these two sides of the AI system, to determine who has the best answer to the problem at hand. Even though they are, by definition, adversaries, both the Generator and the Discriminator have something consistent to contribute to the functioning of a GAN. The Generator works on creating or “generating,” images that look real as if they are a person’s art, and the discriminator tests them by deciding based on certain criteria if they are real or fake.
To do so, the discriminator puts the input picture or “data” through what is called an objective function. In short, an objective function is a type of mathematical function, which in the context of the AI industry, is used to maximize or minimize data in specific ways. As the system is doing this, the Discriminator is also giving the Generator feedback related to how it is performing. From this feedback, the Generator decides how and where the entire GAN needs to improve its processes.
Over time, these GAN systems will continue to function in this way to produce the best art that can be produced. For most, the overarching goal appears to be to optimize the art that they produce so that one day, people will not be able to tell the difference between it and art that is actually created by humans. If you’re interested in seeing what this looks like, check out the art section below and stay tuned for our future pieces on the subject.
In addition to this, for a more in-depth look at GANs and how they work, check out the links from Rohith Gandhi and others, below. Keep in mind that these pieces will be written with much more esoteric language and that here, as always, we have tried to simplify their main points to fit into our discussion.
Stay tuned for future contemporary news articles on how AI Art is taking form right now, thanks to GANs.
Resources:
Primary Source:
https://towardsdatascience.com/generative-adversarial-networks-explained-34472718707a
More on GANs:
Examples of AI Generated Art:
https://www.artnome.com/news/2018/3/29/ai-art-just-got-awesome
https://www.cbsnews.com/pictures/art-created-by-artificial-intelligence/
Create AI Art with Your Photos:
