Can Artificial Intelligence create better art than humans?
“I think those who are unaware of the technology are most vulnerable,if you don’t know about it, you may fall for it.” — Philip Wang
Let’s begin with what a generative adversarial network is.
A Generative Adversarial Network(GANs) is a machine learning model that is capable of generating realistic image, text, voice, and video. GANs are an exciting recent innovation. In 2014 Ian Goodfellow went to a bar in Montreal to celebrate with fellow doctoral students. His colleagues asked for help with a project they were working on to get the computer to create realistic photos. Ian Goodfellow’s idea was to use two neural networks against one another. The first neural network is a discriminative network with the task of classifying images as fake or real. The second neural network is a generative network with the task of producing realistic images. After Goodfellow returned from the pub, he coded the first example, and surprisingly it worked on the first try.
This person does not exist
In 2019 the artificial intelligence-powered website This Person Does Not Exist went viral. Each time the page is refreshed, a person’s face will appear. None of these people exists. Nvidia researchers developed a software StyleGAN. Uber engineer Philip Wang used StyleGan to create This Person Does Not Exist. The idea was to get attention to Artificial Intelligence’s ever-increasing ability to present realistic artificial images. In our society images, and pictures are the surrogates for proof. It can be scary to know what a machine can create and to lose the ability to recognize what is real or not.
“I think those who are unaware of the technology are most vulnerable,if you don’t know about it, you may fall for it.” — Philip Wang
After This Person Does Not Exist went viral, Ryan Hoover created This Cat Does Not Exist. Christopher Schmidt launched the website This Rental Does Not Exist. The image of the room, the description, and the location all look realistic. It could be a rental on Airbnb. To see more examples of this does not exist, check: This X Does Not Exist.
How is it possible to create something from nothing with GANs?
As early mentioned GAN does have two neural networks that competing against each other. The neural networks are called a Discriminator and a Generator. In the architecture of the GANs, as described in the architecture down below. The generator needs input to create fake images, the same as a book needs paper to become a book. It is called noise, random noise, or noise array. The discriminator has two inputs real images and fake images. It is required to train the discriminator to recognize what is real or fake. Eventually, the discriminator will label the images as real or fake. The generator operates as a counterfeiter who is printing fake dollars, and the discriminator is the cop, who is trying to track down the fake dollars. Every time the counterfeiter gets caught with the fake dollars. He improves his printing until the cop can not identify the real or fake.
The architecture above has a minimax loss. So, Generative Adversarial Networks has a game-theoretic approach.
Applications of Generative Adversarial Network
To give you a good intuition of the impressive applications of GANs. The upcoming applications are divided into the following area (these are just a few examples):
- Image-to-Image Translation
- Text-to-Image Translation
- Photos to Emojis
- Video Prediction
Image-to-Image Translation
An Image-to-Image translation transformed an input image into a synthetic image or maps an input image to the desired output image.
Text-to-Image Translation
A Text-to-Image translation transformed a text into an image.
Photo to Emojis
A Photo-to-Emojis translation transformed a photo into emoji.
Video Prediction
A video is a collection of frames. Video Prediction is trying to predict to future frame.
$432.500 Painting created by Artificial Intelligence Sold
In 2018 at Christie’s today, a painting is sold that is created by Artifical Intelligence. The painting is the first art of Artificial Intelligence sold at a major auction house. The estimated price was between $7.000 and $10.000. It is funny to see the loss function of the GAN instead of the artist name.
A dataset of 15.000 paintings between the 14th and 20th centuries has been used to train the GAN. There was a discussion on Twitter about the originality of this painting. Due to the dataset that is been used for the training process.
“AI is just one of several technologies that will have an impact on the art market of the future — although it is far too early to predict what those changes might be,” said Christie’s specialist Richard Lloyd, who organised the sale.
Why is Generative Adversarial Network so impressive?
“The most interesting idea in the last 10 years in Machine Learning” — Yann Lecun
Yann LeCun is known as one of the fathers of deep learning. Of course, there must be high potential in Generative Adversarial Network — when it comes from such an honored researcher in the deep learning area.
The applications are increasing fast, since the invention in 2014 by Goodfellow. Especially in fashion, art, and advertising, also in astronomy, medicine, and video games. For example, Generative Adversarial Network can create imaginary fashion models. In Astronomy GAN can improve astronomical images. GAN has the ability to create molecules from scratch.
Further, GAN is able to create new data that is a representative of the original data. In cases where sensitive data is involved, the generated data would be a better option to analyze. The same applies when there is a shortage of data to analyze. For missing data, creating new data can be a solution to this issue.
As far as we know now, we will see more of the applications of GANs in the future.
Can Artificial Intelligence create better art than humans?
To answer this question, we need to know: What is art?
According to the Oxford dictionary, art is defined as:
“The expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power.”
It is difficult to define art; there is a field of philosophy dedicated to the study of art and beauty. What it also makes it difficult to answer this question is that Artificial intellingce learns from human art. We can say that Artificial Intelligence is connected to the human creativity.
“Artist don’t panic.” — Marcus du Sautoy Oxford mathematician
That is why I have left this question open, in the hopes that you will be able to answer it with all of this information. Share your answer in the comment and let us know, how it makes you feel what Artificial Intelligence already capable of is.
My recommendations:
