AI Art

Authors: Go Chengdu


Art is said to be the embodiment of human experience, the synthesis of our emotions and existence. That’s why you might ask: Is it really possible to simulate the creation of art using a digital neural network, or machine?
The greatest artists of our time have their distinct styles, ranging from realism to impressionism to contemporary abstract. One popularized form of AI Art is style transfer. There are many mobile apps that can do this easily right on your phone, applied almost instantly to the photos you have taken. In effect, there are millions and millions of AI artists in the world, but perhaps because of the commonality and accessibility of it all, we might not typically associate any of that with what we normally ascribe to the term "artist".
Computer scientists have been creating algorithms and training neural networks for a long time and the artistic use of these creations has led to the formation of a new genre of modern art: AI Art. Much of the code used to create this AI Art is open source and readily accessible to anyone with a modicum of experience with computer use and programming. People might say then; how can this be a serious form of art, when virtually anyone with a phone or computer can do it?
They would be right in many ways, but it is not only with AI Art that we can say that. The tools to create art have been readily accessible to anyone at any time over the course of history. A pencil, piece of paper, piece of chalk, a paint brush; all of these, almost anyone can access at any time. So the perceived degree of how "special" an AI Artist really is, again, comes down to what all great art always comes down to; and that is the one inescapable reality about art since the time of cave drawings, which is the ability of the artist to create something unique and desirable enough in the world of art with the tools available to them.
In essence, "AI artist" sounds like a cool title, but is really just a descriptor in the same way that we have realism, impressionism, abstract artists, contemporary artists and the like.
The world today also has several robot painters which use varying degrees and mixes of image recognition, style transfer, neural networks and algorithms, for example CloudPainter (which won the $40k 1st prize in RobotArt's 2018 version of its annual international competition), eDavid and Picassnake (which uses music and sound frequencies to create output strokes). There are international art competitions among the developers of these painters with huge prize money at stake. Some of the results are very impressive, yet the general feeling so far is, that even with the top experts at Google DeepMind, coming up with the best in AI advances, AI Painters have not reached the level of human artistic endeavor.
Headlines were made in 2018, when what was purported to be the world's first AI generated painting sold at Christie's for $432,500 USD. The trouble was, the code used to create it was apparently open source and so the attributions and proper credit for the artistic creation were cast into doubt, creating a storm of controversy.
A lot of AI Art out there today is based on "GANism", or the use of a Generative Adversarial Network and one of the main issues with the use of this technique is the source of the data used to train the model. Many AI artists readily admit to "web-scraping" the images used to train their models because they require huge datasets to be effective. That means that they took thousands of images off of the web and ran them through their coded neural networks (which may have also been borrowed) to create their artwork. There are alot of fine lines in there, perhaps being crossed or not, that need to eventually be addressed. It is a new frontier.
That said, with time, we will certainly see modern art museum exhibits featuring an art exhibition solely of works of art created by a machine. There may even come a day when cybernetic organisms stroll through art galleries filled with ancient paintings created by humans and comment to each other "how quaint".
Each week here in this column, we give some practical, real-world commentary about art, rather than paraphrasing historical or contemporary information on the web. We leave the research to you, but we'll give you some search terms related to the column topic, for you to check out on your own.
1. Edmond de Belamy, from La Famille de Belamy
2. Memories of Passersby 1
3. DeepMind
Also, each week, the writer will feature one of his own works, or provide feedback and commentary on any submitted works by other artists, whether new, emerging or otherwise. Such feedback will be practical and constructive and based on the writer's real-world experience of selling artwork in today's marketplace. As of publication, that would be 314 original works of art in the past 35 months.
If you are an aspiring artist, art student, emerging artist, or fellow professional artist and you wish to have your work reviewed for some constructive feedback, please send a photo or representation of your work to the writer's WeChat (see QR Code) and perhaps your work will be selected for a published review. Please note, not all submissions will be able to be published, however, the writer will endeavor to answer all requests by private message.
This week the writer features his own collaborative AI works "Scenes of Chengdu" and "Primordial", the first of which won a Golden Panda Design Excellence Award in 2018 at the Golden Panda International Design Week sponsored by iF Germany, one of the leading design organizations in the world and the second of which featured in an experiential mini-theater at the 2019 Golden Panda International Design Week in Chengdu and is the world's first AI Perpetual Painter to use a sole source dataset (NOT web-scraped). The neural networks were coded by the writer's son and AI Artist Nathaniel Maxim Vegh and the model was trained using tens of thousands of images of the writer's own abstract artwork. That means the work is proprietary and does not include or use the imagery from any web source or other artistic source to create the final output, which is an important distinction of the work when compared to what is out there on the market.
Title: Scenes of Chengdu
Sizes: 15x25cm (original photograph)
30x40cm (original abstract canvas)
15x25cm (AI Generated Giclee painting)
Scenes of Chengdu is a collaborative AI work between the writer and his son Nathaniel Maxim Vegh for entry into the 2018 Golden Panda International Design Week Competition. The artists took photographs of typical scenes around the city of Chengdu, applied a style-transfer neural network to the photograph using the abstract artwork as the style sample, then output the result by printing a giclee painting on canvas.
Title: World's First AI Perpetual Painting "Primordial"
Size: 800mm spherical display
Composition: Generative Adversarial Network (GAN)
The AI Perpetual Painting "Primordial" was created using a combination of the results of human made abstract art, more specifically, tens of thousands of images of the abstract work of Canadian artist Matthew Vegh and an AI based machine learning algorithm, more specifically a coded GAN (Generative Adversarial Network), the work of Canadian Data Scientist Nathaniel Vegh. The result of this combination is an AI based system that allows the perpetual creation of new abstract artwork for digital display, in this case, within an experiential mini-theater environment, augmented by decorative sculpture work to suit a specific theme.

Matt Vegh is one of China's most widely sold contemporary artists. He is also the curator of the M Gallery International, located in Chengdu. The M Gallery International returns to simpler times with a focus only on "Beautiful Art".
Share to WeChat Close
Scan via WeChat!

About Us | Privacy Policy | Terms of Use | Link to Us | Contact Us

蜀ICP备14024942号-1 Go成都