AI Generated Art [2021 Guide]
What is AI and How does it make art?
Artificial intelligence is basically any type of machine or computer that shows signs of intelligence - it can solve tasks and recognize/categorize things and data, for instance. What counts as intelligence changes as time goes on and AI becomes more capable.
The Turing Test was used for a while to truly determine whether machines could be labelled 'intelligent', though this is often not used anymore to classify AI, as it's almost assumed that most technologies developed now automatically pass this test.
Some people also take philosophical issue with the test, which has led some to classify their machines as 'intelligent' without using the test .
Some of the simplest examples of AI are printers that scan images and documents, your phone's assistant such as Siri or Cortana, and apps that auto generate responses to your customer service questions. You likely interact with these things on a daily basis.
The way that a machine becomes intelligent is through 'machine learning'. This simply means that its computer is programmed with an algorithm in which it can learn tasks and differentiation through evaluating a data set.
'Machine learning' was first used by Arthur Samuel, who designed a computer program that could play checkers against humans (and often won). This was first documented in 1959.
In the case of art, machine learning would simply entail programming an algorithm to process lots of examples of historic artwork. It would learn the similarities and over time learn to differentiate styles, artists etc. Over time, and with the more data it processes, it gets better at learning these differences and being able to apply them.
Beyond simple machine learning, there's deep learning. These are AI designed specifically to mimic the human brain with a series of neurons that send information between one another in order to learn and tell apart information, and to make decisions for itself. It can also teach itself.
One type of deep learning, which is frequently used to create AI art, is GANs, also known as General Adversarial Networks. This uses two types of neural networks modeled on human neurons, one called the 'generator' and the other the 'discriminator'.
In this sense, the generator creates data for the discriminator to evaluate and decide whether it fits the model of the other data. In terms of art, this would involve generating new art based on what it knows about art, including colors, brush strokes, and subject matter.
The discriminator would then decide whether this meets the parameters of what it already knows about art - e.g., whether it would pass as a realistic portrait.
There's also CANs, which are Creative Adversarial Networks. These use very similar networks to GAN's, but instead of trying to replicate a realistic version of a specific artwork genre, they aim to make it difficult to classify its new creation into a genre.
Essentially, it tries to create its own style that makes it difficult to classify. This usually results in more experimental or abstract art than that which comes from GANs.
Neural networks were first invented in 1943, when a paper by Pitts and McCulloch suggests that the future of computer intelligence lies in getting it to a stage where it replicates the human brain via nets of neurons.
The first time the term Artificial Intelligence was formally used was in 1956, where it was used for a conference about computers at Dartmouth University.
The conference saw various computer professors discussing the future of AI - whether we should be designing them to act like human brains and think for themselves, or rather to simply program as per the norms of human life so they can perform simple tasks.
The earliest forms of AI as we recognize it were probably the Alice chatbot and the computer playing chess. Alice came about in 1995, being able to hold something of a text conversation. IBM's chess computer beat the world champion chess player Kasparov in 1997, a move which is widely remembered now.
The subject of many millennial memes in recent years, the Roomba robot, was released in 2002. Roomba seems fairly simple but it was the first commercially available robot, which used simple AI.
These things are all examples of the concept we talked about earlier, in which once AI reaches a certain level of intelligence and capability, previous examples are no longer deemed 'AI' as they just become a part of our lives.
Technically speaking, each of these things is still AI at work - playing a game against a computer, for example - but we just don't think of it as such because technology has developed so rapidly since their inception.
Types of Artificial Intelligence Art
AI art is mainly seen in paintings, as they're one of the easiest things to teach a computer how to recreate. But, there's more to it than the traditional auction house!
Paintings are probably the first version of AI art, and still one of the most popular.
Popularized by the Paris art collective Obvious, paintings made by AI are popping up everywhere. There's now multiple online stores where you can have artworks specially made, just for you, by GAN networks.
The most popular AI paintings are portraits, as most are getting pretty good at creating new people based on traditional portraiture. However, there are also a lot of abstract pieces done by artificial intelligence, as well as renaissance paintings.
This category is slightly different to paintings, as they're not creating something new. But, there are now plenty of online programs in which you can upload an image and select an art style or genre, and the machine learning will alter your original image to make it look like a painting in that particular style.
Popular programmes include Google Deep Dream, which can both turn your images into a specific style, as well as create more abstract images from the visualization of its network learning what constitutes the image in each pixel. This often results in faces appearing where they shouldn't as well as psychedelic looking patterns.
There's also ArtBreeder, which merges two images, letting you control how much of each gets put into the newly formed image.
If you're a fan of Grimes at all, you'll be well aware that musicians have long been talking about the possibility of artificial intelligence being used to make music - whether that's in collaboration with humans, or as Grimes frequently espouses, taking over their role altogether.
In 2017, the first official AI-human collaboration album dropped. Taryn Southern created 'I AM AI' in tandem with an artificial intelligence program which helped her to compose the album. She also created the first song which let royalties be transferred between artists and fans with Ethereum.
Since then, there have been a couple more releases of ai generated music. In 2019, Holly Herndon released an album in which she harmonized using AI.
There are now a few programs that incorporate some level of neural network to help you make music.
AI Duet, for instance, creates a piece of music alongside you. You play a few notes on a keyboard, and the computer will respond and use your notes to create a melody.
Magenta, a program which allows you to explore machine learning with code, now has Magenta Studio, which are a series of music making plugins for the program.
Performance art, live art, and other types of non-static artwork have seen an uptake in the use of AI.
In 2018, Deep Meditations, a visual artwork by Memo Atkin, was produced using machine learning. The algorithm he used took images from Flickr that were tagged with things such as 'everything, love, life' etc. It then reproduced images based on what it knew from this data set, creating a 'machine's understanding' of human life.
2010 saw the theatre piece 'Hello hi there', in which two AI generated chat bots held a conversation based on a famous debate between philosophers Foucault and Chomsky.
Is AI art taking off?
Definitely. As the years go on, the sale and generation of neural network art is increasing slowly but surely. There are now entire websites dedicated to their sale, such as our own.
Portrait of Edmond de Belamy
The art collective Obvious, who we spoke of earlier, sold their first AI painting 'Edmond de Belamy' for $432,500 at Christie's in 2018.
No piece of AI art seems to have sold for as much since then, which may indicate that the initial craze is what drove such a high price. However, individual pieces of AI generated work are still pretty coveted, which is why dedicated websites featuring generated art have popped up.
Will it replace human artists?
Who knows! It's currently unclear how prevalent AI generated artworks will become, as it's a freshly emerging form of art.
Some people are inclined to think that AI will eventually get to a stage where it is intelligent enough to learn and create its own art, without being fed data sets and without being prompted.
Some argue that any AI generated art technically isn't art, as it's forced and depends upon the analysis of previous humans' artworks.
There's no real way of telling yet, as it depends how much value is placed on human art vs AI generated art in the future.
Another really interesting use for AI instead of AI art is using algorithms to detect false images and artworks by humans.
The Art Recognition Algorithm, made by Dr Carina Popovici and Christianne Hoppe-Oehl, has been used to successfully detect counterfeit paintings on the marketplace.
It uses neural networks which analyze the style of an individual artist, focussing on things such as color and brush stroke, before generating a heat map on the image being evaluated for authenticity to highlight areas of suspicion.
It's pretty accurate, having detected the La Horde fake being sold unknowingly by Christie's in 2006.
Should I buy AI generated art?
AI generated art is one of the most individualistic forms of art you could possibly buy.
In that sense, each image created by a neural network is totally individual and almost impossible to recreate due to the complexity of neural networks! You can pretty much guarantee that if you buy a piece of AI art, it's totally yours - no one else is going to have that exact image hanging in their house.
Beyond portraiture, this is the most personalized art around.
There's also a pretty high chance that original neural network produced art will appreciate as the taste for machine artists grows with the advancement of technology.
So, if you enjoy super abstract pieces that make a statement in your home, then AI generated art is definitely for you!