What if we told you that future artworks may be created no longer by human artists, but by Artificial Intelligence (AI) artists instead? This absurd claim may not be as far-fetched as previously thought with the significant advancements in technology, specifically the foray into AI art and AI paintings, that we observe today.
What is AI Art?
AI Art can be defined as any artwork created with the assistance of artificial intelligence. It can generally be split into 2 categories, completely autonomous AI systems vs a collaboration between a human and an AI system.
There are many strategies and methods used to create AI Art. There are 2 main methods currently used to generate AI art – Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs).
An example of an AI algorithm is DeepDream, which was created by Google. There are many other algorithms and software available which use their distinct method to create AI art.
How to make AI Art?
These two networks are considered deep learning and machine learning models, which each have their own distinct processes and methods as to how they produce the final masterpiece.
1) Generative Adversarial Networks (GANs)
One method that is used to create machine learning artworks utilises Generative Adversarial Networks in Deep Learning.
The first step requires numerous artworks and designs to be curated, in the art style that the model is to be trained on. So if you wanted a model that drew Renaissance-inspired artworks, then paintings and art from that time period should be used. These images are then fed into the network.
By analysing thousands of images, this neural network uses a generative and a discriminative model that competes with each other to train the generator. You can think of it as a person who is repetitively learning to draw and create fake currency, while the government is trying to detect and prevent the currency from circulating. This process repeats iteratively, whereby the model repeatedly generates new artworks, until it is indistinguishable from the previous images that was fed into it. What this effectively means is that the model keeps generating artworks until it believes that its own artwork is indistinguishable from those previously fed in.
Upon the completion of training, these neural networks will now have an idea of what art is, based on their prior inputs and training. They are then capable of creating unique AI artworks that have never been seen before.
Advantages of GAN Artworks
The usage of GANs to generate machine learning art and AI paintings is truly AI art, in the sense that these artworks are theoretically drawn entirely by the AI neural network. This allows novel, eccentric and out-of-this-world artworks to be created from scratch – providing many unique designs never seen before.
Disadvantages of GAN Artworks
Currently limitations in technology has led to a less than desirable outcome for these AI artworks. Many artworks produced from GANs look surreal and do not resemble an actual painting. Due to inherent limitations in deep learning and machine learning, the models fail frequently when trying to imitate artworks. The artworks produced are generally deformed and warped.
2) Convolutional Neural Networks (CNNs)
The second method used to create machine learning art utilises Convolutional Neural Networks.
These CNNs consist of numerous layers which serve to process visual information in a hierarchical forwards manner - each with their own image filters which are able to extract unique patterns and features (also known as the 'art style' in artworks). They are trained on numerous images in order to capture different styles and texture information. Utilising the information captured by the different layers, these networks can manipulate input images to create new, perceptually meaningful images.
CNNs generally consist of an image layer and a style layer. The CNN will integrate the style layer onto the image layer to create a unique artwork. For example, if we use an image of a dog, and the style of Van Gogh’s Starry Night, the final output would be a dog drawn in the artistic style of Van Gogh. This method is known as neural style transfer.
Advantages of CNN Artworks
As the image output is predefined by the input images, the resultant AI art is usually very unique and engaging. People are able to choose their favourite style and images to blend, thereby creating very realistic artworks that may look even better than the original.
Disadvantages of CNN Artworks
This method of AI art has been criticized by many as not being ‘true’ AI art, and more of a fancy filter by many AI enthusiasts and researchers. Since this deep learning technique only combines the relative styles of the images, and does not create an artwork from scratch, whether this constitutes AI art is still up for debate.
Can AI make Real Art?
Art can be expressed in many forms – drawings, paintings, sculptures, music and videos. It is a way for artists to express themselves through beautiful renditions of their thoughts and imagination. There have been many debates about what true art or quality art is.
Even amongst human artists, people have been debating whether there is such a thing as ‘real’ art. Some critics have argued that while all art is real art, there is good and poor art. For example, morally challenging, ethically questionable, or low effort works have been considered as poor forms of art. On the other hand, some view this idea of ‘the correct art form’ as an elitist and ridiculous concept. These people feel that anything someone has created can be considered as art. If people cannot even agree on this, it is definitely not surprising that there are even stronger views on whether AI generated art is truly art.
Proponents of the liberal arts agree that AI art can be considered as a form of art, since these AIs are creating artworks similar to the process of how a human perceives and creates art. However, there are a significant number of critics that debunk this idea, clearly from the viewpoint that AI cannot express emotion, experience personal moments and develop creativity. These AI artists are merely systems in which art is being passed through, and such styles are learnt and replicated without any emotion. There is no creativity in creating art through AI – it is merely a duplication of pre-existing work.
What do you think?
Can AI create New Art Styles?
Currently, AI artists create artworks in the style that they have been taught to learn. They currently cannot create art forms not yet seen before. However, with the large advancements in machine learning and deep learning, it is not unimaginable to have an algorithm that can freely create art without user input. If we reach the point where AIs can emulate a human artist, would their art be considered true art then?
AI Art is still in its infancy, and there are many improvements to be made. Artificially created artworks is still a controversial topic amongst the art community, with both sides unable to compromise. Nowadays, art is more proliferated and established than ever. With the world of art constantly changing, who knows the future will hold for art and its long-lasting legacy.