Developing AI Artists with Unique Styles & Genres

Edd H. English

Incredible, human indistinguishable AI-generated artwork has now been made a reality.

It's been just a few years since artificial intelligence (AI) started making paintings, drawings and music. Artificial intelligence is now beating the human world champion in the ancient Chinese game Go and can write a better haiku than most humans. AI-generated artworks are also quite good at predicting whether a person will develop cancer or not, interpreting medical images, and some can even do magic tricks. What we've seen so far is just the tip of the iceberg. Within a few years, AI will surpass human performance in many fields. It's only natural that AI has started to make art too.

A lot of AI research is focused on generative models, which try to generate something similar to existing data without using any instructions or conditions: an image or an audio recording, for instance. Generative models are trained using neural networks, which are loosely inspired by the same processes that dictate how our brain works.

This technique is known as deep learning. It can be used for generating pictures or music by altering the network's parameters in such a way that it produces patterns that resemble those already available in existing databases (e.g. images from Google Images). This approach has been used by Google DeepMind to train neural networks to generate piano pieces that sound like those composed by humans (DeepMind AlphaGo) and also to generate new video frames given what was shown before (Google DeepDream). In both cases, this technique allowed the AI to create something entirely new and different from existing data without any human intervention whatsoever!

We can use similar model for creating artificial artists with unique styles and concepts; with the right parameters, the results can be significantly different from what has been done by humans, and so we can explore novel creative spaces millions of times faster than before using AI. Imagine having an artist who could come up with a new, game-changing art direction every five minutes (rather than once a decade)!

Unfortunately, this approach still requires a significant amount of computational power and time, which is why it's still impractical for real-world applications today. However, that may change soon due to cloud computing becoming more affordable and GPU compute power becoming significantly more available. Soon, the day may come where each of us has the power of a million artists in their palm - and humanity will be better off for it.