Computers are just machines toggling switches on and off in an order humans told them to by writing some lines of program code. So how can they create art?
While this statement was true until a few years ago, all big tech companies are now investing heavily into machine learning by using artificial neural networks which are supposed to copy the functionality of our brain. The groundbreaking revolution of this technology is that machines don’t have to be programmed with millions of lines of code anymore. Instead, an artificial neural network is set up and trains itself with millions of data sets given to it. While trained neural networks showcase amazing abilities, it’s not yet possible to explain how they actually process the data. Now how does this apply to arts though?
A team at Google Brain developed an AI called Deep Dream. During development, Deep Dream has been teached to identify objects by processing millions of classified images. First it learned how to distinguish between all the colors and shades, then it scanned for border areas between objects. Over time it learned how to separate one object from another and built up a catalog of every object from every picture it had scanned. It then figured out how to arrange and categorize objects that had similar characteristics and learned how to recreate random composites of those objects. Finally, when prompted, it displayed a random set of those pictures over a landscape template. Unlike humans, machines only take minutes or even seconds to create an artistic piece after they are trained. These are some of the resulting pictures:
It’s still possible to identify the right picture to be based on Van Gogh’s ‚Starry night‘ but reinterpretations of famous art are a form of art also done by human artists. Of course, one could argue now that Deep Dream is still designed by humans and the training data is made by humans aswell. As always, it’s hard to draw a line between ‚the AI created a piece of art‘ and ‚the AI was just a tool of the developers to create art and thus they are the artists‘. Following the exaggerated logic behind the latter argument, one could say that art created by a genetically modified human who has been cultivated in an laboratory also is not their art. And considering the advances in genetic research, this hypothetic situation doesn’t seem too unrealistic in the future.
Another point to consider is the fact, that humans don’t create art out of the blue either. Just like Deep Dream gets input via training data, humans receive input through their senses all over their life and rearrange this ‚data‘ during the creative process. While machines can’t express social or political critics through their images yet, it will be interesting to see how good artificial intelligences will become in the future and which effect these advances will have on human-made art.