When artists, creatives and computer scientists leverage Artificial Intelligence (AI) programs to develop art in its many forms, does the code itself become the artist? Or is it a tool for the artist? Could AI be the piece of art itself? In his book, The Creativity Code, Marcus du Sautoy grapples with those issues as computer programs are now used in domains such as painting, music, mathematics and story writing.
When AlphaGo, an AI program developed by DeepMind (since acquired by Google), played its highly irregular 37th move in a game of Go against World Champion Lee Sedol in 2016, it stunned the entire audience, the team of developers behind it, and its opponent, who had mastered the game beyond any other. It was only once the game unfolded that the power and vision behind the move was revealed. The shock was not only in the surprise behind the move itself, but it also seemed as if the program went beyond what its creators had originally coded; it was not limited by the insights of its creators. To be sure, while AlphaGo’s win over Sedol in 5 games (4 games to 1) was heralded as a triumph for AI, the program had not acted out of its own free will, and had not developed a sense of agency. It was acting on behalf of its creators, and, thanks to its immensely more advanced processing powers, was able to propose a move that nobody had been able to see. Free will was not necessary for AI to beat humans at Go (or Chess and Jeopardy! for that matter). It may not even be necessary for AI to take over from humans as we stumble on to new advances. It may however be necessary in the artistic domains.
Without agency, no computer will create pieces of art independently. The urge to create will have been predetermined by its coders, who set, to varying degrees, the original parameters or axioms from which the program will proceed to execute instructions, even if those instructions have become so complex, and the program has been provided with the freedom to interpret those parameters as it sees fit (see artificial neural networks). The ability to extract, interpret and act on massive amounts of data has not only allowed AI to make its own decisions, but has also abstracted that decision making process from its original creators. Just as Netflix can make recommendations to users which the coders themselves will not be able to fully explain, so can, for example, a team at Rutgers University code an AI program that generates pieces of art which the the former could not have predicted.
Yet while AI can go beyond the insights of its coders, it does not (yet) have the ability to discriminate between its choices and decisions to introduce something of value. It can introduce surprising elements which its creators can use to progress their art, but without human judgement (either explicit or coded in as a feedback loop), it cannot assign a value to its output. AI could recreate Borges’s Library of Babel, but it cannot yet independently point you towards the same author’s recreated work, or effectively towards the subsets of its creation which are valuable.
What makes a piece of art is a question that will be long debated and will evolve as humans do. The reason for creating art, however, has been innate to humans since our first steps, ranging from cave paintings to story and music writing for rituals: free will, self-expression and communication. Humans have done so to form groups, allegiances and common stories, revere their gods, express their injustice, and more generally provide a window into our own (and others’) inner workings. A piece of art is not only the artist’s production, but also the reaction it elicits from those exposed to it. The individual experiences of the artist and the audience drives much of how and why the art is created and received. Without the innate drive to create, computers will remain an extension of the artist: a tool to test and to work the menial details that artists cannot do, while the artist focuses on the overall outcome. For example, AI could be used to generate essays based on a few key words introduced by the author, but it is the author that will shape the overall narrative. Throughout the book, Du Sautoy mentions multiple examples where AI can get creative “locally” without getting the full picture.
While AI does not yet have the need to express itself, the artistic endeavors that it is used in can still however reveal its inner workings, similar to how human-made art reveal our inner emotions. A team at Google created DeepDream in order to better understand how AI analyzes pictures, and what patterns it searches for. Humans are, similar to AI, programmed to spot patterns. Through DeepDream, we’re able to better understand how AI perceives differently from us. Similar to how art is used to transmit our view of the world, so can it be used to view how AI programs think and perceive the world if and when they do produce their own pieces of art.
Du Sautoy mulls over those ideas with clarity and ease, but cannot reach a definitive conclusion. He had been awarded the Faraday Prize for excellence in communicating science to the public in 2009 by the Royal Society, and his background in mathematics lends itself well to explain not only the advances and workings of AI, but also the intricacies and patterns within art itself — particularly within classical music. As he ponders what is creativity, art, mathematics and algorithms, Du Sautoy simultaneously makes a case for why, instead of replacing artists, computers are instead used (so far) to push us to be more creative, while at the same investigating what it is to be human when engaging with our favorite pieces of art.