Marcus Du Sautoy, professor of mathematics at Oxford University, recently published his book The Creativity Code: Art and Innovation in the Age of AI. After watching a Go master bested by Deepmind’s AI robot, Du Sautoy began questioning everything he previously understood about what sets humans apart from robots. He had frequently used Go as a metaphor for a deeply complex task because the game requires creativity, understanding, and intuition – which computers are not supposed to have. In an age where AI is taking over human functions across a range of labour markets, conventional wisdom has told us that humans are still necessary for creativity and art. Du Sautoy’s new book questions whether this is truly the case.
Du Sautoy begins his quest by developing a test for computer creativity. He names the test after Ada Lovelace, the first computer programmer. To pass the test a computer must be capable of producing something new, surprising, and valuable. Du Sautoy then investigates several examples where AI has been used to create art. Microsoft’s machine learning algorithms created a painting that follows the patterns it recognises from training on all of Rembrandt’s work. In another instance, an algorithm was designed to continue off a jazz riff. The composer was impressed, saying that the computer’s creation was in line with the composer’s vision and typical style.
These examples led Du Sautoy to consider what art is. He proposes that art takes us on an emotional journey. However, he wonders whether the emotional highs and lows are driven by our brain’s response to a novel or surprising pattern. Could computers predict this pattern? And if they can predict it, can they not create it? To answer these questions, he first defines what creativity is, and its three main types. In the case of exploratory creativity, an artist recognises and understands a pattern, and attempts to push the limits. In combinatorial creativity, an artist juxtaposes two unrelated ideas to see if associations with one enlightens new understanding of the other. Finally, some artists drive phase changes. This being the case, no pattern can predict the step change because it is breaking free of the system entirely. Du Sautoy suggests that both the painting and jazz examples demonstrate that AI can accomplish the task of recognising a pattern and building from it. AI can even help us understand patterns we didn’t know were there. However, it is much more difficult for AI to be phase change creative because it requires going outside the existing pattern and rules.
Finally, Du Sautoy discusses AI’s applications to his own field of mathematics. He suggests that it is fairly easy to build code such that an AI algorithm will develop something “new” or “surprising.” However, it is harder to code a robot to understand what is important and why. AI can generate new and surprising proofs for mathematical theorems but cannot select which ones are most significant or may have wider scientific applications. Given this example, Du Sautoy concludes that consciousness must be necessary for total creativity. His outlook remains curious, as he marvels at the feats AI already has accomplished and the potential for the future.