Lately, the human race has been confronted by ever-smarter machines. Last year, a computer called AlphaGo Lee beat the world’s Go champions; then it was beaten by its own successor, which learnt the game in hours. But at least we knew we had art and literature to ourselves.
Not so fast. Now artificial intelligence (AI) appears to have written a new chapter of our species’ crowning literary accomplishment. Harry Potter and the Portrait of what looked like a Large Pile of Ash has spread across the internet, accompanied by a great outpouring of emotion. In deference to this, I present only a short sample: “The castle grounds snarled with a wave of magically magnified wind ... Magic: it was something that Harry Potter thought was very good.”
We can all agree this is a vast improvement on the original. But before handing over the Nobel Prize, it’s worth taking a closer look. Despite the “robot writes new Harry Potter” headlines, the creators of this text are clear that their algorithm does not “write” at all. Instead it reads a collection of texts (such as advice columns, recipe books or Bob Dylan lyrics) and finds patterns in which words most often follow one another. Then it offers these to a human writer who selects them, just as you select from predictive options when texting on your phone. The tool supplies the vocabulary but a human writer must decide the pacing, balancing earnest imitation with bizarre incongruities.
Algorithms are a familiar tool to artists. The “musical dice game” attributed to Mozart arranged fragments of pieces in random order. The world’s first computer programmer, Ada Lovelace, had speculated in the 1800s that an “engine might compose elaborate and scientific pieces of music of any degree of complexity”. As computers become more powerful, they are also getting better at putting words together. But they are still charmingly blind to context: To an algorithm, Chomsky’s famous example of a meaningless sentence — “colourless green ideas sleep furiously” — is a perfectly legitimate utterance.
Generating art by playing with rules
Today, there is a thriving community of “procedural artists”. Darius Kazemi makes Twitter bots that mash together headlines or produce rap lyrics. Michael Cook built an AI that not only develops its own video games, but also opens a chat window and talks you through its creative process. Emily Short generated a guidebook to an imagined almost-England that is as familiar as it is impossible. These are collaborations between human and machine. They generate art by playing with rules, but so did J.K. Rowling when she combined the familiar tropes of children’s writing, fantasy and boarding school stories.
Machines’ blindness to context has dangers, however. When Microsoft created a bot called Tay with the persona of a wholesome teenage girl, online agitators rapidly worked out how she functioned then fed her information that turned her into an abusive holocaust denier. Digital artists have been developing rules and structures that support creativity while limiting this harm. Far from replacing humans, these robots are merely our tools. If we want more wonderful, bizarre creations we will need more humans in the loop, not fewer, all thinking more carefully about what their digital children get up to.
— The Telegraph Group Limited, London 2017
Lydia Nicholas is a researcher in digital ethics and culture.
