杏吧原创

BEAUTY was in the eye of the technician who ran down the corridors of Bell Labs in New York one day in 1965 shouting that a computer had created art. The machine in question, the room-sized , was set up to send numbers to a plotter, a device that turned them into precise line graphs. But thanks to an error in the plotter that day, the numbers produced a random-looking scrawl instead.

A Bell Labs scientist called . He started to call the result 鈥渃omputer art鈥� and set about trying to get IBM 7094 to reproduce the accidental scribble on purpose. After a few coding experiments, he got it to output pleasing zigzag patterns on demand.

Noll said that the results reminded him of Picasso鈥檚 1911-12 painting Ma Jolie, and ended up exhibiting some of the images in a gallery in New York. When Noll sought to copyright the work, however,the US Library of Congress was less impressed by marks made by a machine. When Noll presented the images as the specific results of a computer program he had written, the library relented.

That distinction, between art made by a computer versus art made with a computer, has been at the heart of the debate swirling around creative machines ever since. Created by a machine? It isn鈥檛 art. Created by a human? It isn鈥檛 machine art.

In The Artist in the Machine, Arthur Miller does a good job of showing how that fuzzy distinction has only grown fuzzier. And in the space where human and machine creativity merges, there are big questions to explore. Will machines ever be considered truly creative? What would that mean for our understanding of artificial intelligence? And how would it change how we think of creativity in general?

Miller, a historian and philosopher of science at University College London, adopts a creator-agnostic definition of creativity, one based simply on the ability to produce new knowledge from old ingredients. 鈥淗ow can a system produce results that go far beyond the material it has to work with?鈥� he writes. 鈥淭his is the problem of creativity.鈥�

This is Miller鈥檚 third book mining the intersection of art and science for interesting ideas. He comes at the topic via a background in physics and cognitive science and believes that images are the primary objects of creative thinking. If brains are like computers (a claim we are to take on trust), then why can鈥檛 computers be creative?

After a whistle-stop tour of what various thinkers think about creativity, we get to the strongest section of the book. Through a series of meetings with artists and researchers exploring the frontiers of computer creativity, Miller gives us a relatively comprehensive survey of the artificial painters, poets, composers and storytellers that have been developed over the past few decades. What is clear is just how broad and mature this field is.

There is a nice mix of scientists teaching machines new tricks and artists experimenting with a new medium. We meet Simon Colton, a computer researcher who has been building an AI painter called The Painting Fool for most of his career. It can produce portraits in styles that reflect its 鈥渕ood鈥�, which changes depending on what is in the news that day, and then critique its own efforts. Mark Riedl at Georgia Tech teaches computers how to invent and tell stories, giving them a sense of characters, motivations and plots.

Then there are artists such as Anna Ridler and Mario Klingemann, who have turned machine-learning software into a medium-cum-muse, feeding images into a computer and then reworking the machine-twisted results into strange new forms. For now at least, the most promising art produced by machines comes from this kind of collaboration in which human and machine riff off one another, making something that neither would have been able to create on their own.

Some of the projects lean more to the side of 鈥渁rt made with a computer鈥� than others. But with the advances in AI in the past few years, we are increasingly seeing examples of machines creating music or images or stories without, or with very little, human guidance. For now, these solo efforts won鈥檛 win any awards. But why shouldn鈥檛 they in future? 鈥淚n this day and age, we are going to have to rethink what we mean by thinking and what we mean by creativity,鈥� writes Miller.

It is when Miller does just that, and mulls over the significance of what he has been writing about, that I have less confidence in him as a guide. For a start, he wants to separate 鈥渆veryday creativity, like discovering a new route to work鈥� from 鈥渢he big, domain-breaking feats of creativity, such as discovering the theory of relativity鈥�. He is interested in the big stuff and dwells quite a bit on genius, at times invoking the likes of Beethoven, Picasso and Einstein with near-mystical reverence.

But what about all the muddle in the middle? The bad poetry, a toddler鈥檚 painting? A computer that could independently produce the equivalent of a doodle you might make while on the phone would still be doing something creative. Most art isn鈥檛 high art. What is more, what if discovering the theory of relativity is in fact akin to coming up with lots of new routes to work, rather than some indivisible flash of genius?

Then there is the odd way Miller talks about AI. In one example, he writes: 鈥淎nd all this means that computers are now finally beginning to create art, literature, and music in ways that exhibit not only their creativity but their inner lives.鈥� It isn鈥檛 at all clear what a computer鈥檚 inner life might be.

This strange idea crops up more than once. Here he is discussing the overall aim of the book: 鈥淭his will involve looking into the 鈥榣ives鈥� of computers, exploring their creativity, their innermost thoughts, to what extent they may be similar to ours and to what extent different.鈥�

Miller also jumps on a quip by David Ferrucci, the lead on IBM鈥檚 Watson AI, a machine that took part in, and won, the TV quiz show Jeopardy!. When asked whether Watson could really think, Ferrucci replied: 鈥淐an a submarine swim?鈥� Miller seems to miss the point. 鈥淥f course, a submarine swims 鈥� not like a fish, but better,鈥� he writes. 鈥淚t鈥檚 an apt analogy.鈥�

But I suspect that isn鈥檛 what Ferrucci meant at all. He was probably alluding to an aphorism by the pioneering computer scientist : 鈥淭he question of whether machines can think is about as relevant as the question of whether submarines can swim.鈥� Miller misses this 鈥� or misses it out 鈥� completely.

Still, the questions that Miller pursues in his book are some of the most exciting ones you can ask about artificial intelligence today. Computers that surprise us 鈥� that make things we haven鈥檛 imagined or find solutions to problems we have overlooked 鈥� will change the world. And there is good reason to think that computers will do these things in ways that will feel alien to us.

鈥淭he most promising art, for now, comes from collaborations between human and machine鈥�

Miller asks whether computers will develop the qualities we see in creative people, or develop their own form of autonomous creativity, 鈥渘ot as replica people but as an altogether different and independent form of intelligence鈥�. It is one of the big questions and Miller gives us a taste of what is possible. It goes down even better taken with a pinch of salt.

Douglas Heaven is a consultant for New 杏吧原创