There is a natural desire among computer programmers working with artificial intelligence (AI) to achieve the most complicated, difficult and useless things. To get a computer to make art, for example, would be an almost unbelievable achievement, because we imagine that art is uniquely human, illogical, emotional, personal, subjective, indirect – everything an algorithm is not. This is precisely what makes it such a programming challenge, and whether we need art made by computers or not, the challenge must be surmounted, like Everest, because it is there. The point is the genius of the programming, not the historical impact of the art. And it is genuinely exciting – a means of uniting scientists and artists in order to understand how art works.
I have documented a few examples of linguistic/artistic AI projects in recent columns: the attempts at bilateral conversation by two Cleverbots; the computer charged by Dutch bank ING to paint a “new Rembrandt;” an Internet company that can write your tweets for you after your death; a program that attempts to describe photographs in words it writes itself. All these result in quirky and imperfect artifacts that are really only interesting if you know how they are created.
Now, what is perhaps the most powerful AI setup ever created is attempting to create or recreate some art, and its massive computing power and genius programmers promise to deliver something an on entirely more convincing scale. The computer is Watson, IBM’s so-called “cognitive” machine – the one that was designed to answer Jeopardy! questions posed in “natural language,” and subsequently did so well on the show. It is a combination of hardware and software. As reported in Forbes, IBM has partnered with a New York agency called SoftLab, which is programming Watson to create a sculpture “by” (i.e., in the style of) the great Catalan architect Antonio Gaudi. They have fed the machine not just hundreds of images of Gaudi’s work, but also contextual images – images of Barcelona, where Gaudi’s most famous work can be found, and other historical and cultural information. The idea is to actually recreate Gaudi’s intelligence, not just his signature style – to create an artificial Gaudi who was inspired the way Gaudi was.
In theory, any artist’s brain could be recompiled in this way, and you could consult a virtual Leonardo da Vinci not just about how to draw and paint but about how to invent new flying machines.
In literature, writers are working with AI experts to try to do a similar thing with genres: They are getting computers to “read” (digitize and analyze) corpuses from a particular genre – say romance or westerns – to codify the plot conventions. As narratologists have done long before the days of computers, they will then be able to come up with a set of formulas for creating their own original westerns. So far, computers are very bad at actual writing, so writers will have to do the individual pages, but they will at least have a generated plot to work with.
We have long made jokes about computer-generated novels – we all think it would be easy to write a program that would generate a historical-injustice-themed Canada Reads winner – but outside humour, not even cash-strapped publishers have seen a real use for them. The idea is cool, is all. The challenge, like the creation of a Gaudi sculpture, is a technical one, not an existential one.
Watson is scheduled to produce its 3-D plans for the Gaudi sculpture this week at a tech conference in Barcelona. The team has released some preliminary drawings of swirly shapes and they do look pretty Gaudi-like.
But here’s the thing: Just like answering Jeopardy! questions, this kind of mimicry and empathy is something that humans already do extremely well. Hundreds of years of convincing forgeries and frauds prove this. I can easily read all a writer’s work and research her past and then, depending on my talent, do a passable pastiche of that writer’s work, just as a good architecture student can already draw an imaginary Gaudi church.
And this exercise actually has a practical function: We humans use imitation as a way of understanding things. We place ourselves in an author’s brain in order to dissect her style and to expand our own. I use this practice as a regular exercise with creative-writing students – just as art students have to go into galleries and execute perfect copies of great masters’ work. It’s training for your own painting.
Now, it’s probably true that a computer can do all this research much faster. But there is one parameter which will always render the starving artist more attractive as an art-producer: cost. And this is paradoxical: Here is a case in which the usual economics of automation are inverted. Artists cost almost nothing to employ. All we need is an apartment in the bad end of town. Whereas Watson consists of 90 eight-core IBM servers and an army of PhDs from Stanford. I promise we consume much less energy (our cooling costs are much lower, for one), even if we smoke and drink. We are simply more economical as labour units. How about that? We have finally made ourselves cheaper than machines.Report Typo/Error
Follow us on Twitter: