Most of us don’t notice algorithms unless they go wrong. When an online retailer sends an email suggesting purchases they think you’d like, every one of them is a Sylvester Stallone movie because you Googled a trip to the Rocky Mountains a few days ago. Or, as happened last year, when Amazon had to remove T-shirts from sale because they were plastered with “Keep Calm and Choke Her”. The slogans were being generated by an algorithm trawling the internet for verb-noun combinations. But no one had thought to add a moral dimension to the equation, which would tell the computerised process that threats of violence might be in poor taste.
Even though the T-shirt designs might not have taken the fashion world by storm, this is still the way that more and more creative work is being done: an algorithm, albeit one designed by a person, now invents things which were once the preserve of human creativity – not just clothing slogans, but art, poetry and jokes.
I’ve tended to be sceptical about the marriage of machines and creativity. Isn’t the whole point of a poem that it speaks from one human being to another? When I read Virgil’s Aeneid, it doesn’t matter that it was written 2,000 years ago by a man I wouldn’t have understood. It couldn’t matter less that the big emotional scene is between the queen of Carthage and a prince of Troy. All that matters is that Dido’s heart is broken by Aeneas leaving, just like our hearts were once broken by people who weren’t good enough for us (though we sensibly didn’t throw ourselves on a pyre when they finally left).
The problem is, if anything, even more difficult when it comes to humour. Making people laugh is a difficult business at the best of times. Context is everything, which is why it’s so hard to make jokes work in translation. Word order, puns, references, resonance – all these things change from one culture to another, and from one language to another. On top of that, jokes aren’t a rarefied art form. They have to appeal to a mass audience. Think how feeble Christmas cracker jokes need to be to ensure they can be understood by everyone from small children to their great-grandparents.
And jokes are themselves only a part of what makes something funny. As Frank Carson succinctly put it: “It’s the way I tell ’em.” Comedians are more than the sum of their material: they’re performers. How can a machine in a lab replicate any of that? Quite aside from the strange dynamics of an audience, whose presence can have an alchemical reaction – their laughter redoubling not because of the comedian they’re watching, but because of the response of the rest of the audience.
I’ve just made a documentary for Radio 4 (who else?) about computational creativity – I’m not an expert on algorithms at all, but I do understand something about the difficulties that any performer or artist needs to overcome if they’re going to connect with an audience. The work that computers are creating (once the algorithm is in place) is incredibly wide ranging: visual art, poetry made out of tweets produced over the previous 24 hours, even comic one-liners.
The poetry is, unnervingly, no worse than Adrian Mole’s teenage efforts. Simon Colton of Goldsmiths College showed me how his algorithms search the day’s newspapers to calculate, from the headlines, whether it’s a happy day or a sad one. The programme’s mood is thus decided, and it can write something akin to poetry by picking key words from the newspaper articles themselves. Mr Colton’s computer can also turn Twitter into verse, finding short chunks of prose which fit the mood he’s chosen, then picking tweets which rhyme and scan and putting them in a meaningful order.
The jokes are the really tricky thing. Computers are getting better at identifying what might seem funny, but comedians don’t need to worry yet. Some scientists think that humour is the final piece of the artificial intelligence jigsaw: when computers can make us laugh, they will be as intelligent as humans. So when HAL takes over the world, at least he’ll have a punchline.