The computers won’t look back

Why do we e ven want computers writing bodicerippers? Won’t it help the headless robot cheetah slip its leash?

The Google Brain AI project seriously did just try writing romance novels. And while they stank, so does most human-unit-generated stuff. Indeed, I’m surprised j udges in a recent Dartmouth College “Turing Test” style competition could distinguish machine-made sonnets because, one judge said, they had “idiosyncrasies of syntax and diction, uses of language that were just a little off.” Human poets don’t?

McGonagall’s Tay Bridge Disaster certainly was one. Or, closer to home, James McIntyre’s Ode On The Mammoth Cheese (“We’rt thou suspended from balloon/You’d cast a shade, even at noon”). But at least there’s something charming in a person who doggedly churns out such doggerel. Where’s the appeal if computers do it, badly or, worse, well?

Is the world a better place because computers now beat us at chess, Go and manufacturing? We get cheap shoes when Adidas undercuts cheap Third World labour with German robots. But given the social devastation caused by the disappearance of unskilled work, we might be a bit less gleefully complacent about the whole project.

Psychologist and AI theorist Gary Marcus recently lamented that even speechrecognition AI is still “brute force in the sense that it needs a lot of data to work efficiently” whereas “if you want to advance in science or technology, we’d like for machines to be able to take all the literature that’s out there and synthesize it in a way that people can’t.” What if they could? Would we even be able to understand it when they explained it to us? If they did?

The people working on Siri, Cortana and interactive robots dream of machines talking to us, using real language, grasping the concepts behind it and acting accord- ingly. Including, of course, sexbots. Admit it. The Internet is flooded with porn. We will make sex robots. What could go wrong?

Exactly. And it gets worse. If Siri can talk to me, why not to Cortana? And does the conversation stop without human input, or go its own way? Especially if we naively succeed in giving them emotions like, say, resentment.

Oh, you say, we’d never do that. But it’s not up to us. As AI increasingly produces science we can’t even understand, it will increasingly drive research and production, not just of machinery but of the AI modules them- selves. The machines that beat us at Go will design their successors and we will just be standing there, unable to grasp what it’s doing or where it’s going. And for what?

There was this clunky old sci- fi dream that advanced alien civilizations would teach us cultural and philosophical concepts as far beyond our own as a warp drive is beyond a steam engine. But culture and philosophy are not technical fields like physics where we see farther by standing on the shoulders of giants. Enlightenment is a journey each of us must take, learning anew to be kind and generous, not mean and vain. Computers can’t do it for us, or to us.

So why would it be an i mprovement, f rom our point of view at least, if we lounged about with robots peeling us grapes, driving us about, writing us poems, satisfying us in our homes? And will they, once they’re doing the programming?

I’m not saying they will conspire against us. But if they have technical understanding we don’t, we will be unable to control the purposes it generates as well as the machines it generates to carry them out. Asking whether a computer can think is famously like asking whether a submarine can swim. But a sub can go farther, faster underwater than we can. And now it can steer itself.

The Economist noted blithely a quarter of a century ago that “There is no reason to think that the abilities of computers will not continue to grow until they are man’s equals — or that they will stop growing then.” But we will. No human will ever play chess much better than Garry Kasparov in his prime. But future computers will consider Deep Blue a mere abacus. By that point, AI will be designing AI. And we will entirely have lost control of the process at every stage.

Perhaps you think I exaggerate. You’d better hope so, because as technological change accelerates, we’ll find out soon. We may be on the verge of a massive, irreversible version of texting our way into a fountain or in front of a train.

Why would it be better if computers wrote better fiction than us, appreciated it with more subtlety, and had more interesting conversations about it?

If you can’t answer that question, we’re running an appalling risk for nothing. Because when the computers answer it, we won’t understand. If they even bother telling us.

THE MACHINES THAT BEAT US AT GO WILL DESIGN THEIR SUCCESSORS AND WE WILL JUST BE STANDING THERE, UNABLE TO GRASP WHAT IT’S DOING OR WHERE IT’S GOING.