Infants Outsmart Computers

He should watch his posture, but not worry about being outsmarted just yet

http://www.wsj.com/articles/what-a-child-can-teach-a-smart-computer-1421942333
What a Child Can Teach a Smart Computer
Children are remarkably good at coming up with creative concepts in a way that computers can’t even begin to match
By ALISON GOPNIK

[spoiler title=”Click for whole story” open=”0″ style=”1″]

WHAT A CHILD CAN TEACH A SMART COMPUTER

Every January the intellectual impresario and literary agent John Brockman (who represents me, I should disclose) asks a large group of thinkers a single question on his website, edge.org. This year it is: “What do you think about machines that think?” There are lots of interesting answers, ranging from the skeptical to the apocalyptic.

I’m not sure that asking whether machines can think is the right question, though. As someone once said, it’s like asking whether submarines can swim. But we can ask whether machines can learn, and especially, whether they can learn as well as 3-year-olds.

Everyone knows that Alan Turing helped to invent the very idea of computation. Almost no one remembers that he also thought that the key to intelligence would be to design a machine that was like a child, not an adult. He pointed out, presciently, that the real secret to human intelligence is our ability to learn.

The history of artificial intelligence is fascinating because it has been so hard to predict what would be easy or hard for a computer. At first, we thought that things like playing chess or proving theorems—the bullfights of nerd machismo—would be hardest. But they turn out to be much easier than recognizing a picture of a cat or picking up a cup. And it’s actually easier to simulate a grandmaster’s gambit than to mimic the ordinary learning of every baby.

Recently, machine learning has helped computers to do things that were impossible before, like labeling Internet images accurately. Techniques like “deep learning” work by detecting complicated and subtle statistical patterns in a set of data.

But this success isn’t due to the fact that computers have suddenly developed new powers. The big advance is that, thanks to the Internet, they can apply these statistical techniques to enormous amounts of data—data that were predigested by human brains.

Computers can recognize Internet images only because millions of real people have sorted out the unbelievably complex information received by their retinas and labeled the images they post online—like, say, Instagrams of their cute kitty. The dystopian nightmare of “The Matrix” is now a simple fact: We’re all serving Google ’s computers, under the anesthetizing illusion that we’re just having fun with LOLcats.

The trouble with this sort of purely statistical machine learning is that you can only generalize from it in a limited way, whether you’re a baby or a computer or a scientist. A more powerful way to learn is to formulate hypotheses about what the world is like and to test them against the data. One of the other big advances in machine learning has been to automate this kind of hypothesis-testing. Machines have become able to formulate hypotheses and test them against data extremely well, with consequences for everything from medical diagnoses to meteorology.

The really hard problem is deciding which hypotheses, out of all the infinite possibilities, are worth testing. Preschoolers are remarkably good at creating brand new, out-of-the-box creative concepts and hypotheses in a way that computers can’t even begin to match.

Preschoolers are also remarkably good at creating chaos and mess, as all parents know, and that may actually play a role in their creativity. Turing presciently argued that it might be good if his child computer acted randomly, at least some of the time. The thought processes of three-year-olds often seem random, even crazy. But children have an uncanny ability to zero in on the right sort of weird hypothesis—in fact, they can be substantially better at this than grown-ups. We have almost no idea how this sort of constrained creativity is possible.

There are, indeed, amazing thinking machines out there, and they will unquestionably far surpass our puny minds and eventually take over the world. We call them our children.
[/spoiler]

This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *