SMART NEWS

Could a Computer Ever Learn to Identify Death?

Identify signs of life in a human might still be beyond a computer’s grasp

Contributor

January 22, 2014

robot — Hello? Are you asleep or dead? John Greenaway

Seven people die in Clockwork Orange. Ninety-five die in Kill Bill Volume 1. Eight hundred and ninety six die in the Lord of the Rings: Return of the King. We know this all because someone counted each death. By hand. And then told the website Movie Body Counts.

But Caleb Garling at SFGate wonders if there’s another way to identify death. A way that doesn’t involve humans, but computers. And there might be—it will just be hard. Garling explains:

A lot of training would need to happen. One of the common ways to teach a computer is to give it repeated examples — with information about the examples — and let the machine’s software puzzle out the patterns and determine the characteristics of one quality versus another. This is how you might teach it to recognize a cab versus a yellow car. Perhaps showing a million pictures of alive people versus a million pictures of dead people would teach the machine.

Teaching a computer to figure out whether someone is alive or dead is useful beyond movie body counting. But we're not yet capable of programming the artificial intelligence required to identify life. Plus, humans aren’t even that good at recognizing death. If a man's lying down, motionless, you might not be able to tell if he's sleeping or dead unless you get close enough to see him breathe. Teaching a computer to know something we don’t know is harder than teaching it to answer Jeopardy questions or paint pictures.

“That’s what the whole hullabaloo about ‘deep learning’ is — somehow we’ll train machines to be like our brains through pattern recognition,” researcher Mary Cummings told Garling. “But we don’t understand it ourselves. Of all the pieces of the body, the eye-brain connection is a mystery science hasn’t gotten close to unlocking.”

But in an age when facial recognition still isn’t great at telling a crying person from a laughing one, telling when the life has left a human might still be beyond a computer’s grasp. Then again, it’s getting harder and harder for humans to identify whether something is alive—or just a really lifelike robot.