MELISSA BLOCK, HOST:
A machine with superhuman intelligence is a staple of science-fiction, but what about a machine with just ordinary human intelligence, a machine that's so humanlike in its behavior that you can't tell if it's a computer acting like a human or real human? As part of his project Joe's Big Idea, NPR's Joe Palca has this story about a competition to build that kind of machine.
JOE PALCA, BYLINE: This competition isn't one where you test a computer to see if it can beat a human at something like chess or poker or "Jeopardy!" No, this competition is for a program that can pass what's called the Turing Test, named for the British computer scientist Alan Turing, who first proposed it.
DAN ROCKMORE: The Turing test is the one in which you actually try to be human, try to appear to be human.
PALCA: Dan Rockmore is a computer scientist at Dartmouth College. He says to pass the Turing Test, you have to make a program that does something a human might do in a way that's indistinguishable from what a human would produce. And he says creating such a program doesn't just involve writing clever computer code, it involves picking apart human behavior so understand its essence.
ROCKMORE: It might actually tell you a lot more about what it's like to be human than it is about what it's like to be a machine trying to be a human.
PALCA: Rockmore has picked three quintessentially human activities and created three different categories for the competition.
ROCKMORE: One called Digilit.
PALCA: In Digilit, the goal is to write a computer program that can write a short story. The second...
ROCKMORE: A competition called Poetix.
PALCA: In Poetix, the challenge is to generate a sonnet. The third competition involves music and is called AlgoRhythms. Get it? That's a computer science pun. Anyway, AlgoRhythms...
ROCKMORE: In which a machine acts like a dance floor DJ and mixes music in such a way to get an audience cranked-up and happy on the dance floor in a way that a good DJ might.
PALCA: Again, these are not supposed to be programs that just string together top of the charts hits or can understand syntax and grammar and regurgitate something that follows the rules. They're intended to really capture what human beings bring to prose and poetry and music - why these activities form what's called the humanities. Rockmore says a winning computer program doesn't necessarily have to create a sonnet worthy of Shakespeare or a story that could've been written by Alice Walker.
ROCKMORE: I'm hoping a machine can generate an average short story. I'm not looking for experimental short fiction, and similarly, for a sonnet, I wouldn't be looking for a random collection of things that had the right meter and the right rhyme scheme, which from some postmodern point of view might appear to be a great sonnet.
PALCA: And although these Turing Tests really are about computer science, Rockmore admits there's a hidden agenda in this competition.
ROCKMORE: Maybe people who've never, ever thought about a sonnet, never, ever thought about actually writing a short story, might begin to think hard about those kinds of crafts.
PALCA: It was very devious - you're trying to sneak in art under the guise of science.
ROCKMORE: You know, I'm - I always feel like I was a humanities guy who made a wrong turn, but here I am, and clawing my way back little bit by little bit.
PALCA: People planning to enter the competition to write software that can pass the Turing Test have until March of next year to get their submissions in. There's a link on the NPR website that will provide all the relevant details. And next spring, when the submissions are in, we'll post on the Internet the top stories and poems and dance tapes created by the computers so you can judge for yourself whether you can tell them apart from something that's genuinely human made. We'll see just how good a job these programmers do in capturing the essence of humanness in a machine. Joe Palca, NPR News.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.