The clever code of computing â" the algorithmic empire â" is on the march across business and society, but human help still has a role to play. And that role is increasingly apparent as algorithms try to do more work that was once the realm of human intelligence.
That is the subject of an article ) I wrote in Mondayâs New York Times. The reporting ranged fairly broadly, even if the piece was more focused, in part because the underlying issue, I think, is a good one: the line between human and machine intelligence. Of course, that line keeps moving in some pretty interesting ways, fitfully and not always predictably.
The reporting for the article gave me an opportunity to revisit a research project I first wrote about nearly three years ago for our Science sction. The project is a bundle of advanced computers and machine-learning algorithms called the Never-Ending Language Learner, or NELL for short.
NELL began three years ago as a bold bet on the power and potential of the algorithm, unleashed on its own â" a computer system that reads the Web and seeks to understand the meaning of words and phrases in context.
Human help was not part of the original plan, but the plan changed after about six months. âI thought it would be this computer Lone Ranger, a totally automated system,â said Tom M. Mitchell, a professor at Carnegie Mellon University, who heads the NELL project. âBut we got it wrong at the beginning.â
Today NELL, based at Carnegie Mellon, with funding from Defense Advanced Research Projects Agency and Google, is doing quite well. It has assembled more than 50 million âbeliefsâ about the meaning of words in context. Not all of them are accurate. As of last week, for example, NELL thought âTexas marriage recordsâ! was a ârecord label.â
So NELL relies on human fact-checking to correct such misconceptions.
People can help NELL by clicking a thumbs-up or thumbs-down icon to statements about the meaning of words that the computer believes to be correct. This helper feature â" and an explanation of how NELL works and assembles its beliefs â" is on the projectâs Web site. That helper feature is also on Twitter @CMUNELL.
NELL has a couple of thousand followers, and is looking for more.
Dr. Mitchell has his own beliefs about what lies ahead for computers like NELL. Today, they can be brilliant at times, and laughably wrong at others, he said. But eventually, he suggests, human help will not be needed.
âI honestly think that 10 years out, maybe 15 years out, computers will be reading more like humans, and be able to do what we call reasoning,â Dr. Mitchell said.
âThey wil be able to do more than just extracting meaning from word patterns,â he said. Computers, he predicts, will be able to read and reason enough to backstop, even challenge, human decision-making.
âA computer,â Dr. Mitchell explained, âwill say to you: âYes, I hear what youâre saying and it seems reasonable. But I donât believe it, and hereâs why.â â