NELL creator Tom Mitchell gets honorary Dalhousie degree
'The end goal is literally to get computers to read and write like you and I do'
NELL is a five-year-old with a Twitter account who reads all day, every day, and follows professional sports.
NELL is also artificially intelligent.
The Never Ending Language Learner (NELL) is a machine learning system that's housed in a computer at Carnegie Mellon University in Pittsburgh.
"The end goal is literally to get computers to read and write like you and I do," said Tom Mitchell.
Mitchell is the founder and leader of the Machine Learning Department at Carnegie Mellon University. He leads the team that created and runs NELL.
True or False? "wonka-candy-bars" is a kind of #Candy?- NELL
He was in Halifax this week to receive an honorary degree at Dalhousie University.
Mitchell said the ambition is to have virtual reading assistants that really understand material.
"If computers could read, you would tell your personal reading assistant that you want to understand this, it would read the 200,000 documents that are out there on this topic and write for you a summary of the key issues and the arguments on both sides, along with citations back to the source material," Mitchell said.
Mitchell says it would replace search engines with questions and answers — a digital game changer.
"Politicians will have a personal reader that tells them what their voters really think about an issue. Marketers will have a personal reader that tells them what people are really saying about the new Apple Watch that just came out because it will read all those comments and digest them."
The list goes on.
The citation for Mitchell's honorary degree describes NELL as the first system to simulate human learning. The team at Carnegie programs NELL with noun phrases that refer to people or emotions, for example. It gives NELL a dozen examples, and NELL's job is to read the web and form beliefs.
"These are facts like coffee is a beverage, or T-shirts are worn with blue jeans or Bill Gates founded Microsoft."
Ringo Starr — wasn't he the drummer?
NELL now has around 100 million beliefs that are stored online. Each day, it should be able to read better than the day before.
It also learns relations between categories, like people and musical instruments. "It found Ringo Starr and drums occur way too often to be a coincidence," Mitchell said.
NELL tends to favor the things that get talked about the most on the web.
"And that turns out, I can tell you from NELL's perspective, it's music to some degree, but much more it's hockey and baseball and all sports, especially professional sports."
The accuracy of the beliefs has gotten better of the past five years.
But mistakes do happen.
"When it was looking at the category of baked goods, it found chocolate chip cookies and raisin oatmeal were examples of baked goods, so then it found internet cookies and thought they were baked goods," Mitchell said.
That caused a bit of trouble. 'It kind of went off on a bad tangent, hallucinating many, many internet cyber terms as baked goods," Mitchell said.
NELL's Twitter feed tries to limit its mistakes by asking people to confirm or reject its hypotheses. For example:
True or False? "St . Michael the Archangel" is a <a href="https://twitter.com/hashtag/Female?src=hash">#Female</a> (<a href="http://t.co/seNuUOE8hK">http://t.co/seNuUOE8hK</a>)—@cmunell
Sometimes, NELL asks the same mind-hurting questions as any five year old:
True or False? "wonka-candy-bars" is a kind of <a href="https://twitter.com/hashtag/Candy?src=hash">#Candy</a> (<a href="http://t.co/LRQXAu2k3Q">http://t.co/LRQXAu2k3Q</a>)—@cmunell
Artificial intelligence doesn't come without controversy and concern over what this could mean for humans. "I do run into those people, and I don't think they're crazy," Mitchell said. "It doesn't bother me, in fact it makes me smarter. Because now I can have a computer assistant help me multiply numbers, and I think a reading assistant can play the same kind of role."
Over the years, computers have become better than humans at some things, like playing chess or multiplying numbers.
"We need to start thinking more carefully about what we want computers to do for us, and which types of computer assistants we want to develop."
There are a few projects working on this type of artificial intelligence. NELL's buddy NEIL, the Never Ending Image Learner, that was created by one of his colleagues, Abhinav Gupta.
That system is like NELL except it looks for objects and relations among them.
A year and a half ago, the projects had the idea to have NEIL and NELL talk. Now, every so often, a picture will pop up in NELL's database.
"They're sharing information and trying to teach each other, really."