Carnegie Mellon University Researchers, under the funding of Google and DARPA, and benefitting from the internet database provided by Yahoo, are working on a self learning computer algorithm that is able to extract knowledge from the internet by itself, to learn language and facts.

The system is called NELL, which stands for “Never Ending Language Learning System”, scans hundreds of millions of web pages everyday, and learns facts by itself, at an accuracy rate of 87 percent. So far, NELL learned 390,000 facts, which are grouped into symantec categories, such as universities, actors, cities etc… The number of categories so far is 274 and this number is growing. These facts are things like “Rose is a flower” or “New York is a city”.

NELL is also capable of understanding relationships, from the facts it has learned. For example, even if it never reads anywhere that a football player is playing for a certain team, by knowing the facts that this person is a football player and also the other name is a football team, it can relate these facts and conclude that this person is most probably playing for that team, with a high degree of accuracy.

The researchers indicate that the efficiency of learning of the system will improve over time as it learns more and more. So far although there are human intervention sometimes, when the arrived conclusions are wrong, but researchers remind that even with human learning there needs to be corrections sometimes.

This project looks like a step forward in developing the AI. We must remember that we will not arrive at the humanlike AI all of a sudden. It will be in hundreds of small steps in the right direction, just like this one.

For original story, please visit here.

Reference: www.nytimes.com

  language, learn