Computer Reads and Learns From the Internet

Carnegie Mellon University researchers have taught a computer to read and learn. The computer called Nell, which stands for Never Ending Language Learner, learns by discovering information on the Internet, comparing the information with other sources, and attempts to decipher those information as beliefs or facts. This learning process is similar to how humans learn.

CMU researchers say that Nell has learned over 440,000 distinct things and has an accuracy of 74 percent in trying to discern beliefs and opinions. The way that NELL discerns an opinion from a belief is done in one of two ways. First, if NELL recognizes that a certain statement of fact comes from a single authoritative source, it will assign that statement as a belief. Also, if several lesser authoritative sources arrive at the same statement, NELL will similarly classify it as as belief.

Right now, there are several problems with NELL’s learning. First, NELL can’t unlearn beliefs once it has assigned statements as beliefs. So if NELL had at one point learned that the world is flat, and new scientific facts emerge saying that it is now round, NELL can’t update its learning.

Additionally, NELL can also make mistakes, especially when it comes to jargons and idioms. NELL has learned that Klingon is an ethnic group, for example, and that right posterior is a body part.

This article was filed in Homepage > Computers > Robots and was tagged with computer.
Like us, and get the best stories

User Comments