Aiming to Learn as We Do, a Machine Teaches Itself

Today, I presented my current events article from the NY times. My article was about NELL (the Never Ending Language Learning system), and how this super computer is breaking boundaries in terms of technology mastering semantics. It is actually quite superb when you read the article and see how amazing technology is getting these days. Researchers at the Carnegie Mellon University developed NELL with intentions of bringing human language to the world of computers. NELL revolves its “knowledge base” around 290 semantic categories such as actors, universities, cities, sports teams, etc; and connects all of these categories through various “relations”. NELL currently has 280 relations and scans hundreds of thousands of textual patterns and phrases to match up categories based on their relations to other categories. I thought it was quite amazing how NELL was highly automated, meaning that it could perpetuate its own curiosity and have endless knowledge with the millions of web pages out there. With NELL’s progress, we can actually have a computer that teaches itself based on the knowledge that has already been established through human establishment. I might just be a little paranoid, but while reading the article, I couldn’t help thinking about sci-fi horror films with dangerous super computers. Some movies that came to mind were Eagle Eye, and of course, Terminator’s Skynet.

But jokes aside, I asked the class if it was safe for NELL to be basing its knowledge base off of something as bias as the world wide web? From our midterm project, we determined that there are sources out there (Wikipedia) that are just not reliable. So, what if NELL were to get false information from sites like Wikipedia? Could NELL be identified as a credible source of information? I feel that one solution to the problem could be that the same or similar algorithm could be used, but instead of using the WWW as reference, it could narrow its database to credible sources such as ESPN for sports, CNN for news, etc. I understand that its knowledge wouldn’t be as extensive, but it wouldn’t be so volatile to false information or bias entries.

All in all, NELL is a great idea with a ton of potential. It’s algorithm is shaky, but with work, I’m sure that its future is bright. With NELL in mind, I am curious to find out where the future of A.I. is going. We now have a computer that can think for itself, what’s next? Robots?!?!