Aiming to Learn As We Do, A Computer Teaches Itself

http://www.nytimes.com/2010/10/05/science/05compute.html?_r=1&ref=science&pagewanted=print

Computers are better than humans at specific tasks, but most computers are not as effective at jobs that require nuanced thinking. Semantics, understanding the meaning of language, is incredibly difficult for computers, because they not only have to learn the words, but understand them based on context and knowledge. Computers do not have the accumulated database that the human brain does. However, at Carnegie Mellon University, researchers are working on a computer that understands language like a human. The Never Ending Language Learning system, NELL, has scanned millions of Internet web sites, learned about 390,000 facts with a very high accuracy rate (87%) and grouped the facts into one of 280 categories. It has also learned about relations between different categories. Many computers have become faster and are leading to progress, but NELL is different-most learning systems are passive, NELL is automated. NELL is supposed to define words in different contexts, and it corrects itself when it learns more information. NELL occasionally needs help from humans, such as when it assumed that “Internet cookies” were baked goods and thus thought “files” were baked goods as well, but someday it may run completely independently.

This article is extremely relevant. Computers become more and more important every day, and if a computer were to learn independently, it would be a huge step forward for technology. Voice recognition software can help the computer perform different, more ambiguous tasks. This could end up becoming one of the most important technologies available. Computers that understand language could conduct more sophisticated search engines. Computers could become “personal assistants” in many different fields, answering sophisticated questions about medicine, the news, and various other topics.

While I thought this article was extremely interesting, there was a little too much detail as to how NELL relates text phrases to each other. This article took a worthwhile topic and presented it in an understandable way without too much technical language. I enjoyed learning about some of NELL’s different categories that it places subjects in and how an independent computer categorizes things. I would have liked to have learned more about the future of this research and what other applications they envisioned for this type of computer learning. Overall, this was a very important and interesting article, and I hope that we can learn more about NELL’s progress.