Entries with tag "artificial_intelligence"

A new bit of AI has been developed which can teach itself new languages. You feed it some text, and it analyzes its structure looking for patterns, and thus learns the language's grammar. Once it has learned it, it can produce new, meaningful sentences in that language. That's a very cool use of two of my big interests—artificial intelligence and linguistics. I couldn't find any more information about this project aside from the single webpage linked above, but I'll have to see if more turns up, because I'm very interested in learning more about how it works. In particular, I wonder how it addresses the issue of vocabulary.

[Entry Continues...]

Today I discovered the slides for an interesting talk given back in 1997 by Nathan Myhrvold (CTO of Microsoft at the time) titled "The Next Fifty Years of Software". He makes some intriguing assertions, such as "Nathan's First Law of Software:" Software is a gas, that expands to fill its container. This in turn drives hardware development, and is what makes Moore's Law possible.

The part I really liked, though, was where he starts talking about the storage requirements of the human genome, etc. For example, the stuff that makes you genetically unique fits on a 3.5" floppy disk.

From there he proceeds to discuss the "ultimate computer," the human brain. He brings up something I've pondered as well: once computers reach a complexity comparable to the human brain, will it even be possible to program them directly as we do modern computers? Or will we have to devote time to teaching them, like we do with people? As he says, human takes 20 years to boot up.

I just discovered MIT's OpenCourseWare, which appears to be a repository of the course materials for most or all of MIT's courses, available for free and without registration on the web. I went straight to the Computer Science section and immediately found two courses on Artificial Intelligence (Fall 2002, Spring 2003). They actually appear to be the same course, but taught by two different teachers, and the materials seem pretty different. It will probably be interesting to go through both of them.

This website might be a godsend, since I tend to be pretty good at learning things on my own. Of course it's very helpful to have a knowledgeable professor to explain it and to ask questions of, but I can certainly pick up most of it from the course material alone. It's like it was made specifically for people like me — too lazy do do well enough in school to get into MIT, but willing to do a little extra work to learn some of the stuff anyway :)

Update: Actually, I hadn't seen the XTutor section for each course, which seems to be a full set of recorded audio lectures by the professors, along with matching lecture slides, full transcripts, and lecture handouts. There are also weekly online interactive homework problems. In other words, that's pretty damn close to what I would get by taking the course in person. Now I'm even more excited.