Thursday, May 01, 2008

Machine Learning, Information Filtering and People

There is a lot of information on the internet now; billions upon billions of webpages. Some of the information out there is trivial but there is plenty of useful information out there. All that one needs to do it have a topic of interest and go off searching the internet to summarize the important details of a topic they need to know.

Unfortunately, there isn't a whole lot of good websites that summarize things very well. I usually end up finding myself surfing from wepbage to webpage looking for details, trying to confirm what I've learned. Quite often this ends up to be a very time consuming task. What if we could simplify this? What if we could find ways of going through large quantities of information very quickly, figure out what is reliable, what (seems to) make sense and then summarize it into a nice simple packet of information that we can digest quickly.

People as Information Filters

I've had a hypothesis for quite some time that people are information filters and the kind of information that is retained by the person is largely dependent on the character of the person. Now that I've mentioned the words information and character in the same sentence, I ought explain this relationship in more detail.

Though my experiences, I theorize that the character of a person is largely determined by the kind of information they find useful or interesting. As simple as this sounds, I think it is quite profound, as it isn't as stated often like this. If one were to think about it, the kinds of information that we find useful are the kinds of information that allow us to accomplish specific goals, albeit consciously or sub-consciously.

A person characterized as a thrill seeker will most likely find information about the latest scoup on cheap places for bungee jumping, white water rafting and the most exciting get-aways to be particularly useful. This person will obviously will have their eyes and ears peeled for information related to their interest, whether be it through magazines, clubs, newspapers or other forms of media. And I would be willing to go as far as to point out that the greatest source of information are other people with similar interests as it would be obvious that these people are also on the look out for similar information, digesting whatever pieces of data they get to summarize what they learn to actional pieces of data. You will often get the best information digest from other people that have done a fair amount of information and data collection, up to the point where they can anticipate the kinds of questions you will have and immediately fill you in on what you need/want to know.

I personally have to say that the best teachers I've ever had are the ones that know how to summarize information well, communicate it succinctly and be able to lead into the next important topic by anticipating the questions or the interesting points that arise from what was just presented.

I have found that dense exchange of well summarized useful and interesting knowledge to be rare and few. And I find this to be quite unfortunate, but this point is probably for a different essay at a different time, though I should point out that your best friends are likely to be the ones that share information that are the most relative with you... but anyways, I digress...

Machine: A man's best friend?

The old adage was that a dog is a man's best friend-- that you'd be able to send out your dog and have him fetch your shoes, slippers, a dead bird or the newspaper. But what if this time, instead of the newspaper, you had a dog that could surf the web and fetch you all the important things that you should/wanted to know. Imagine that.

Now, with the digitization of information, it has become easy for computers to digest digital representation of text-- in other words, you don't have to teach your computer have to read characters anymore (I still remember the first time I tried to read my elementary school teacher's handwritten comment on one of my assignments and I couldn't make heads or tails out of it)! Now with that problem out of the way, wouldn't it be interesting to see if we could give a computer some simple goals, have it go out into the internet and learn/find what you want to know and then summarize it for you.

What if you told your computer that you wanted to make gourmet food and have it come back at you and asked you what kind while providing a list of different ethnic cuisine. What if in addition to that, it went further to recommend certain dishes because of availability of ingredients based on your geographical region and the season. All of this relevant information would be nicely summarized into something that you can immediately act upon. This is what I believe to be the epitome of useful information (considering that most of information you get through the news and other sources are hardly actionable!).

Intelligence simply does not come from data!

Throughout my high school life, I realized something far more important than just learning what your teachers taught you, but rather more efficient ways of learning. In practice, I focused on finding more efficient ways of learning. I can tell you with certainty that effective learning does not result from wrote memorization. What is more important that just having information is developing relationships between pieces of information, like a puzzle, to create a coherent overview of the body of knowledge learned.

With the combination of having a wide breadth of knowledge and understanding how it fits together, it puts one in the desirable position of understanding the relevance of what they know to quickly understand how new information may fit into their knowledge or knowing how to respond to questions with the most relevant answers. After all, this is what tests are all about.

How do you teach someone/something to learn?

Now this is obviously the million (multi-billion?) dollar question (and if you know the answer, go start your own education company). There is no easy way to teach someone or something to learn. And I believe that the process of learning boils down to motive and objective. What I have found more important that instead of just teaching a subject, is to teach the underlying motivations behind why this body of knowledge was created because if a person could never find a purpose for what they learned, why bother learning it? I think that a topic should be as interesting as it is useful.

Motivating humans, I think, is a far easier task than motivating an inanimate object like a computer to do something. Humans have hard-coded biological objectives to survive and live a happy life. Obviously, by understanding the basic constructs of ourselves, it becomes easier to understand the relevance of information to ourselves. I should point out that the desire to be well informed is not just for survival, but also for entertainment. I don't know about you, but I find myself quite happy to discover or learn profound ideas (where I find myself running around to friends to tell them all about it... despite how little they may care sometimes :)

In order to develop an autonomous computer system capable of learning, it would be very necessary to hard code specific objectives into an artificial intelligence to give it the motivation to learn. Determining what those motivations should be will likely be a challenging topic, but I am sure that without them, a CPU made from silicon will have as much motivation to learn as a chunk of rock. And the same goes for students.

No comments: