Sunday, March 2, 2008

Machine Learning vs. Artificial Intelligence

Artificial Intelligence and Machine Learning are two independent topics that are often used interchangeably by the general public. Artificial intelligence, in general terms, is a field dedicated to creating autonomous machines (think robots from any science fiction movie.) Machine learning is a sub field of artificial intelligence that is primarily concerned with the development of an algorithm that allows computers to learn.

Artificial intelligence starts at the top of the complexity pyramid because it is concerned with making a machine replica of a human being. Machine learning, however, starts at the bottom because it is only concerned with giving machines the power to learn.

Machine learning algorithms have tremendous potential in medical research because if computer programs are able to learn, then they will be able to study.

For example, every day we have more and more medical research that is being produced by scientists all around the world. This is great, except that we are outputting research at a rate far greater than any human can take in and understand. Thus, there are likely related studies that are being published, but not being linked together because researchers of one study are not aware of the related studies. Imagine a study found that rats who do not eat enough carrots begin to develop a disease (exclusive to rats) which is caused from a lack of vitamin A. Imagine also, that a recent study found that a community in South America had an abnormally high rate of blindness in their elderly population. This study looked at different issues, including the diet of the community. Finally, we have a third study that indicates that vitamin A is required to maintain healthy eyes. If you had a server farm that was capable of machine learning and you dedicated this farm to studying medical research, these machines could discover that the solution to the problem faced by the community in South America was that they did not have enough carrots in their diet. The machines would be able to figure this out because they could have learned that:

1) People were going blind in South America (Study 2)
2) Eyes require vitamin A to function properly (Study 3)
3) Vitamin A is found in carrots (Study 1)

Now, this example is obviously ridiculous because it is already well documented that we need vitamin A to maintain the health of our eyes and that carrots are a source of vitamin A. However, the point is that machine learning is powerful because computers can be dedicated to studying our existing knowledge base and looking for connections that we have not yet discovered. Thousands of machines can be put to task and look for connections between existing research at a fraction of the cost a single human would demand in salary. Also, a computer can work 24 hours a day and does not need a bathroom break. I am interested in finding an algorithm that will allow us to use computers to connect medical research.

My next few posts will present some of my thoughts about machine learning.

2 comments:

Robert said...

I find it interesting that ML would be a subfield of AI, considering how different you say the approaches are. I am considering studying Artificial Intelligence. Thanks for the informative article.

Richard C. Lambert said...

Machine learning, however, starts at the bottom because it is only concerned with giving machines the power to learn. speech recognition software