Friday, October 4, 2013

Machine Learning

An article on Wired.com, by Matt Kennedy, talks about artificial intelligence. It introduces this concept called machine learning. Machine learning (ML) is a field that intersects computer science and mathematics. The goal of ML is designing software systems that can make use of data without being explicitly programmed with information about the structure or contents of that data. ML is supposed to be counterintuitive to the general understanding that we have about how computer programs work. Computer programs are known by everyone to be a set of instructions and a computer executes that set of instructions. Even a person who does not know how to program a computer would know that a computer program is basically that. However, with ML, a machine learns from the input data through mathematics.

It is interesting what the author writes. Math does have the tendency to scare people. That could be the reason why some people would hate for a robot to be built with all this complex math and being able to understand it. Then, it would seem they know more than most people do and that could set up some extreme domination of robots over humans. However, understanding how it works through simple examples is completely possible. Here is what the article posted as a simple example:

“Imagine that you’re a real estate agent in a posh neighborhood and you have a new 5000 square foot house to list. How do you determine the listing price on your house? There have been three recent sales in the neighborhood: a 4000 sq ft house for $4M, a 4500 sq ft house for $4.5M and a 6000 sq ft house for $6M. With nice gigantic round numbers like this, it’s pretty clear that a reasonable price for the 5000 sq ft house is $5M. If you worked that out in your head, mathematically speaking, you just performed an algorithm called linear regression, and then performed a predictive analysis by using the known feature (5000 square feet) to ascertain the target information ($5M). Aren’t you clever?”

When I read this example, it was very simple. Anyone could understand this. Linear regression is the algorithm that allows computers to analyze data in the same way. We make predictions based on linear regression and that is how a computer does it. However, it is not that simple. What if the data from the previous example was not aligned so perfectly like that? He goes further into his example about real estate here. The amount of square footage is not the only determination of a household. The quality of it plays a factor as well. There could also be other factors such as age and location too. That would not allow for such a straight line, but does that mean that we as humans cannot understand? Certainly not. We are capable of learning anything a machine can. After all, we create the machines.


What this example explained was supervised learning. The other class of ML algorithm is unsupervised learning. Unsupervised learning is when a machine is given data to find patterns on its own, whereas supervised learning is when the patterns are actually already known by its programmers.  The reason I bring up this article is because there have been movies about artificially intelligent machines becoming self-aware and taking over the world. Some people may actually believe in that stuff. However, it is easy to understand how a robot is being made to be intelligent. The harder stuff is for those teams of programmers, mathematicians, domain experts, and data scientists. All the common person needs to know is that these machines are only being made to think like they do already. They won’t be smarter than you.

No comments:

Post a Comment