“If (there) was one thing all people took for granted, (it) was conviction that if you feed honest figures into a computer, honest figures (will) come out. Never doubted it myself till I met a computer with a sense of humor.”
― Robert A. Heinlein, The Moon is a Harsh Mistress
This post is the first in a series of articles in which we will explain what Machine Learning is. You don’t have to have formal training or experience in data analysis. We will write using simple language, without unnecessary technical jargon. Thanks to our articles you’ll gain basic knowledge of Machine Learning.
Let’s start with the definition, of course. Then, standing on a balcony, we will be dropping sheets of paper, heavy bowling balls and golf balls. We will use this analogy to show you how machines learn.
What is machine learning?
There are many formal definitions of this branch of science. Let’s see what Wikipedia says:
“Machine learning is a subfield of computer science (more particularly soft computing) that evolved from the study of pattern recognition and computational learning theory in artificial intelligence. In 1959, Arthur Samuel defined machine learning as a “Field of study that gives computers the ability to learn without being explicitly programmed”. Machine learning explores the study and construction of algorithms that can learn from and make predictions on data.“
Unfortunately, this type of definition doesn’t bring us closer to understanding what Machine Learning is from the point of view of the everyday functioning of private and public institutions.
Machine learning is a set of tools which are guided by the philosophy of finding patterns, without the expectation of finding their meaning. The meaning exists, of course, but it’s beyond the analytical scope of the machine. You can make easy comparisons here.
Of course this is a very simple example and usually classification schemes are basing on a much more complex schemes using hundreds of variables, which unfortunately is not possible to illustrate using static graphics.
How machines learn?
Imagine that you are standing on the top of a high building. Scary, I know. Now, you are holding two types of objects – a white sheet of paper and black bowling ball. We drop them at the same time and watch them fall. The ball hits the ground much earlier. Everybody can easily explain this phenomenon: a sheet of paper has higher surface to weight ratio and is more easily stopped by air resistance.
But the computer doesn’t know that!
Computer observes the entire event. It knows which object fell quickly and which slowly. It knows the properties of those two objects: black and heavy bowling ball and white and light sheet of paper. But it doesn’t know what’s the connection between those properties and the result.
Here we go again dropping more heavy black bowling balls and white light sheets of paper. The computer learns that objects of these properties will fall respectively fast and slow. But the computer does not know which of these properties is important.
We begin to drop two more types of objects: black and light pieces of cardboard and white and heavy golf ball. Balls are still falling rapidly, while the cardboard and paper – slowly. Computer is watching it. It’s confused, but starts to understand that the essential property is the weight of the object not its color ! It uses that information to predict the future, for instance, will this object quickly hit the ground when dropped of the balcony?
It still doesn’t understand that the ratio of weight to surface is important (but if we show computer things that are heavy but have big surface (metal sheet for example) it will notice this property too and will be a bit closer to the truth), but it doesn’t need this information until there will be objects small and heavy, which will fall slowly… But even then it will not understand the law of universal gravitation, friction, air density and all other factors that even a high school student easily understands.
Its task is not to understand the sense of reality, but to find recurrent patterns, which we humans can make sense of and use them in our lives to predict the unknown.
In the next part we will answer the question – do we need machine learning at all?