The following post was originally shared on the Pythian Blog
Artificial intelligence, machine learning and data science are all terms that get thrown around a lot these days. While it’s easy to get into hair-splitting arguments about the distinctions between them, really they refer to the same thing: teaching machines to learn, think and convert data into knowledge.
Harnessing machines’ massive computational capabilities will allow organizations to do deeper, broader analyses faster than ever before, automate processes with real intelligence, and make better decisions more confidently.
In this blog series, we’re going to look at some fundamental data science concepts — starting with what it means for machines to learn in the first place.
Learning versus memorization
Learning is different from memorization. Memorization is what computers are really great at because they can store massive amounts of data and recall it instantly. Some people are really good at memorization, too. There’s the story of the man who memorized pi to more than 70,000 decimal places. Now, while that’s an amazing feat, does knowing all those digits tell you anything about the nature of a circle, or give you the idea to create a wheel, an axle or a car?
To do those things requires real learning.
So when we talk about machine learning we mean getting machines to take information they’ve stored or “memorized” and apply it, discovering and developing new concepts (new, at least, to the machine).
Making judgments based on experience
What does that “application” of knowledge look like? Well, say you’re a parent and you walk onto a playground with your kids. Within moments, you observe the climber and swings and make all kinds of decisions and judgments: will your kids have fun, will they be safe, etc. That happens because you have knowledge of all the parts of the situation: the structures, your kids, their tendencies. By combining this knowledge with what you observe about the specific environment, you make instant decisions.
The goal of machine learning is to get a computer to look at that same playground and be able to make similarly accurate decisions about whether or not kids will be safe, for example. It can’t be done just by memorizing rules about, say, weight restrictions on the swings. The computer has to apply knowledge of human behavior.
That kind of knowledge comes from modeling, which is a way of predicting how systems are going to work. It also comes from accumulating experience and, importantly, building on the knowledge of past mistakes to enable better, more accurate predictions going forward.
Giving machines that ability to build on the data they possess and their experience of the world will make them immensely powerful tools for everything from analyzing business data and optimizing systems to solving complex problems like climate change and finding cures for diseases.