Do Machines Really Learn?

Do machines really learn?

Do you believe that in a few years your autonomous vehicle will transport you wherever you’d like as you sit back and relax? And when you leave work, the car will ask you whether you’d like to head to your favorite restaurant or go home?

Some of you may already have a smart home, where lights turn on when you arrive and off when you leave, or a smart thermostat that heats the house while you’re there and reduces the temperature while your family is at work and school. Many in agriculture have combines and grain bins that practically drive themselves, and autonomous tractors are already here.

You may even wonder if in the future computers will be able to write articles like these. Could a robot win the Pulitzer? Could a machine ever learn to play basketball with the grace of LeBron James? They’ve already beat us at chess.

Machine learning is the basis of these and other strange questions.

Machine learning is here today, and it’s shaping and simplifying the way we live, work, travel and communicate. In fact, for good or bad, we use it dozens of times a day without thinking about it. One everyday example is the email spam filter.

Spam mail (also known as junk mail) is a type of electronic spam where unsolicited messages are sent by email.Every day,some of your email goes directly to the junk mail folder instead of your inbox. And if you’ve ever checked, most of the email in the junk mail folder really is junk. Don’t you wonder how it’s done?  (If you’re really interested in learning how Gmail spam filter works, here’s a link.)

When we think about machines, we usually visualize huge cranes or the robots now common in car factories. When we talk about machine learning, however, our concept of machine is quite broad. Machine learning can include any mechanical, electrical, or electronic device. A TV remote, car, thermostat, and even Gmail, can be called a machine.

One of the promises of modern smart machines is that they improve the speed and efficiency to complete tasks and analyze outcomes. Powerful smart machines can now accomplish in seconds tasks that would take humans days, weeks, or years to complete.

But machine learning is not like a robot learning from a book. Actually, it is a method of data analysis developed by computer scientists. It’s closely related to computational statistics, which focuses on making predictions using computers. It is a branch of artificial intelligence (AI) based on the idea that a machine can be trained to learn from its past experiences.

So when you move a message that ended up in your spam to your inbox, or mark a message you don’t want as spam, your email program learns from you and doesn’t repeat the mistake. At least, that’s the idea. Some emails, I don’t know where they end up!

If the question is if machines really learn, the answer is definitely yes. However, we need to understand what we mean by learning.

As a child we learn to recognize the objects with repeated exposure to the word as well as with its image and related sounds. For example, to learn what a car is, we are shown multiple images of different cars, we hear car noises and repeated use of the word “car” over and over. With multiple attempts we learn to associate the word “car” to the object. But this is natural. We don’t teach babies or toddlers how to do this. It’s part of how our brain is wired to learn language.