Researchers have taught computers to play all sorts of games, from chess to StarCraft. All that gaming has made computers smarter, allowing them to get better at many things–not just games.

As AI researchers come up with new ways to give computers humanlike abilities, they also need to make sure they work. It’s similar to your teacher giving you a test to see how you’re doing in class.

Games can test and compare the way computers and humans solve problems. Here’s how: researchers feed the computer the game’s rules and teach it the moves and strategies humans would use. Some of the first computers with AI learned to play checkers and chess this way, and it showed that computers with AI could do some of the same tasks as well as humans–but many times faster.

That’s partly because computers can store more in their short-term “memory” than we can.

For example, say you’ve told a computer that it needs to win a game of chess in as few moves as possible. You tell it some basic rules, like the piece called a bishop can only move diagonally. The computer then plays the game against itself, making moves and getting feedback about whether those moves helped it win. It is given some kind of reward, like scoring a point if the sequence was good. It might lose a point if the sequence was wrong. The computer keeps playing over and over until it can’t top the fewest number of moves that it previously won.

In time, the computer could come up with a winning strategy that even world-championship players haven’t tried, faster than a human ever could. Just imagine if you played the same game millions of times. Even chess champions don’t have the patience or time, but that kind of practice is how you keep getting better.

So, without even realizing it, you might have encountered something created via reinforcement learning. Maybe you have a robotic dog?

Let’s take a closer look at how this method could work with robotic dogs.

Even though reinforcement learning can make a computer very smart, it’s no substitute for the human brain. Humans have many qualities that computers don’t, like emotions and creativity. More than that, we have common sense. Your common sense helps you use your judgment to do logical things, like reading a book from the first page, not from the last. Or stopping at the curb when there’s oncoming traffic, even if the crosswalk signal tells you it’s OK to walk. When it comes to games, your common sense helps you to know the game’s end goal without ever being told. So we really need to help computers so that reinforcement learning doesn’t lead them down the wrong path. Working together, the human brain and the computer can come up with the best solutions to a variety of problems.

Now let’s take a look at how much smarter computers have gotten at learning how to play games over the years.

In the past six decades, computers have made lots of advances using artificial intelligence. Let’s take a look at how much smarter computers have gotten at learning how to play games over the years.

Imagine you’ve been asked to make something no one’s ever built before. You and your friends might sketch different ideas, build and test a few models, and then make changes based on what went wrong. Realistically, you couldn’t sketch everything you wanted to, especially if you had a deadline for completing the project.

Let’s see how it all came together.

Reinforcement learning can help solve lots of problems. And one day, it might even help solve some of the world’s toughest challenges, like making sure everyone has enough food to eat or combatting climate change.

Here are a few examples of how reinforcement learning can do the thinking for us:

As you can see, there are many different ways smarter computers can help humans. Which problems could reinforcement learning help you solve? Now that you’ve come to this point, think you can answer these questions? (Hint: the answers are all in this story!)