Failing Better

By Victoria Clare

Posted March 1, 2021

Media Player Error
Update your browser or Flash plugin

Transcript

Y: What are you waving that hula hoop in front of your cat, Don?

D: I’ve been trying to teach him to jump through it. He’s a pretty smart cat, so I figured he could learn a trick or two.

Y: How’s it going?

D: Well, so far, he’s just been looking at me like I’m crazy.

Y: Maybe you should try something a little simpler first. But not too simple—you have to let him fail a little if you want him to really learn. Actually, some scientists have an exact figure for how much failure is optimal for learning: they say that learning is most effective when it involves failure 15% of the time. Or, in other words, succeeding 85% of the time. They came to that number after a series of machine learning experiments where they taught a computer simple tasks such as classifying patterns into one of two categories, or sorting handwritten numbers as even or odd, or high or low. The computers learned best when they completed the task correctly 85% of the time. Even though the study focused on machine learning, it fits well with something education researchers and psychologists have noticed about how humans learn— that people seem to learn best when a task is challenging, but not impossible. If you get the answer right 100% of the time, you’re not learning anything new. But if you get the answer right, say, 25% of the time, you might become discouraged and give up before any learning really sinks in.

D: So you think that’s what’s happening to my cat?

Y: No, I think more likely that your cat really does think you’re crazy.

New studies propose an ideal amount of failure to better aid in the learning process. (Michael Surran, Wikimedia Commons)

Some scientists have an exact figure for how much failure is optimal for learning: they say that learning is most effective when it involves failure 15 percent of the time.

Or, in other words, succeeding 85 percent of the time. They came to that number after a series of machine learning experiments where they taught a computer simple tasks such as classifying patterns into one of two categories, or sorting handwritten numbers as even or odd, or high or low.

The computers learned best when they completed the task correctly 85 percent of the time. Even though the study focused on machine learning, it fits well with something education researchers and psychologists have noticed about how humans learn— that people seem to learn best when a task is challenging, but not impossible. If you get the answer right 100 percent of the time, you’re not learning anything new. But if you get the answer right, say, 25 percent of the time, you might become discouraged and give up before any learning really sinks in.