You only learn by mistake. But how much is ideal to make mistakes? 15%, says study.

Observing computers that learned simple tasks found that maximum performance

"It is by mistake that one learns," the saying goes. And science works just like this: you come up with a hypothesis and then perform an experiment that can refute or confirm it. But how much exactly must one fail before learning? If we make too many mistakes, will we get discouraged and end up getting nowhere?

A recent study by US researchers investigated the question and got a mathematical answer . Scientists have assumed that it is common sense between educators and educators: the student must be challenged to learn for real. Imagine having a ninth grader attend a class that teaches adding and subtracting. It's too easy, so it won't extract anything new from there.

On the other hand, put it in to follow a university quantum physics class to see what happens. Of the two, one: Either 100% of the things the teacher asks will be wrong, or the difficulty will be such that it will completely give up the challenge. That if it doesn't get trauma from the matter. It is clear that there must be a middle ground where the new task or content is neither so simple nor so complicated.

Research has found that this ideal point where we make just enough mistakes to keep us stimulated but not let down is 15% of the time. Or, on the other side of the coin, give the right answer 85% of the time - so the discovery was called the “85% Rule” . To get there, scientists conducted a series of machine learning experiments in which computers learned on their own to perform certain simple tasks.

These were things like classifying distinct patterns into two categories or differentiating handwritten numbers as even or odd. And it was potato: the algorithms had the best learning achievement whenever they hit 85% of the time. Previous animal studies have also revealed the same rate, and even for humans the 85% Rule seems to hold true - especially for so-called perceptual learning.

It works when we learn something slowly and on our own, cemented in our own experiences and examples. Like when a radiologist is recording in your brain how to identify an image showing a tumor from one in which there is no tumor. Only time and practice will help him: he needs to gain experience and collect examples to improve his judgment.

"If I give too easy examples, you'll get it right 100% of the time and there's nothing to learn," University of Arizona study leader Robert Wilson said in a statement. “If I give too many difficult examples, it will be correct 50% of the time and you won't be learning anything new yet, whereas if I give something in between, you will be at that ideal point where you get the most information out of each example in particular, ”said the professor of psychology and cognitive science.

But take it easy before you think taking 8.5 on a test is better than 10. Wilson and his colleagues have so far investigated simple tasks with binary answers: correct or incorrect. To apply the results to something as complex as education, we need to deepen and refine the study, published Tuesday (5) in Nature Communications .

Wilson hopes to expand the work to include more complex forms of learning as well. “If you're taking classes that are too easy and getting it all the time, you're probably not extracting them the same as someone in trouble, but finding a way to keep up,” he said. Now we can leave that famous expression a little more scientific: "It is missing 15% that you learn."