Learning something new — and quickly — may depend on the lesson’s difficulty level, according to a new study.
Now, scientists say they have cracked the code on the optimal level of difficulty to speed up learning. The team tested how the difficulty of training impacts the rate of learning in a broad class of learning algorithms, artificial neural networks, and computer models thought to simulate learning in humans and animals.
The researchers discovered a learning sweet spot — the point at which training is neither too easy nor too hard. In the sweet spot, learners make errors about 15 percent of the time, but their learning progresses fastest and additional effort has the biggest payoff.
“Eighty-five percent is the place where if you put in more cognitive effort and concentrate more, that will boost your performance the most, whereas with 50 percent accuracy or 100 percent accuracy, concentrating more is actually not going to boost your accuracy at all,” Robert Wilson, a researcher at the University of Arizona, Tucson and author of the study, tells Inverse.
The study was published in the journal Nature Communications.
The new findings poke holes in the idea that perfection should be the ultimate goal of learning. Although the study was in artificial neural networks, Wilson points to the real-world implications — namely, that we need to rethink what the education system values.
“We reward perfection maybe too much,” Wilson says. “Errors and mistakes are just a part of life and as we’ve shown here, a crucial part of learning.”
The 85 percent rule
When people learn something new — a language, instrument, or mathematical procedure — they gravitate towards challenges at the edge of their competence, a 2012 study suggests. Instead of trying to solve overly complex or simple challenges first, they tackle problems that are just pushing the boundaries of their understanding.
If people attempt tasks that are too hard, they’re likely to grow anxious, frustrated by not seeing results after repeated effort.
“Say you want to learn Spanish and you jump into a really difficult high-level Spanish class. You’re not gonna learn anything from that,” Wilson says.
At the same time, if people complete tasks quickly and easily, they get bored.
“It’s like taking a kindergarten spelling class when you’re in college,” he says. “That’s not to be a very engaging activity because you know everything already.”
The authors’ “85 percent rule” could provide a road map to reaching the ever-elusive “flow state.” People tap into flow when their skill level and the challenge at hand match. Athletes, surgeons, artists, and others report experiencing peak performance, effortlessness, and even euphoria during flow state.
During this 85 percent window of difficulty and during a flow state, concentrating and increasing effort can lead to big gains, no matter the activity you are engaging in.
In the classroom
Wilson’s study could help learners and teachers figure out where and when to focus their attention. It pays to hunker down on tasks just beyond understanding, rather than knock out easy questions or attempt overly difficult problems, which may seem impossible to master.
“As a learner, the thing to focus on is to make sure you’re pushing yourself and getting into this region of intermediate difficulty where you are making mistakes and getting things wrong and accepting that’s part of how you learn,” Wilson says.
Wilson encourages educators to distinguish between tests that train students and tests that assess them.
Rather than clumping all assessment into a heavily weighted midterm or final exam, incorporating more frequent, low-stakes testing will help students learn better.
“In the latter, educators should be pushing the difficulty so that students aren’t getting 100 percent all the time,” he explains. “They should be making mistakes on those tests because they can learn from those mistakes; not so many mistakes that they are getting discouraged, but a few mistakes here and there and they’re going to be okay.”
Researchers and educators have long wrestled with the question of how best to teach their clients be they humans, non-human animals or machines. Here, we examine the role of a single variable, the difficulty of training, on the rate of learning. In many situations we find that there is a sweet spot in which training is neither too easy nor too hard, and where learning progresses most quickly. We derive conditions for this sweet spot for a broad class of learning algorithms in the context of binary classification tasks. For all of these stochastic gradient-descent based learning algorithms, we find that the optimal error rate for training is around 15.87% or, conversely, that the optimal training accuracy is about 85%. We demonstrate the efficacy of this ‘Eighty Five Percent Rule’ for artificial neural networks used in AI and biologically plausible neural networks thought to describe animal learning.