Your brain has a shortcut for hard problems, and it starts by ignoring most of them

What's the best way to learn a puzzle or solve a problem? Consider a task where you must predict the weather from mysterious symbols. Should you try to interpret all the clues at once, or master them one by one? A new study published in the journal Nature Human Behaviour tested this head-to-head.

Volunteers played a "weather prediction" game using simple geometric shapes as hints. Half the group trained on single shapes (one-cue trials), while the others always saw three shapes together (multi-cue trials). After this training, everyone faced the same three-shape puzzles without feedback. This design directly compared the two teaching strategies to see which worked better.

One by one: How isolated practice builds super-learners

From the start, the one-at-a-time trainees surged ahead. Those who practiced each shape separately picked up its meaning quickly. By the end of training (and on the final tests), the single-clue group was significantly more accurate than the all-at-once group.

Remarkably, this held even though they had seen fewer combined-clue examples: Mastering each hint individually helped them infer the full puzzle.

The study's authors say, "The weather prediction task is learned more effectively by dividing and conquering—training the cues one at a time." In other words, breaking learning into bite-sized chunks lets people build a stronger overall grasp of the task.

A hybrid learning strategy

Why did this "divide-and-conquer" trick work so well? To uncover the underlying mechanisms, the researchers built simple computational models of learning. They compared two main strategies: a "marginal" rule and a "joint" rule. The marginal rule updated each clue in isolation, as if it alone determined the outcome. The joint rule, conversely, updated all clues at once, considering their combined influence.

Neither strategy alone fully matched the observed human behavior. The true winner was a hybrid model that cleverly blended both approaches.

This model suggested that our brains don't stick to a single learning method. Instead, they dynamically switch between strategies depending on the situation.

Picture a trial in which evidence strongly suggests the wrong person was charged. In this case, the hybrid model would mainly apply the simpler marginal updating, which focuses on that one piece of information. If the outcome depended on the complex combination of many cues, though, the model would engage in the more costly joint updating of all cues. Such dynamic arbitration facilitates effective learning.

The brain gets to figure out the exact "cue-weights"—the actual weights of the clues, without having to deal with confusion and overload in cognition. The brain, as such, appears to be able to balance fast-paced, sequential learning with deliberate integration. In this sense, it seems that dividing and conquering works well because of this process of dynamic arbitration.

Curriculum by design

The hybrid model didn't just explain the data—it made new predictions for better teaching. For example, it suggested that training on examples with very clear outcomes (strongly "sun" or "rain" cues) would accelerate learning. The team tested this by using a skewed curriculum loaded with extreme cases, and indeed, those learners improved even faster.

As the authors note, these results show "how computational insights on learning can guide the design of curricula that accelerate learning." In short, it matters not only what you learn, but how the lessons are structured. This insight could help teachers craft lessons that start with obvious examples before adding subtler cases.

Curiously, many machine-learning systems get little benefit from such staged training, implying that humans have a special knack for this divide and conquer approach. As the paper concludes, "Curriculum design is central to efficient learning in fields ranging from human education to artificial intelligence."

Break a big problem into small steps—that's the clear takeaway. This experiment offers concrete proof that there is no substitute for the age-old saying that one must "start from the bottom."

Once all of the pieces have been mastered individually, they can then be combined to solve the overall problem. The next time that you encounter a tough problem or train another person in a set of skills, keep in mind the saying "one piece at a time."