How Experts See What Beginners Miss

Show a chess master a position from a real game for five seconds, and the master can rebuild much of the board from memory. A beginner places about four pieces. Scatter the same number of pieces at random, and the master’s advantage nearly disappears. William Chase and Herbert Simon reported this in 1973 (Chase & Simon, 1973), and it shows what expertise mostly consists of: a different way of perceiving the same information, built from years of exposure to one kind of material.

What de Groot found in grandmasters.

The work started in the Netherlands with Adriaan de Groot, a chess master and psychologist. In the 1940s he asked players to think aloud while choosing a move and recorded everything they considered. He expected the grandmasters to calculate further ahead than strong club players, but their searches were no deeper and they did not consider more options. The difference was that the moves they considered at all, within the first few seconds, were better moves. De Groot concluded that the difference lay in perception rather than calculation.

Chunks, and the limit they work around.

Chase and Simon explained how that perception works. When players rebuilt a position, the pieces went down in quick bursts of several at a time, separated by pauses. The masters were recalling a handful of groups rather than twenty separate pieces. Chase and Simon called those groups chunks: sets of elements seen together so often that they are perceived as one thing. Where a beginner sees a knight, a pawn and a king on three squares, a master sees a castled kingside. A random board contains few of those familiar groups, so the master is left placing pieces one at a time like everyone else.

The popular version of the random-board result is slightly too strong. Fernand Gobet and Simon later found that masters keep a small but reliable advantage on random boards, probably because random positions still contain a few familiar fragments (Gobet & Simon, 1996). The advantage collapses without disappearing entirely.

Chunking matters because of a limit on working memory. George Miller put it at about seven items (Miller, 1956), and Nelson Cowan later argued the real figure is closer to four (Cowan, 2001). Miller’s more important point was that the limit applies to items, and what counts as an item changes with learning. A child learning to read holds the letters C, A and T as three items. An adult reads whole phrases at a glance with the same limit. Expertise leaves the ceiling where it is and changes what fits under it.

A memory for running times, and for nothing else.

The clearest demonstration came from K. Anders Ericsson and colleagues. They worked with an undergraduate known in the literature as SF, whose digit span was an ordinary seven. After more than 230 hours of practice, SF could hear 79 digits read aloud and repeat them back (Ericsson, Chase & Faloon, 1980). SF was a competitive runner, and he had learned to hear groups of digits as running times: 3-5-2 became a mile in three minutes and 52 seconds, a single item. When he was tested on letters instead of digits, his span was back to about six.

What SF built was a large, very specific set of chunks for one kind of material, and it did not carry over to anything else. Expertise works the same way: it is a library of patterns in one domain, and outside that domain an expert is an ordinary performer.

Recognition in the clinic and on the fireground.

Harold Kundel and Calvin Nodine showed experienced radiologists chest films for two-tenths of a second, too short to move the eyes around the image. The radiologists identified 70% of the abnormal films from that single glance, compared with 97% when they could look for as long as they liked (Kundel & Nodine, 1975).

I have seen the same thing in physical therapy education. Put an experienced clinician and a first-year student in a room watching the same patient walk twenty feet. The student sees knees bending and hips moving. The clinician sees a compensated Trendelenburg gait and is already thinking about hip abductor and external rotator strength before the patient reaches the wall. The student is often working harder to see it. Both receive the same information, and only one of them has the chunks.

Gary Klein found the same pattern in people who make decisions under pressure, including fireground commanders. He expected them to compare options and pick the best one. Instead, experienced commanders usually generated a single course of action, ran it forward in their heads, and acted on it if it worked. They were recognizing the situation more than deciding among choices. In one case Klein describes, a lieutenant at what looked like a routine kitchen fire pulled his crew out because the fire was not behaving as it should, and the floor collapsed moments later. The fire had been burning in the basement beneath them. The lieutenant credited a sixth sense. Klein’s account is that the situation failed to match any of the many house fires he had seen.

The same patterns can block better answers.

Merim Bilalić, Peter McLeod and Fernand Gobet gave expert chess players a position with a familiar five-move checkmate and a less familiar three-move solution. Most found the familiar one quickly and never found the shorter one. The players said they were searching the board for alternatives, but eye tracking showed their gaze kept returning to the squares involved in the solution they already had (Bilalić, McLeod & Gobet, 2008).

The recognition that makes an expert fast also narrows what the expert sees next. In clinical reasoning this is called premature closure: settling on the first plausible explanation and no longer considering reasonable alternatives. In a study of 100 diagnostic errors made by internists, it was the single most common cognitive cause (Graber, Franklin & Gordon, 2005). It is a hazard that comes with experience, because a novice does not yet have patterns strong enough to take over attention.

The one thing to do.

The next time you reach an answer in your own field, such as a diagnosis, a plan or a judgment about someone’s work, stop and ask one specific question: what would I be seeing if this were something else? Name the alternative out loud, then look for the finding that would separate the two.

Asking yourself whether you are sure does not work, because the answer that came to mind first usually feels right. Trying harder did not help the chess players either; their eyes still went back to the familiar squares. Naming a specific alternative gives your attention somewhere else to go.

If you are earlier in the process and still building those patterns, the post on the week-three wall covers what the practice itself looks like.

What this does not show.

Recognition-based judgment is only as good as the experience behind it. Daniel Kahneman and Gary Klein, who disagreed for years about expert intuition, eventually wrote together that intuition deserves trust only in environments regular enough to learn from, and only after long practice with clear feedback (Kahneman & Klein, 2009). Confidence alone is not evidence that those conditions were met.

Several of these findings rest on small samples. SF was a single participant, and the radiology study used a small group of radiologists. The chess findings have been replicated many times, but chess is an unusually structured domain. The question above is a habit drawn from this research rather than an intervention these studies tested, and chunks built in one domain do not carry over to others, as SF’s letter span showed.