Expertise transforms the structure of cognitive representations rather than merely accelerating the same processes used by novices. Chase and Simon's chess studies showed that masters recall meaningful board positions far better than beginners, but not random positions — revealing that expertise consists of a large repertoire of perceptual chunks stored in long-term memory. The long-term working memory theory (Ericsson and Kintsch) proposes that experts create retrieval structures in long-term memory that effectively extend their functional working memory capacity within their domain.
Compare chess master versus novice recall of structured versus random boards — the crossover interaction (masters superior for real positions but not random arrangements) is one of the clearest demonstrations of chunking. The ten-year rule for expertise contextualizes why chunk acquisition requires sustained deliberate practice.
Working memory, as you learned in studying the working memory model, has a sharply limited capacity — roughly 4 items in the relevant stores, managed by the central executive. This is the bottleneck that separates novices from intermediate learners: a beginning chess player trying to evaluate a position must hold individual piece locations in working memory, quickly exhausting capacity while barely scratching the surface of what needs to be considered. But if working memory capacity doesn't expand with practice, how do experts operate so effectively? A chess grandmaster surveys a complex position and immediately perceives the right plan — clearly doing something fundamentally different from the beginner. The answer is that experts don't work with individual elements at all.
The key phenomenon is chunking, demonstrated rigorously by Chase and Simon (1973). They showed chess players positions for five seconds — either from real games or random arrangements of the same pieces — then asked players to reconstruct the board from memory. Chess masters showed dramatically better recall of real game positions than novices. The critical finding was the *crossover interaction*: for random board positions, masters were no better than beginners. This asymmetry proves the mechanism. Masters weren't simply better at visual memory; their advantage appeared only when positions contained patterns from real play. What they had stored were chunks — familiar configurations of pieces that commonly occur together, held as single retrievable units. A master recognizing a "kingside fianchetto with castled king" isn't processing 7 pieces; they're processing one chunk.
The implications extend far beyond chess. Long-term working memory theory (Ericsson & Kintsch) proposes that experts build elaborate retrieval structures in long-term memory — organized schemas that allow rapid encoding and retrieval of domain-relevant information. This effectively extends functional working memory capacity within the expert's domain: a physician doing a clinical workup can hold vast amounts of patient information "in mind" because they are encoding it into pre-existing diagnostic schemas rather than maintaining raw facts in short-term storage. A radiologist scanning an X-ray isn't processing pixels — they are comparing the image against thousands of stored patterns of normal and abnormal anatomy, matching at a level that compresses the cognitive work dramatically. Your prior study of cognitive load theory is directly relevant here: experts have lower intrinsic load for their domain precisely because chunked representations drastically reduce the number of elements requiring simultaneous working memory attention.
Why does this require so much time to develop? The ten-year rule (roughly 10,000 hours of deliberate practice) reflects the accumulation time needed to build a large, well-organized chunk library. Simply performing a task repeatedly isn't enough — the practice must be deliberate: focused at the edge of current competence, with immediate informative feedback and specific targets for improvement. Even then, the resulting expertise is domain-specific. A chess grandmaster confronting a novel domain starts nearly as blank a slate as anyone else, because their long-term working memory retrieval structures are organized for chess patterns, not the new domain's patterns. This specificity is the most important constraint of expertise, with real implications for how we evaluate expert testimony, design education, and think about whether "experience" in one field transfers to another. Expertise generalizes far less than people typically expect — and this is precisely what the structure of chunking predicts.