Experts Don't Think Harder. They See Different.
In 1973, researchers proved chess masters don't have better memory. They have better patterns. That discovery explains how expertise actually works in every field.
In 1946, Dutch psychologist Adriaan de Groot showed chess players a mid-game position for a few seconds, then took the board away. Reconstruct what you saw.
Masters nailed it. Nearly perfect recall. Novices got maybe 4 or 5 pieces right out of 25.
The obvious conclusion: masters have better memory. Bigger mental hard drives. Some kind of innate gift for holding information.
That obvious conclusion was wrong.
The Scramble Test
In 1973, William Chase and Herbert Simon at Carnegie Mellon ran de Groot's experiment again with one twist. They showed players two types of boards: real game positions and boards with pieces placed randomly.
On real positions, the results matched de Groot's. Masters crushed it. Novices struggled.
On random boards, masters performed no better than novices.
Read that again. Grandmasters with decades of experience, people who could recall a real chess position almost perfectly, dropped to beginner-level memory when the pieces were scrambled. Their advantage wasn't memory. It was pattern recognition. They weren't remembering individual pieces. They were recognizing familiar configurations.
A castled king. A fianchettoed bishop. A pawn chain controlling the center. Where a novice sees 25 separate pieces that need to be memorized individually, a master sees 5 or 6 familiar structures. Each structure is a single unit in memory.
George Miller called these units "chunks" in his famous 1956 paper on the limits of short-term memory. You can hold about 7 things (plus or minus 2) in working memory at once. But a "thing" can be a single letter or an entire meaningful pattern. The trick to expertise isn't expanding your memory. It's expanding what counts as one chunk.
Fifty Thousand Patterns
Fernand Gobet and Herbert Simon estimated in 1996 that chess masters have accumulated somewhere between 50,000 and 100,000 chunks in long-term memory. Built over years of deliberate study. Organized into what Gobet and Simon called "templates," which are larger structures with variable slots that can accommodate different specific configurations.
That's why masters can handle positions they've never seen before. They're not doing lookup on an exact match. They're recognizing a familiar template with new pieces filling the variable slots. It's the difference between memorizing every sentence you've ever read and understanding English grammar well enough to parse sentences you've never encountered.
This happens everywhere expertise exists.
Expert radiologists don't examine a chest X-ray pixel by pixel. They see familiar configurations. Experienced cardiologists have been shown to identify abnormalities in cardiac images within 200 milliseconds, far too fast for conscious analysis. They're not thinking. They're recognizing.
Expert programmers see code the same way. I notice this when I'm reading through a codebase. I don't process it line by line anymore. I see patterns. A React component with a state hook and an effect. A middleware chain. An error boundary wrapping a lazy-loaded route. Each of those is one chunk, not fifty lines of syntax. When I started coding, every line was its own thing to parse. Now entire files compress into a handful of recognizable structures.
That compression is what expertise feels like from the inside.
Mental Representations
Anders Ericsson and Robert Pool describe this in their 2016 book Peak as building "mental representations." Rich, precompiled structures that let experts see meaning where novices see noise.
This reframes what expert intuition actually is. When a doctor walks into a room and immediately senses something is wrong with a patient, that's not mystical gut feeling. That's pattern recognition running on a library of thousands of encoded cases. The doctor's brain is matching the current input against a massive database of previous patterns, and flagging a mismatch. It feels effortless because the processing is happening below conscious awareness, but it was built through years of effortful encoding.
Schmidt and Bjork (1992) noted that this is precisely why the conditions that build expertise feel so different from the conditions that demonstrate it. The performance looks smooth, automatic, intuitive. The practice that built it was slow, frustrating, and full of errors. Nobody sees the construction. They only see the finished building.
Why Transfer Fails
Here's the uncomfortable part. Those 50,000 chunks a chess master has built? Completely useless in medicine. Or law. Or programming. Or anything else.
This is one of the most replicated findings in cognitive science. Thorndike and Woodworth showed it back in 1901. More recently, Sala and Gobet (2016) ran a meta-analysis on chess instruction and found no meaningful transfer to academic or cognitive skills. Same story with music training. Sala and Gobet (2017) again found no reliable transfer from music to general cognitive ability.
Simons and colleagues reviewed the entire "brain training" industry in 2016 for Psychological Science in the Public Interest. The conclusion was definitive. Training on specific tasks makes you better at those specific tasks. It does not make you generically smarter.
Barnett and Ceci (2002) mapped out exactly how transfer works in their taxonomy framework. The more similar the training context is to the target context, the more transfer you get. The further apart they are, the less. "Near transfer" is real. "Far transfer" is mostly a myth.
This destroys the idea that learning chess makes kids better at math, or that brain games make you sharper at work, or that studying Latin helps you think more logically about everything. Your brain builds domain-specific architecture. Chess chunks help with chess. Code patterns help with code. Medical patterns help with medicine.
The chunks are the expertise, and they don't travel.
What This Means for Practice
If expertise is really about accumulating thousands of domain-specific patterns, then the way you practice determines what patterns you build.
Blocked practice, doing the same thing over and over, builds shallow pattern recognition. You get fast at recognizing one specific configuration. Interleaved practice, mixing different types of problems, forces your brain to build the discriminative patterns that let you tell configurations apart. That's why Rohrer and Taylor (2007) found that shuffling math problems improved learning. Mixing forces you to build chunks for "when to use what," not just "how to do this one thing."
Retrieval practice matters here too. Roediger and Karpicke (2006) showed that testing yourself, actually pulling information out of memory, strengthens pattern encoding far more than restudying. Every time you successfully retrieve a chunk, you reinforce the neural pathway that makes that pattern accessible. Every failed retrieval attempt, as Kornell, Hays, and Bjork (2009) demonstrated, actually primes the brain to encode the correct pattern more deeply when you encounter it.
Productive failure fits the same framework. Kapur (2008, 2014) showed that struggling with problems before receiving instruction builds richer mental representations than instruction-first approaches. The struggle isn't wasted time. It's building the scaffolding that expert chunks will eventually attach to. Loibl, Roll, and Rummel (2017) confirmed this in their review, finding that problem-solving followed by instruction consistently outperformed direct instruction alone.
Seeing Differently
The practice paradox keeps showing up in new forms. The strategies that feel productive (rereading, blocked repetition, avoiding errors) build surface-level familiarity. The strategies that feel frustrating (testing yourself, mixing problems, struggling before you get help) build the deep pattern libraries that actually constitute expertise.
A chess master doesn't experience the board the way you and I do. A veteran programmer doesn't experience code the way a bootcamp student does. An experienced doctor doesn't experience a patient presentation the way a first-year resident does.
They're not smarter. They're not thinking harder. They've built a different perceptual world through years of the kind of practice that felt, at the time, like it wasn't working.
Your brain right now contains every chunk you've built through practice. Every pattern you've encoded through struggle and retrieval and failure and correction. That library is your expertise. And it's still under construction.
The only question is what patterns you're building next.
Sources
- Thought and Choice in Chess (De Groot, 1946/1965)
- Perception in Chess (Chase & Simon, 1973, Cognitive Psychology)
- The Magical Number Seven, Plus or Minus Two (Miller, 1956, Psychological Review)
- Templates in Chess Memory (Gobet & Simon, 1996, Cognitive Psychology)
- Peak: Secrets from the New Science of Expertise (Ericsson & Pool, 2016, Houghton Mifflin Harcourt)
- New Conceptualizations of Practice (Schmidt & Bjork, 1992, Psychological Science)
- The Influence of Improvement in One Mental Function upon the Efficiency of Other Functions (Thorndike & Woodworth, 1901, Psychological Review)
- Do the Benefits of Chess Instruction Transfer to Academic and Cognitive Skills? (Sala & Gobet, 2016, Educational Research Review)
- When the Music's Over: Does Music Skill Transfer? (Sala & Gobet, 2017, Educational Research Review)
- Do "Brain-Training" Programs Work? (Simons et al., 2016, Psychological Science in the Public Interest)
- When and Where Do We Apply What We Learn? (Barnett & Ceci, 2002, Psychological Bulletin)
- The Shuffling of Mathematics Problems Improves Learning (Rohrer & Taylor, 2007, Instructional Science)
- Test-Enhanced Learning (Roediger & Karpicke, 2006, Psychological Science)
- Unsuccessful Retrieval Attempts Enhance Subsequent Learning (Kornell, Hays, & Bjork, 2009, Journal of Experimental Psychology: Learning, Memory, and Cognition)
- Productive Failure (Kapur, 2008, Cognition and Instruction; 2014, Journal of the Learning Sciences)
- Towards a Theory of When and How Problem Solving Followed by Instruction Supports Learning (Loibl, Roll, & Rummel, 2017, Educational Psychology Review)
Part of the Practice Paradox series. Previous: 10,000 Hours of What, Exactly?. Next: Your Brain Learns Nothing When You Get It Right.
