Imagine calculating a variation. You mentally advance a pawn, your opponent replies with a knight, you develop the bishop, then bring in a rook. As you try to continue, you lose the thread.

You no longer know exactly where the black knight is in the calculated position.

This frustrating lapse is not a verdict on intelligence or effort. The representation of the variation was not maintained correctly; overload or distraction may contribute.

Baddeley’s model: what working memory is

Baddeley and Hitch propose a multicomponent model in 1974. Baddeley (2012) traces its development and discusses competing models. The episodic buffer was added in 2000, not part of the original 1974 model.

Working memory involves active maintenance and manipulation during processing, rather than only temporary storage. Its components include:

The central executive: an executive system coordinating the whole, allocating attentional resources and managing exchanges between subsystems.

The phonological loop: maintains verbal and auditory information, such as words in your head or arithmetic you verbalise mentally.

The visuospatial sketchpad: maintains visual and spatial information relevant to representing a board and pieces during calculation. A functional component of a model should not be confused with one anatomical region.

The episodic buffer: integrates information from subsystems with long-term memory, linking working information and stored knowledge.

Cowan (2001) discusses capacity of about four units under conditions limiting rehearsal and long-term knowledge. This is not a four-piece counter: task, attention and grouping change what a unit represents.

Why chess places demands on working memory

A game can involve several demands at once.

The current position: even with a visible board, you need a representation to use in calculation.

Calculated variations: every “if I play A, they play B, I reply C” requires following successive positions changed by moves.

Opponent threats: keeping track of active threats adds information to retain and check.

Strategic plans: holding a medium-term plan, such as preparing a kingside attack, improving a knight through a particular square or creating a passed pawn, while calculating immediate tactics.

In a simple position with few active pieces and one issue, the demands may be manageable. In a tactically complex position with several threats or a sacrifice to assess, following everything can become harder.

You may forget a piece identified as threatened three moves earlier, lose track of counterplay you noticed, or play a line calculated only partway. These observations do not establish a fixed capacity or a unique cause for every mistake.

Experts’ apparently “expanded” memory

Why do grandmasters appear to have much greater working memory than beginners?

Good position memory alone does not measure general working-memory capacity. Sala and Gobet (2017) distinguish domain expertise, correlations with cognitive abilities and training effects.

A library of chunks, familiar configurations stored in long-term memory, can help organise information. A position may be recognised through a familiar castled setup, a knight on f6 and a Sicilian pawn structure rather than treated only as separate pieces. “Three groups instead of twenty pieces” illustrates the idea; it is not a measured count for every expert.

Knowledge can support retaining a base position and retrieving details while following new variations. It does not remove the need to check exactly where pieces have moved.

Disorganising positions disrupts use of familiar configurations. That does not establish that all expert advantage disappears in every task: structural recognition and general capacities must be distinguished.

Training working-memory demands in chess

Calculation practice rehearses operations useful in chess. It does not demonstrate a general increase in memory capacity. The following are chess-training suggestions without guaranteed timescales or gains.

Chess-specific exercises:

Calculate without touching: follow a variation mentally before playing it. Increase the target depth gradually, for example from two moves to three, then four, while checking accuracy.

Blindfold tactical problems: observe a position for about thirty seconds, hide it or close your eyes, then try to calculate a solution. Check the board afterwards. This increases reliance on mental representation without isolating a single memory subsystem.

Replay from memory: choose a game you know through moves 15–20 and try replaying it without the book, checking afterwards.

Timed calculation: choose a position and record time needed to follow a variation to a defined depth. Compare accuracy as well as speed; a faster answer is not automatically a better representation.

What general training does not guarantee

Sala and Gobet (2017) distinguish near from far transfer. You can improve on a memory task and similar tasks without improving distant skills. A N-back application is therefore not an established way to gain Elo points.

Noise, stress or perceived fatigue may also make calculation less stable. The consulted sources provide neither a fixed percentage of capacity lost during play, an exponential law of errors nor a universal deadline after which training gains disappear.

To evaluate work, note what you can follow: a moved piece you forget, an exchange you misrepresent or a threat you neglect. Return to a suitable position and check your representation on the board. That provides a concrete objective without converting a laboratory model into a promise of improvement.

Sources

  • Baddeley, A. (2012). Working Memory: Theories, Models, and Controversies. Annual Review of Psychology, 63, 1–29. Full review.
  • Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24. Full article.
  • Sala, G., & Gobet, F. (2017). Does Far Transfer Exist? Current Directions in Psychological Science. Review.

Further reading