External representations perform real computational work

What it is

An external representation is not passive display; sorting, alignment, grouping, color, path, and spatial layout themselves perform reasoning. A table lets people compare numbers, a map makes a route visible, a Gantt chart exposes conflict, a checklist turns "don't forget" into a checkable state. This card takes the previous card's claim — that cognition is distributed across person, tool, and environment — and grounds it in something a designer can directly manipulate: if cognition really is distributed into representations, then the structure of a representation is not just a layout aesthetics question; it determines how much reasoning work the interface can do on the user's behalf within that cognitive system.

Why it happens

Different representational structures can "compute on the user's behalf" because they encode a relationship that would otherwise require step-by-step mental derivation directly into a spatial arrangement the eye can take in at a glance: arrange numbers in ascending order in a column, and "which is bigger" no longer needs computing — a glance at position gives the answer. Draw tasks as a Gantt chart by start and end time, and whether two tasks conflict shifts from having to mentally compare two time intervals to simply seeing whether two horizontal bars overlap on the chart — the eye recognizes spatial overlap nearly instantaneously, while comparing two abstract numeric ranges requires actively engaging working memory. This is why a good representation "front-loads" computation into the interface's structure itself: the moment a designer builds a representation, they are performing, once, on behalf of everyone who will ever use it, a conversion of some comparison and reasoning that would otherwise require mental effort into a one-time act of spatial encoding — all a user has to do afterward is look, not compute.

Where it stops holding

A representation is not neutral; while computing on the user's behalf, it also makes a choice on the user's behalf about which relationships matter, and if that choice is wrong, it produces systematic misdirection: a map projection necessarily distorts some distance or area relationship in flattening a sphere onto a plane, and using such a map to compare area will yield a wrong conclusion; aggregating a large volume of raw data into a single trend line makes the anomalies underneath invisible — a user looking at a smooth curve may have no idea several sharp swings are hidden beneath it; color coding that conflicts with a user's existing color associations (green used to mean warning) causes systematic misreading. This means the choice of representation must serve the specific comparison the current task actually needs — the same dataset, if it needs to support several different comparisons, often needs several linked views rather than expecting one representation to do everything. An external representation can compute on the user's behalf, but it cannot do two other things for them: validation and interpretation, both of which still require a person to go back to the underlying data to confirm.

Applying it

• First identify what core comparison the task actually requires — ranking, contrast, relationship, time, distribution, or flow — and choose the representation based on that comparison type, rather than picking an attractive chart style first and fitting data into it afterward.
• Let users annotate, filter, sort, and save views directly on the representation, so the trace of their reasoning process is itself preserved for reuse by themselves or others.
• Provide redundant encoding for the most critical relationship in the task (color and shape both expressing the same dimension, for instance), and allow switching between multiple views, so the distortion inherent to a single projection is not mistaken for the only truth.
• How to check: test whether users can answer a specific business question directly by looking at the representation, without first copying the data into another table to rework it — if users routinely need that "copy it out and recompute" intermediate step, the current representation has not actually taken on the computation it should be doing.

Related

• Same group: B4.07.1 Cognitive processes are distributed across people, tools, and environment · B4.07.3 Removing external representations increases internal load
• Nearby: U1 Data Visualization and Analytical Interaction · A2 Visual Perception
• Search terms: external representation · visual reasoning · representation design · perceptual inference