Two cases can be one metre apart in the world and infinitely far apart for the decision you are trying to make.
Two books sit beside each other on a shelf.
Physically close.
One is a physics textbook.
The other is a novel.
For shelf location, they are neighbours.
For subject matter, they may be far apart.
Now compare the physics textbook with another physics book in a library across Singapore.
Physically far.
Conceptually close.
Distance therefore depends on the space in which we ask the question.
Quick Route
- Feature: what properties or relations are represented?
- Dimension: along which axes can cases differ?
- Distance: how far apart are two cases in that representation?
- Neighbourhood: which cases are locally close?
- Density: how crowded is that local region?
- Cluster: do several cases form a coherent group?
- Outlier: which case sits unusually far from the expected pattern?
Canonical Job
Distance owns one reader job in Cognitive Art:
How far apart are two cases once a representation has decided what differences count?
Distance is not the same as raw difference.
It is difference organised by a representational geometry.
One-sentence answer: Distance is the separation between two represented cases under a chosen set of dimensions, weights and comparison rules.
Distance Is Not Comparison
Comparison asks how two cases are aligned to reveal similarity, difference or relation.
Distance asks what degree of separation that alignment implies.
Comparison can be qualitative.
Distance creates an ordered geometry.
Distance Is Not Dimension
Dimension supplies the axes.
Distance combines differences across those axes.
Two students may differ by:
- speed,
- concept mastery,
- transfer,
- checking.
Distance asks how those differences should be aggregated into a sense of closeness.
Distance Is Not Similarity, Although They Are Closely Related
Many models treat similarity as decreasing as distance increases.
Closer cases are usually treated as more similar.
But human similarity judgments can violate simple geometric assumptions.
One thing may feel more similar to another in one direction than the reverse.
This matters because mathematical distance usually obeys symmetry, while psychological similarity need not.
Amos Tversky’s classic work on the features of similarity showed why human judgments cannot always be reduced to one simple metric geometry.
Cognitive Art therefore uses distance as a powerful representational tool without claiming that every human similarity judgment is literally one metric computation.
Psychological Space
Imagine every case as a point.
Similar cases sit near one another.
Different cases sit farther apart.
This is the intuition behind psychological space.
Roads and Love’s 2024 review, Modeling Similarity and Psychological Space, surveys how representations can be understood geometrically and how the geometry chosen changes the similarity relations a model predicts.
The important public idea is not that the mind contains a literal Cartesian grid.
It is that cognition often behaves as though cases occupy structured relational spaces in which some moves are short and others long.
Physical Distance and Cognitive Distance Can Disagree
Singapore and Kuala Lumpur are geographically closer than Singapore and London.
But for one legal question, Singapore and London may be conceptually closer because of shared institutional history.
For cuisine, Singapore and Kuala Lumpur may be closer.
Distance changes with the represented dimensions.
Weighting Changes Distance
Two schools differ in:
- distance from home,
- curriculum,
- culture,
- cost.
If travel time receives very high weight, the nearby school becomes cognitively close to your decision ideal.
If curriculum fit receives high weight, a farther school may become the closer option.
Distance is not only about dimensions.
It is also about their weights.
Learning Can Change the Geometry
Before learning birds, two species may look almost identical.
After expertise develops, a tiny beak or wing difference becomes decisive.
The physical stimuli barely changed.
The representational distance did.
Category-learning research shows that attention can expand category-relevant dimensions and compress irrelevant ones.
This is one reason expertise changes what appears “obviously different.”
Distance and Generalisation
You learn a response to one case.
How strongly should that learning transfer to another?
Roger Shepard’s influential 1987 paper Toward a Universal Law of Generalization for Psychological Science proposed that generalisation tends to decrease systematically with psychological distance.
The exact universality and geometry of the proposal remain areas of research.
But the central insight is powerful:
what is learned about one case should usually transfer more strongly to cases represented as nearby than to cases represented as far away.
Distance Is Why Transfer Can Fail
A student solves:
x² − 9 = 0.
Then sees:
49a² − 4b².
Structurally close.
Surface appearance differs.
If the learner’s representation weights surface features too strongly, the second problem feels far away.
Transfer fails because the geometry is wrong.
Different Metrics Produce Different Distances
Mathematics offers many ways to define distance.
- straight-line distance,
- city-block distance,
- angular or cosine-style separation,
- edit distance between strings.
Each answers a different structural question.
No metric is automatically correct merely because it is mathematically valid.
The metric must preserve the distinctions relevant to the job.
The Straight-Line Mistake
Two MRT stations may be geographically close.
But if no direct route exists, travel distance can be much larger.
The representation “straight-line geography” answers one question.
The representation “actual transport network” answers another.
Good cognition chooses the distance that matches the action.
Distance Can Be Local
One case may be far from the global average but close to a local subgroup.
A very tall basketball player may be ordinary among professional centres.
Local distance can matter more than global distance.
This is the bridge to neighbourhood.
Distance and Context
The same two events can be close under one question and far under another.
Context selects the active dimensions and their weights.
Context therefore helps determine which geometry is currently meaningful.
Distance and Perspective
A teacher and student can look at the same two answers.
The student sees:
both are wrong.
The teacher sees:
- one is a representation error,
- one is a final arithmetic slip.
The expert perspective contains dimensions the novice perspective lacks.
Those extra dimensions move the answers farther apart diagnostically.
Distance and Resolution
At low resolution:
both students scored 60.
Distance appears zero.
At higher resolution:
- Student A is slow but conceptually strong.
- Student B is fast but conceptually fragile.
The students move apart.
Resolution changes geometry by revealing dimensions that were previously compressed away.
Distance and Scale
Two neighbourhoods can be close within one city and far apart in socioeconomic structure.
Two countries can be far geographically and close institutionally.
Scale controls the extent of the system being considered.
Distance operates inside that chosen extent.
Distance and Cluster
Clusters appear when many cases have small mutual distances relative to surrounding cases.
Change the distance function and the cluster can change.
This is why clustering is never separable from geometry.
Distance and Outlier
An outlier is unusual only relative to some distance structure.
Change dimensions or reference group.
The outlier status can disappear.
Distance creates the possibility of “far away.”
Operating Envelope
Distance is most useful when:
- the represented dimensions have a defensible relationship to the job,
- the units or weights are meaningful,
- the distance model is appropriate to the structure,
- local and global comparisons are not being confused.
Distance becomes dangerous when a convenient mathematical measure is mistaken for the true geometry of the problem.
Distance in Mathematics
Mathematics gives distance precise axioms and many metrics.
The Cognitive Art lesson is broader:
a distance formula is a declaration about which differences count and how they combine.
Even a perfect calculation can answer the wrong question if the metric is wrong.
Distance in English
Words can be close in one semantic dimension and far in another.
“Angry” and “furious” are close in broad meaning.
They differ in intensity and register.
Lexical mastery grows when learners stop asking only whether two words are synonyms and begin asking along which dimensions they are close or far.
Distance in Science
Scientific models constantly decide how to measure separation.
Genetic distance.
Phylogenetic distance.
Chemical similarity.
Phase-space distance.
Different distances preserve different relationships.
Scientific language should always make the metric or comparison frame explicit when conclusions depend on it.
Distance in Education
A strong teacher asks:
How far is this new problem from what the learner already controls?
Too close:
no transfer demand.
Too far:
the learner cannot identify a route.
Instruction can deliberately vary distance:
- same structure, new numbers;
- same structure, new wording;
- same principle, new domain;
- new structure requiring a contrast.
Transfer training is controlled movement through representational distance.
Distance in Organisations
Two projects can look close because both are software launches.
But one serves ten users.
The other serves ten million.
Scale, reliability and failure cost can make the operational distance enormous.
Analogy is safest when the relevant distance is small, not merely when the labels match.
Failure Modes
1. Wrong-Space Distance
Cases are compared using dimensions unrelated to the actual decision.
2. Wrong Weighting
An easy-to-measure dimension dominates a quieter but more consequential one.
3. Physical-Close Equals Cognitive-Close
Spatial proximity is mistaken for conceptual, causal or operational similarity.
4. Metric Reification
A convenient mathematical distance is treated as though it were the one natural geometry of the world.
5. Global-Distance Blindness to Local Structure
A case is judged against the entire population even though its nearest local comparison class is more informative.
Repair Path
- State the job.
- Choose the relevant features.
- Define dimensions.
- Decide how dimensions should be weighted.
- Choose a distance model that matches the structure.
- Check local as well as global distance.
- Test whether conclusions survive reasonable alternative geometries.
The Distance Audit
- Distance in what space?
- Which features define the points?
- Which dimensions define separation?
- How are the dimensions weighted?
- Which metric combines those differences?
- Does the metric preserve the relations relevant to the job?
- Would an expert use the same dimensions as a novice?
- Is local distance more informative than global distance?
- Does the conclusion survive another defensible metric?
- What action changes because these cases are close or far?
A Primary-to-Adult Progression in Distance Thinking
Primary: near and far
Children begin with spatial and simple perceptual distance.
Lower secondary: close in one way, far in another
Students learn that similarity depends on chosen features and dimensions.
Upper secondary: question the metric
Learners compare alternative distance functions, weighting schemes and local versus global frames.
Adulthood: use geometry as an instrument, not an oracle
Professional reasoning asks whether the representational distance improves prediction, transfer or decision—and whether the conclusion survives another reasonable representation.
Research Notes and Further Reading
For a modern synthesis of representational geometry and similarity, see Roads and Love, Modeling Similarity and Psychological Space (Annual Review of Psychology, 2024).
For the classic relationship between psychological distance and generalisation, see Roger Shepard, Toward a Universal Law of Generalization for Psychological Science (Science, 1987). Later work continues to test how broadly such distance-based generalisation applies, including high-dimensional naturalistic stimuli.
For an important boundary showing why psychological similarity cannot always be treated as a simple symmetric metric, see Amos Tversky, Features of Similarity.
World Return
Distance earns its place only if it survives contact with outcome.
If two cases judged “close” repeatedly require different actions, the geometry needs revision.
If two cases judged “far” consistently transfer knowledge well, the representation may be overweighting irrelevant differences.
Reality sends the distance model back for calibration.
Final Thought: Distance Is a Property of the Map as Much as the Territory
The two books remain beside each other.
The library shelf says close.
The subject map says far.
A reader’s purpose decides which distance matters.
Before asking how far apart two things are, ask which map made “far” meaningful.