Structure-Mapping Analogy

Thinking Tools4.2 Expansive Idea FormationAlso known as Analogical Ideation, Structure-Mapping

From the structure-mapping theory of Dedre Gentner (Structure-Mapping: A Theoretical Framework for Analogy, 1983), extended by Keith Holyoak and Paul Thagard (Mental Leaps, 1995) and by Douglas Hofstadter and Emmanuel Sander (Surfaces and Essences, 2013).

Borrows the pattern of relationships from a well-understood field and maps it onto a problem, generating ideas and showing where the analogy breaks.

An immune system mapped onto a bank's fraud team, one relation at a time

Each tag pairs something from the source, the immune system, with its candidate in the target, card fraud. Select one, then the column that judges the mapping: a relation that carries over gives an idea, an object-for-object match gives nothing, and a relation that fails shows where fraud works differently. The first is placed as an example. Illustrative.

Items (select one, then a box)
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break
  • Not placedPlacedTake the ideaDrop itStudy the break

Immune system to fraud

Mappings
Take the idea
Drop it
Study the break
Holds as a relationMatches objects onlyBreaks

Carried over to the fraud team, the mapping

Matches known signatures. Immune cells match the marks of invaders seen before; fraud rules match patterns of known schemes.Not placed yet.Matches known signatures. Immune cells match the marks of invaders seen before; fraud rules match patterns of known schemes.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Holds. The idea: share new fraud patterns across the card network fast, the way a vaccine spreads a signature ahead of the infection.Matches known signatures. Immune cells match the marks of invaders seen before; fraud rules match patterns of known schemes.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Matches known signatures. Immune cells match the marks of invaders seen before; fraud rules match patterns of known schemes.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Remembers past invaders. A second infection meets a faster response; a fraud ring often returns with the same devices.Not placed yet.Remembers past invaders. A second infection meets a faster response; a fraud ring often returns with the same devices.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Holds. The idea: a memory list that blocks a known fraud device the first time it reappears.Remembers past invaders. A second infection meets a faster response; a fraud ring often returns with the same devices.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Remembers past invaders. A second infection meets a faster response; a fraud ring often returns with the same devices.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Spares its own cells. Healthy immunity leaves the body's own cells alone; declining honest customers is the bank attacking itself.Not placed yet.Spares its own cells. Healthy immunity leaves the body's own cells alone; declining honest customers is the bank attacking itself.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Holds, and it names the false decline problem. The idea: model each customer's own normal spending so their habits stop looking foreign.Spares its own cells. Healthy immunity leaves the body's own cells alone; declining honest customers is the bank attacking itself.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Spares its own cells. Healthy immunity leaves the body's own cells alone; declining honest customers is the bank attacking itself.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Scales its response. A small infection meets a small response; a card either works or is blocked outright.Not placed yet.Scales its response. A small infection meets a small response; a card either works or is blocked outright.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Holds as a gap to close. The idea: stepped responses, a text check first, a hold next, a block last.Scales its response. A small infection meets a small response; a card either works or is blocked outright.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Scales its response. A small infection meets a small response; a card either works or is blocked outright.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Analysts are white cells. Both are the ones who go after the intruder.Not placed yet.Analysts are white cells. Both are the ones who go after the intruder.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Analysts are white cells. Both are the ones who go after the intruder.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Objects matched by role, with nothing about how either works. It feels apt and suggests nothing to try.Analysts are white cells. Both are the ones who go after the intruder.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Both are defenses. The immune system defends the body; the fraud team defends the bank.Not placed yet.Both are defenses. The immune system defends the body; the fraud team defends the bank.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Both are defenses. The immune system defends the body; the fraud team defends the bank.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.A shared label with no shared mechanism behind it: it is true of every defense and tells the team nothing new.Both are defenses. The immune system defends the body; the fraud team defends the bank.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Germs mutate blindly. Germs change by chance; fraud rings watch what the bank blocks and change on purpose.Not placed yet.Germs mutate blindly. Germs change by chance; fraud rings watch what the bank blocks and change on purpose.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Germs mutate blindly. Germs change by chance; fraud rings watch what the bank blocks and change on purpose.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Germs mutate blindly. Germs change by chance; fraud rings watch what the bank blocks and change on purpose.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Breaks, and the break is the finding: an adversary that learns can map fixed thresholds, so thresholds should vary in ways it cannot learn.Germs seek no profit. An infection spreads wherever it can; fraud goes where the payoff per attempt is highest.Not placed yet.Germs seek no profit. An infection spreads wherever it can; fraud goes where the payoff per attempt is highest.Take the idea: The relation works the same way in the target, so what the source does suggests what the target should do.Germs seek no profit. An infection spreads wherever it can; fraud goes where the payoff per attempt is highest.Drop it: It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.Germs seek no profit. An infection spreads wherever it can; fraud goes where the payoff per attempt is highest.Study the break: The relation fails in the target, which marks a place where the target has structure the source lacks.Breaks: fraud follows money. Raising the cost of the cheapest route moves more fraud than tightening every rule a little.

Your sort matched 0 of 8.Your sort matched 1 of 8.Your sort matched 2 of 8.Your sort matched 3 of 8.Your sort matched 4 of 8.Your sort matched 5 of 8.Your sort matched 6 of 8.Your sort matched 7 of 8.Your sort matched 8 of 8.

Take the idea
The relation works the same way in the target, so what the source does suggests what the target should do.
Drop it
It pairs one thing with another by resemblance, with no shared working behind it, so it predicts nothing.
Study the break
The relation fails in the target, which marks a place where the target has structure the source lacks.

The sort

  • Matches known signatures: Take the idea. Holds. The idea: share new fraud patterns across the card network fast, the way a vaccine spreads a signature ahead of the infection.
  • Remembers past invaders: Take the idea. Holds. The idea: a memory list that blocks a known fraud device the first time it reappears.
  • Spares its own cells: Take the idea. Holds, and it names the false decline problem. The idea: model each customer's own normal spending so their habits stop looking foreign.
  • Scales its response: Take the idea. Holds as a gap to close. The idea: stepped responses, a text check first, a hold next, a block last.
  • Analysts are white cells: Drop it. Objects matched by role, with nothing about how either works. It feels apt and suggests nothing to try.
  • Both are defenses: Drop it. A shared label with no shared mechanism behind it: it is true of every defense and tells the team nothing new.
  • Germs mutate blindly: Study the break. Breaks, and the break is the finding: an adversary that learns can map fixed thresholds, so thresholds should vary in ways it cannot learn.
  • Germs seek no profit: Study the break. Breaks: fraud follows money. Raising the cost of the cheapest route moves more fraud than tightening every rule a little.

An example

The card fraud team at a regional bank had tuned its rules for two years, and every round traded one problem for another: tighter rules caught more fraud and declined more honest customers. The head of fraud ran a session that borrowed from the immune system.

The group listed what an immune system does, as verbs: it matches known signatures, remembers past invaders and answers faster the second time, spares the body's own cells, and scales its response to the threat. Then they tested each relation on the fraud operation. Four held; matching signatures was what the rules already did, and three gave ideas: a model of each customer's own normal spending, so their usual purchases stop looking foreign; a memory list that blocks a known fraud device on sight; and stepped responses, a text check before any block. One relation broke: germs mutate blindly, while fraud rings watch what the bank blocks and adapt on purpose. That break became a fourth idea: thresholds that vary so they cannot be learned. False declines fell by a third the next quarter.

Why it works

Two situations can look alike and share nothing useful, or look nothing alike and share the same pattern of causes. An analogy pays off when it carries a connected system of relations from a field people understand in depth to one they do not: detects, remembers, spares, scales. Each relation that holds in the target brings with it everything people already know about how that relation behaves, which is where the new ideas come from.

Writing the relations as verbs keeps the mapping honest. Calling analysts white cells feels clever and predicts nothing, because it maps an object; asking whether the fraud operation spares its own customers maps a relation that can be checked. A relation that fails to carry over marks a place where the target works differently from the source, and naming that difference is often the sharpest finding of the session.

The thinking behind it

Dedre Gentner Dedre Gentner, a cognitive psychologist at Northwestern University, set out structure-mapping theory in a 1983 paper in Cognitive Science. An analogy, on her account, maps a system of relations from a base domain onto a target, and people prefer mappings that keep connected higher-order relations intact over mappings that rest on shared surface features. The tool's rule of listing verbs and leaving the objects aside comes straight from that distinction.

Keith Holyoak Keith Holyoak, a UCLA cognitive scientist who studies analogy in reasoning, wrote Mental Leaps: Analogy in Creative Thought (1995) with the philosopher Paul Thagard. They hold that analogy is guided by three constraints at once, similarity, structure, and purpose, and they trace analogy at work in science, politics, and everyday reasoning. The purpose constraint is why a mapping session starts from the target problem the group wants ideas about.

Douglas Hofstadter Douglas Hofstadter, the cognitive scientist who wrote Gödel, Escher, Bach, argued with the psychologist Emmanuel Sander in Surfaces and Essences (2013, first published in French) that analogy making, the mapping of one situation onto another, is the core of concepts and of all thought. On their view the borrowing a mapping session makes deliberately is what ordinary thinking does all the time, mostly without noticing.

How to run it

  1. State the target problem. Name what you want new ideas about, such as cutting false declines without letting more fraud through.
  2. Pick a source the group knows well. Choose a field with rich, well-understood workings, such as the immune system, a kitchen, or traffic.
  3. List the source's relations as verbs. Write what the parts do to each other, such as detects, remembers, spares, and scales, and leave the objects aside.
  4. Map each relation onto the target. Ask what in the target does the same thing, and write the candidate beside each verb.
  5. Mark each mapping holds or breaks. Check each against how the target actually works, with someone who knows it in detail.
  6. Turn each hit into an idea. Ask what the source's way of doing it suggests the target should do.
  7. Study each break. Write what the target has that the source lacks, and what that difference calls for.

Where it goes wrong

The mapping runs on objects. Analysts are white cells, alerts are fever, and the session produces a vivid picture and no testable idea. Spot it when the table's left column is nouns; rewrite each entry as what the thing does.

The group falls for the first source and maps everything onto it, explaining away each mismatch. Mark every relation holds or breaks before anyone proposes an idea, and try a second source if most break.

A hit is treated as proof. That the immune system remembers does not show the bank should block on first sight; it suggests a test. Run the idea on past data before it goes live.

When to use something else

  • Synectics. When the group needs to get far from the problem's own terms, with personal and fantasy analogies as well as direct ones.
  • Biomimicry. When the source should be how living things solve a specific design problem in nature.
  • First Principles Thinking. When the problem needs rebuilding from what is actually required, with no borrowed picture at all.

Where it sits in SOLVE eX

Go to the source

  • Structure-Mapping: A Theoretical Framework for Analogy, Cognitive Science, 1983
  • Mental Leaps: Analogy in Creative Thought, 1995
  • Surfaces and Essences: Analogy as the Fuel and Fire of Thinking, 2013

Back to the catalog