When you use a metaphor, you're not saying two things are the same. You're saying: treat this unfamiliar thing as if it implements the interface of this familiar thing. That's duck typing applied to cognition.
The claim
Metaphor is duck typing for cognition. Not identity assertion. Behavioural compatibility assertion.
Duck typing in code says: don't check the type declaration, check the behavioural interface. If it has the methods you need, use it. The formal type is irrelevant to whether you can reason about it effectively. Metaphor works identically — but the "methods" are causal patterns and the "caller" is human reasoning.
The Examples Under the Microscope
Take five common metaphors and apply the duck test:
| Metaphor | Interface it borrows | Where it quacks wrong |
|---|---|---|
| "Memory is a filing cabinet" | Organised, retrievable, persistent storage | Files don't change when you re-read them. Memory reconstructs each time. The filing cabinet doesn't hallucinate. |
| "The brain is a prediction engine" | Hypothesis generation, error correction, forward modelling | Engines are built to spec. Brains are grown under pressure. The metaphor survives most neuroscience — a good one. |
| "Organisations are organisms" | Adaptation, metabolism, lifecycle, immune response | Organisms die when the environment changes faster than they can adapt. So do organisations. This metaphor is unsettlingly accurate. |
| "Networks are highways" | Throughput, congestion, routing, on-ramps | Packets don't slow down because of trucks. Bandwidth isn't lanes — it's frequency. Breaks down under latency reasoning. |
| "LLMs are autocomplete on steroids" | Pattern continuation, statistical next-token prediction | Autocomplete doesn't reason through novel problems. Calls out the limits clearly. A deliberately partial interface. |
None of these metaphors are literally true. All of them are interface contracts for reasoning. The question isn't whether they're accurate — it's whether they're causally preserving. A metaphor that preserves the causal structure of the target domain lets you reason forward and arrive at correct conclusions. One that doesn't gives you confident wrong answers.
The Quality Criterion: Does it Preserve Causality?
Good metaphors preserve the causal structure of the thing they describe. "The brain as prediction engine" survives most questions about perception, attention, and learning — because predictive processing genuinely does map onto those phenomena. "Memory as filing cabinet" breaks the moment you ask about false memories, emotional salience, or reconstructive recall — because files don't do any of those things.
Good Metaphor
Supports forward reasoning
Causal chains still hold under "but what if…"
Breaks gracefully at known limits
Transfers explanatory power
Gets more useful the deeper you go
Bad Metaphor
Sounds plausible, breaks under a second question
Causal chain diverges from reality
Fails at exactly the decision points that matter
Produces confident wrong conclusions
Gets less useful the deeper you go
Why Some Explanations Collapse Under Pressure
When an explanation collapses under a deeper question, it almost always means the underlying metaphor was already wrong — just not wrong enough to notice until someone pushed on the causal chain. The explanation was valid at the surface behavioural level. It just didn't preserve causality one level down.
Bad metaphors don't quack wrong randomly. They quack wrong at the exact points where the causal structure of the source domain diverges from the target. The failure location is the map of the gap.
The Duck Button Is This Principle Made Literal
The 🦆 button on this site is this principle made literal. It doesn't change what anything is. It changes the behavioural description to use duck-shaped mental models you already have. It's not simplification — it's interface translation. The internal implementation stays exactly the same. The causal surface you reason against changes.
And this is what good communication has always been. Not dumbing down. Not oversimplifying. Finding the correct behavioural interface that already exists in the listener's mind — and mapping the new thing onto it carefully enough that the causal chain still holds when they push on it.
Metaphor is duck typing for cognition. The test is not accuracy — it's whether the interface holds under reasoning.