Distortion
Distortion (失真) is a scale-class criterion — when a model is moved into a scale or reference frame it was not made for, it is distorted.
Field definition
A model projected onto a scale or reference frame to which it does not apply.
The key: not false. The model itself may be entirely correct. The position is wrong — the model is being used where it does not apply.
Three-layer distinction
| State | Definition | Everyday example |
|---|---|---|
| Valid | The model is used within its proper domain | A textbook's explanation of dry ice applied to "dry ice in air" |
| Distorted | The model is used outside its applicable conditions | A textbook's explanation of dry ice applied to "dry ice in water" |
| False | The model does not hold under any reasonable condition | "The earth is flat" |
Key distinction: "It is distorted" = "the concept is not wrong, but it has been used in the wrong place" — more precise than "wrong", gentler than "incorrect".
High-leverage distortion
When a valid model is used across domains by a high-leverage system (textbooks, media, the state, AI), its distortion is amplified and shows up as "errors that look correct".
This is the form of distortion that demands extra vigilance in the AI era: a correct model, carried by a system that amplifies it, placed in a context where it does not apply, with output that looks reasonable — but is being applied where it should not have been.
Invalidation clauses
- If used to evade calibration ("it's not wrong, only distorted") → invalid. Distortion needs to be corrected, not tolerated
- If used to dismiss another's observation ("you are just distorted") → invalid. Distortion describes the relationship between model and scale, not between one person and another
Why it matters
Many situations that look like "different positions", "cognitive conflict", or even "the other side is wrong" are actually distortion — both sides hold correct models, but they are applied at different scales, so each feels the other is talking nonsense.
This term offers a more precise position: not "who is right and who is wrong", but "does this model apply at this scale?"