The Robotics Inheritance
Masahiro Mori observed in 1970 that human comfort with humanoid robots does not increase linearly with realism. As a robot becomes more lifelike, comfort increases — until a threshold is crossed. At a point just below full human likeness, the slight imperfections cause a sharp drop in comfort: the familiar-but-wrong triggers an acute sense of wrongness more profound than a clearly mechanical robot would. The valley of discomfort lies between "clearly artificial" and "fully human."
AI-generated prose currently occupies this valley. It has long since crossed the threshold of clearly artificial — it is grammatically indistinguishable from human writing at the sentence level. But it has not crossed the threshold of fully human in the holistic sense: a trained reader, reading a passage of any length, experiences a background unease that resolves into "this was not written by a person with a life."
The Tells
The uncanny quality of valley prose is difficult to reduce to a rule because it emerges from the aggregate rather than any single element. Common indicators include:
- Uniform hedging density: The same epistemic qualifiers distributed evenly across all claims, regardless of the actual certainty of the claim.
- Generic transitions: "Furthermore," "It is worth noting," "In conclusion" — the connective tissue of a system that has learned transition patterns but not what to transition to.
- Symmetric paragraph structure: Each paragraph doing the same amount of work, making the same number of moves. Human writing is uneven.
- Absence of the irrelevant: A human writer's mind wanders productively — a digression, a personal association, a piece of detail that does not quite fit the argument but is true and specific. Valley prose has no wandering mind behind it.
- The missing scar: No indication of where the writer has been hurt, confused, or surprised by the subject.
Usage in context: "It reads too smooth — it's in the Uncanny Valley. Find the scar. Where did this actually come from a person?"