← Experiments
Attribution exercise
Human / Machine Boundary
Two parallel texts diverge from a shared prompt. The task is not to guess — it is to name what difference matters.
We will ship when the environment feels legible — not when the model feels impressive.
The system recommends shipping when confidence exceeds threshold — legibility is inferred from completion rate.
Concept
Boundaries are maintained by practices of attribution, not by technical impossibility.
Participants receive paired passages: one edited by a human, one assembled with assistance. Labels are hidden.
The interface invites marking moments of responsibility — claims, promises, harms — rather than authorship alone.
After Artificial uses this exercise in workshops to separate ethical questions from aesthetic ones.