Agent-type contrasts are isolated by random assignment to identical targets.

Evidence

Participants were randomly assigned to one of the three agent conditions (harmful; neutral; benevolent). In all three conditions, participants first read some filler information about the target: “John is a 29 year old man with brown hair and brown eyes. He works as a taxi driver in Miami.” — Mansur Khamitov, Jeff D. Rotman, and Jared Piazza, Perceiving the Agency of Harmful Agents, 293