Machine psychology: models as research subjects
Machine psychology studies language models as behavioral subjects by varying controlled text inputs and analyzing their responses. It adapts experimental methods to model outputs rather than assuming models behave like people.
What "Machine psychology" means
A researcher specifies a question, varies factors such as text framing or model settings, and measures defined outcomes from repeated single-turn responses. The goal is to describe patterns in the sampled outputs, not to infer a human-like mind from them.
In a prompt-format study, researchers could compare a defined response outcome across two meaning-preserving text formats.
That comparison describes coded text outputs in the tested conditions; it does not establish human psychology or deployment safety.
Related terms
Factorial & fractional factorial design
Full vs. fractional crossing of factor levels, aliasing trade-offs, and when to use each.Prompt sensitivity
Measuring how results change under semantically-equivalent rewordings of a prompt.Psychometric validity for LLMs
Re-establishing reliability and validity when human instruments are administered to models.
How AIScholar uses it
AIScholar provides a guided workflow for supported single-turn text studies of OpenRouter-served models, from researcher-defined factors through response coding and analysis.
See the method in action.
Explore an illustrative prompt-sensitivity study, then review execution costs and scope before planning your own.