PARRY was a conversational program built by the psychiatrist Kenneth Colby at the Stanford Artificial Intelligence Laboratory and demonstrated around 1972. Where Joseph Weizenbaum’s ELIZA had imitated a non-directive therapist by reflecting the user’s statements back, PARRY went the other way: it simulated the patient, modeling the thinking of a paranoid individual through a system of internal variables for fear, anger, and mistrust that shifted in response to what the user typed. Colby’s underlying model had been published in the journal Artificial Intelligence in 1971.
PARRY held opinions and an emotional state, so it could steer conversations toward its fixed preoccupations - racketeers, the mob, and a fear of being harmed. Because paranoid speech is evasive and suspicious by nature, the program’s deflections and non-sequiturs read as in-character rather than as failures of comprehension, which made it unusually convincing for its era.
Colby ran a version of the Turing test on it. Transcripts of PARRY interviews were mixed with transcripts of real paranoid patients and shown to practicing psychiatrists, who were unable to distinguish the program from the humans at better than chance. That result is one of the earliest documented cases of a machine passing a controlled imitation test in a specialized domain.
In a now-famous episode, PARRY was connected to ELIZA’s Doctor program over the ARPANET, and the resulting dialogue between the two chatbots was preserved in 1973 as Internet RFC 439, “PARRY Encounters the Doctor,” written by Vinton Cerf.
Why business readers should care: PARRY showed that a chatbot becomes more believable when it has a consistent persona and point of view, not just good grammar - the same insight that drives character and system-prompt design in today’s conversational AI.