PARRY, a model of paranoia
In spring 1971 Kenneth Colby, Sylvia Weber and Franklin Hilf described PARRY, a program simulating a paranoid patient in a psychiatric interview over teletype. They proposed to judge the model by indistinguishability tests, and in 1972 psychiatrists acting as judges tried to tell interviews with it from interviews with patients.
Why it matters
A conversational program was tested by a variant of the Turing test with experts as judges: psychiatrists rated how paranoid the interviews were and tried to tell the machine from the patient. The authors concluded to 'a satisfactory degree of resemblance' between the two groups of interviews.
The full texts of both papers are paid and their abstracts were read; they carry no figures, so the record does not claim how many psychiatrists took part, how many interviews they were given or how often they guessed right. On 18 September 1972 PARRY, running at the Stanford AI Laboratory (SAIL), talked with DOCTOR on a BBN Tenex machine, both reached from a terminal at UCLA; Vinton Cerf published the session on 21 January 1973 as RFC 439. The RFC does not use the word ARPANET. At the first Loebner contest in 1991 Colby's program on relationship problems was rated human twice.