Colby's disenchantment with psychoanalysis would be further expressed in several publications, including his 1958 book, A Skeptical Psychoanalyst. He began to vigorously criticize psychoanalysis for failing to satisfy the most fundamental requirement of a science, that being the generation of reliable data. In his 1983 book, Fundamental Crisis in Psychiatry, he wrote, “Reports of clinical findings are mixtures of facts, fabulations, and fictives so intermingled that one cannot tell where one begins and the other leaves off. …we never know how the reports are connected to the events that actually happened in the treatment sessions, and so they fail to qualify as acceptable scientific data.”.
Likewise, in Cognitive Science and Psychoanalysis, he stated, "In arguing that psychoanalysis is not a science, we shall show that few scholars studying this question get to the bottom of the issue. Instead, they start by accepting, as do psychoanalytic theorists, that the reports of what happens in psychoanalytic treatment -- the primary source of the data -- are factual, and then they lay out their interpretations of the significance of facts for theory. We, on the other hand, question the status of the facts." These issues would shape his approach to psychiatry and guide his research efforts.
Later, Colby would be one of the first to explore the possibilities of computer-assisted psychotherapy. In 1989, with his son Peter Colby, he formed the company Malibu Artificial Intelligence Works to develop and market a natural language version of cognitive behavioral therapy for depression, called Overcoming Depression. Overcoming Depression would go on to be used as a therapeutic learning program by the U.S. Navy and Department of Veteran Affairs and would be distributed to individuals who used it without supervision from a psychiatrist. Needless to say, this practice was challenged by the media. To one journalist Colby replied that the program could be better than human therapists because "After all, the computer doesn't burn out, look down on you or try to have sex with you."
In 1972, at the Stanford Artificial Intelligence Laboratory, Colby built upon the idea of ELIZA to create a natural language program called PARRY that simulated the thinking of a paranoid individual. This thinking entails the consistent misinterpretation of others' motives – others must be up to no good, they must have concealed motives that are dangerous, or their inquiries into certain areas must be deflected - which PARRY achieved via a complex system of assumptions, attributions, and “emotional responses” triggered by shifting weights assigned to verbal inputs.
Shortly after it was introduced, PARRY would go on to create intense discussion and controversy over the possibility or nature of machine intelligence. PARRY was the first program to pass the “Turing Test," named for the British mathematician Alan Turing, who in 1950 suggested that if a computer could successfully impersonate a human by carrying on a typed conversation with a person, it could be called intelligent. PARRY succeeded in passing this test when human interrogators, interacting with the program via remote keyboard, were unable with more than random accuracy to distinguish PARRY from an actual paranoid individual.
As philosopher D.C. Dennett stated in Alan Turing:Life and Legacy of a Great Thinker, "To my knowledge, the only serious and interesting attempt by any program designer to win even a severely modified Turing test has been Kenneth Colby. He had genuine psychiatrists interview PARRY. He did not suggest that they might be talking or typing to a computer; rather he made up some plausible story about why they were communicating with a real live patient via teletype. Then he took the PARRY transcript, inserted it into a group of teletype transcripts and gave them to another group of experts -- more psychiatrists -- and said, 'One of these was a conversation with a computer. Can you figure out which one it was?' They couldn't."
Much of the criticism of ELIZA as a model for artificial intelligence focused on the program's lack of an internal world model that influenced and tracked the conversation. PARRY simulates paranoid behavior by tracking its own internal emotional state on a few different dimensions. To illustrate this, Colby created another program called RANDOM-PARRY which chose responses at random. Responses from RANDOM-PARRY did not model the human patients' responses as well as standard PARRY. Some have argued that PARRY fooled its judges because paranoid behavior makes inappropriate responses or non sequiturs appropriate. But there is still a certain logic to them that PARRY simulates effectively. It is simpler to simulate paranoid behavior, perhaps, but it is not trivial.
The two computer programs, ELIZA and the paranoid PARRY, eventually met for series of "talks," executing different programs. One exchange such went like this:
PARRY: I know the mob controls the big rackets.
ELIZA: Who are you really talking about?
PARRY: Have you ever really thought about the Mafia?
ELIZA: Suppose you ever really thought about the Mafia?
PARRY: Let's talk about something else.
ELIZA: I am not sure I understand you fully.
PARRY: You keep asking me the same things, why not change the topic?
ELIZA: In what way?
Colby would claim that PARRY mimics the natural process by which a person (in this case a paranoid schizophrenic) engages in conversation. The structure of the program that makes the linguistic decisions in PARRY is isomorphic to the 'deep structure' of the mind of the paranoiac. As Colby stated: "Since we do not know the structure of the 'real' simulative processes used by the mind-brain, our posited structure stands as an imagined theoretical analogue, a possible and plausible organization of processes analogous to the unknown processes and serving as an attempt to explain their workings" .
Yet, some critics of PARRY expressed the concern that this computer program does not in actuality "understand" the way a person understands and continued to assert that the idiosyncratic, partial and idiolectic responses from PARRY cover up its limitations. Colby attempted to answer these and other criticisms in a 1974 publication entitled, "Ten Criticisms of Parry."
Colby also raised his own ethical concerns over the application of his work to real life situations. In 1984, he wrote, "With the great amount of attention now being paid by the media to artificial intelligence, it would be naive, shortsighted, and even self-deceptive to think that there will not be public interest in scrutinizing, monitoring, regulating, and even constraining our efforts. What we do can affect people’s lives as they understand them. People are going to ask not only what we are doing but also whether it should be done. Some might feel we are meddling in areas best left alone. We should be prepared to participate in open discussion and debate on such ethical issues.
Still, PARRY has withstood the test of time and for many years has continued to be acknowledged by researchers in computer science for its apparent achievements. In a 1999 review of human-computer conversation, Yorick Wilks and Roberta Catizone from the University of Sheffield comment: "The best performance overall in HMC (Human-machine conversation) has almost certainly been Colby’s PARRY program since its release on the net around 1973. It was robust, never broke down, always had something to say and, because it was intended to model paranoid behaviour, its zanier misunderstandings could always be taken as further evidence of mental disturbance, rather than the processing failures they were."