Thesis: Believable Suspect Agents
The social skills necessary to properly and successfully conduct a police interrogation can and need to be trained. In my PhD project, I work toward a virtual character that can play the role of a suspect in a practise police interrogation. The student can interrogate the virtual suspect using natural interaction. Virtual humans in social skill learning offer learning by experience; the student can experience a social interaction.
To make a compelling virtual suspect that can do the above, we need to know how to relate the behaviour of our virtual suspect to the behaviour of the trainee in a way that is consistent with the personality of the persona the virtual suspect is playing. The virtual agent needs to model the dynamics of such interpersonal relations. Ideally, the agent can analyse the speech and non-verbal messages of the trainee to determine his or her level of friendliness or aggression and use these interpretations to update their interpersonal relationship values. The response of the virtual suspect is then based on the interpersonal status of the interrogator (e.g. if you make him angry…) and the suspect (…he will respond angrily). The virtual suspect’s response has to be comparable to, or can pass for, “human suspect behaviour” in such a way that the human interactant is capable of understanding the virtual suspect’s interpersonal status otherwise the user cannot learn interpersonal skills from the interaction. In my thesis, I will describe our efforts towards these goals.
My thesis titled Believable Suspect Agents: response and interpersonal style selection for an artificial suspect can be downloaded here.
See also this fancy video by the COMMIT/ IUALL project: