Gestural Turing Test. A Motion-Capture Experiment for Exploring Believability In Artificial Nonverbal Communication.

One of the open problems in creating believable characters in computer games and collaborative virtual environments is simulating adaptive human-like motion. Classical artificial intelligence (AI) research places an emphasis on verbal language. In response to the limitations of classical AI, many researchers have turned their attention to embodied communication and situated intelligence. Inspired by Gestural Theory, which claims that speech emerged from visual, bodily gestures in primates, we implemented a variation of the Turing Test, using motion instead of text for messaging between agents. In doing this, we attempt to understand the qualities of motion that seem human-like to people. We designed two gestural AI algorithms that simulate or mimic communicative human motion using the positions of the head and the hands to determine three moving points as the signal. To run experiments, we implemented a networked-based architecture for a Vicon motion capture studio. Subjects were shown both artificial and human gestures, and were told to declare whether it was real or fake. Techniques such as simple gesture imitation were found to increase believability. While we require many such experiments to understand the perception of humanness in movement, we believe this research is essential to developing a truly believable character.