How, and to what extent, generic human-like behavioural characteristics can trigger adopting the intentional stance to the robot? The first step is to quantitatively analyse the parameters of human reactions to distracting stimuli in a naturalistic environment. As a second step, those behavioural characteristics will be copied on a robot, and participants will observe either a human-like robot reactive to the environment, or a mechanistic behaviour.
In this study, we examined if humans adapt their performance to the robot's actions. Firstly, participants were asked to create and teach a sequence of musical tones to the iCub robot. The robot repeated the sequence with a decreasing delay between its own taps and taps performed by the participants. Then, we asked participants to play a duet with iCub. In the duet phase, the robot made an error in 60% of the trials, in order to violate participantsí expectations. Across participants, we manipulated the type of mistakes. For half of the participants, iCub pressed a wrong key (Human-like error), whereas, for the other half, iCub interrupted the melody and moved back and forth between two keys in a "loop" (Mechanical error).