The influence of robot autonomy on sense of control and trust towards a robot

We conducted four online experiments in which we investigated how different types of autonomy influence sense of control and trust. Participants played a game in which they selected which box should be collected by a robot avatar. Each box contained a certain number of points. Participants were told that some robots have artificial intelligence and that these robots might collect different boxes than selected by the participants. At the end of the experiment, participants played a trust game with each robot. We compared a situation in which a robot disobeys participants’ commands in order to benefit the participant by obtaining more points (helpful autonomy) versus when it disobeys, but acquires the same payoff (same-outcome autonomy). Moreover, in Experiments 3 and 4 we compared two types of helpful autonomy: autonomy to avoid loss and autonomy to obtain extra gain. We found that helpful autonomy led to higher trust than same-outcome autonomy, except when it helped to avoid loss, but did so unreliably (Experiment 4). Second, trust was unrelated to the perceived sense of control over autonomous robots. Finally, participants reported experiencing more sense of control over robots with same-outcome than helpful autonomy in Experiments 1-2, where the behavior of autonomous robots could be compared with that of robots that had no autonomy or behaved randomly. However, this relation was reversed in Experiments 3-4, where these control conditions were absent, showing that sense of control can be modulated by robot's perceived autonomy, but also exposure to different types of other robots’ behavior.