Humans do not "play fair" with social robots: Unequal task sharing in an attentional task

Using artificial agents, such as ChatGPT, to support everyday tasks has become commonplace, yet little research has examined offloading of attentionally demanding tasks to artificial agents. Prior work investigating how humans offload attentional demands to algorithms and other humans found that people prefer an equal split of attentional demands when interacting with humans but not when interacting with algorithms. This raises the question of how offloading decisions are made when interacting with algorithms embodied in social robots with human-like characteristics: Would they treat them like a human or like an algorithm? To investigate this question, participants performed an attentionally demanding task (i.e., a multiple object tracking task, MOT), which they could (partially or fully) offload to the social robot MAKI, whose task accuracy was either known (Experiment 1) or unknown (Experiment 2). In both experiments participants offloaded a significant number of the attentional demands to the robot, which improved the participants' own tracking accuracy. Knowing the robot's tracking accuracy, however, did not impact offloading behavior or performance gains as participants perceived the robot as highly capable of performing the MOT task even without explicit information about its accuracy (see Experiment 1). Importantly, participants did not split attentional demands equally with the robot (like they do with human interaction partners), matching offloading patterns from studies when humans offloaded attentional demands to algorithms. In conclusion, our results indicate that when it comes to offloading behavior, a social robot is treated similarly to an algorithm rather than a human interaction partner.