Expectations and attribution of mental capabilities to artificial agents

Key research questions

1. Do expectations about intelligent robots influence the type of descriptions people use for robot behavior?

2. Which individual factors modulate this effect?

METHODOLOGY

We devised a new paradigm to evaluate whether more hesitant behavior was associated with intentional motivation of actions. We measure individual characteristics, expectations, and descriptions of behavior using questionnaires before and after the experiment. Descriptions of robot behavior were measured with the Intentional Stance Questionnaire (Marchesi et al., 2019).

Group 1 observed the robot behaving decisively (80% hesitant, 20% decisive) while Group 2 observed the inverted proportion of behavior (80% decisive, 20% hesitant).

ANALYSES

We focused on understanding whether the exposure of the behavior and the expectations modulated the use of intentional descriptions of behavior. Correlation analyses were performed to unveil the links between questionnaires and observed behavior.

RESULTS

After observing the robot, participants from Group B (mostly decisive) tended to describe the behavior of the robot as intentional. Interestingly, independent of the group, participants with higher expectations used more intentional descriptions of behavior. Other questionnaires did not reveal any differences between the groups.

We capture the attention of the participants with iCub's behavior. It was a big step to understand what can fascinate people and engage them in a task. Expectations play a crucial role while interacting with humanoid artificial agents, we should always include them in the paradigms and analyses.

My contribution was designing the paradigm, analyzing the data, and writing the paper. This research was a foundation for new studies and more interactive paradigms.

Our work was accepted and presented at an international conference.

Reference: S. Marchesi, J. Perez-Osorio, D. De Tommaso and A. Wykowska (2020) Don’t overthink: fast decision making combined with behavior variability perceived as more human-like. In Proceedings 29th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN 2020), pp. 54-59, doi: 10.1109/RO-MAN47096.2020.9223522.