Not as bad as it seems: When the presence of a threatening humanoid robot improves human performance

See allHide authors and affiliations

Science Robotics  15 Aug 2018:
Vol. 3, Issue 21, eaat5843
DOI: 10.1126/scirobotics.aat5843


“Bad” humanoid robots just paying attention to human performance may energize attentional control—as does human presence.

Millions of people worldwide may soon benefit from the presence of humanoid robots designed to ensure support to the elderly, disabled people, or pupils with learning difficulties (1). Despite this accelerating trend, little is known about the emotional experience associated with human-robot interaction (HRI) and its impact on human cognition. Because this is a critical issue for the introduction of humanoid robots in our societies (2), we examined whether (i) socially interactive humanoid robots affect attentional control (i.e., the paramount cognitive ability) and (ii) this impact depends on the emotional valence associated with HRI. To do so, we used the gold standard of attentional measures, the Stroop task (3), requiring individuals to identify the color in which a word is printed, ignoring the word itself. Because of the automaticity of reading, identification times are consistently longer for color-incongruent words (the word “BLUE” in green ink) than for color-neutral items (“DESK” in green ink). The amplitude of this well-known effect, called “Stroop interference,” indicates the efficiency of cognitive-attentional control. It typically decreases under stress (4), especially in the presence of others competing with—or simply paying attention to—our current performance (57). However, whether and when the presence of social humanoid robots also boosts attentional control remains unanswered. We predicted that the presence of robots simply paying attention to human performance may energize attentional control—as human presence does—especially when these robots are thought to be likely to produce negative evaluations (8).

To test this hypothesis, we asked young adults to perform the standard Stroop task twice. In session 1, all participants performed the task alone. In session 2, they performed the task either alone or in the presence of a humanoid robot with which they had previously interacted either positively (a “good” robot responding in a nice way, with empathy) or negatively (a “bad” robot responding with contempt, lack of empathy, and negative evaluations about participants’ intelligence) (see the Supplementary Materials). In the two gestures and speech based on a strictly identical script (e.g., head movement toward the participant 60% of the time, light arm movements) (Fig. 1A). At the end of session 2, participants in the two robotic conditions rated the robot present on various personality traits (see the Supplementary Materials), either positive (e.g., warm, competent) or negative (awkward, aggressive).

Fig. 1 Experimental setup and participant performance.

(A) We used a MeccanoidG15KS animated at a distance by a human operator using two smartphones to control the robot’s gestures and speech. In the two presence conditions, the robot was positioned in front of participants (to their right on the edge of their peripheral vision) and watched them 60% of the time by turning the head according to a pre-established script. (B) The main effect of condition on Stroop performance improvement (error bars represent 1 SE) indicates that the positive interaction condition did not differ from the control condition, whereas the negative HRI condition differed from the positive HRI and control conditions averaged (see the Supplementary Materials for detailed statistical analyses).

Not surprisingly, the bad robot was rated as less warm, friendly, and pleasant than the good robot. Participants also attributed fewer human nature traits (e.g., “cognitive openness”) and more mechanical dehumanization traits (e.g., “rigidity”) to the bad robot compared with the good robot (see the Supplementary Materials). More importantly, individuals’ attentional control improved notably in the presence of the bad robot. Planned comparisons were used to analyze relevant between-group contrasts (alone versus pleasant robot; alone and pleasant social robot averaged versus unpleasant social robot) on Stroop interference [response times (RTs) for color-incongruent words minus RTs for color-neutral items] at session 2 minus interference at session 1 (baseline). A positive value (see Fig. 1B) means reduced interference (improved performance) at session 2 relative to baseline. As expected, Stroop performance improved exclusively in the presence of the unpleasant robot. This condition differed from the two other conditions averaged (alone and pleasant social robot), which did not differ from one another (see the Supplementary Materials for detailed statistical analyses).

These findings run counter to a purely mechanistic approach that reduces the effects of robotic presence to physical action or noise distraction, which may facilitate or inhibit performance depending on task difficulty (9). According to this approach, both robotic presence conditions—regardless of their emotional tone (positive or negative)—should have resulted in a performance change compared with isolation (all the more so because the robot’s appearance and behavior during task performance were identical in both conditions). Instead, Stroop performance changed exclusively in the bad social robot condition.

Perhaps even more striking, the bad social robot had the same impact on Stroop performance as in earlier research with human presence (57). This presence reduced, rather than increased, Stroop interference, which extends the relevance of the attentional view of social facilitation from humans to social robots. According to this view (5, 6), the presence of potentially threatening others improves the selectivity of attention to relevant information at the expense of competing cues (in the Stroop task, the color in which a word is printed at the expense of the word itself). This is what happened in the bad social robot condition. Therefore, not only can the behavior of robots change humans’ perception of robots during HRI (10), but these attributions are susceptible to making the simple presence of robots likely to affect human cognition as a function of the interaction type. Thus, the present findings constitute evidence that the presence of social robots may energize attentional control, especially when the emotional valence and anthropomorphic inferences associated with the robot being present require a heightened state of alertness.




Fig. S1. The experimental installation.

Table S1. Verbal exchange script.

Table S2. Mean correct response times (in milliseconds), SDs (in parentheses), and error rates as a function of the type of stimuli, session, and group.

References (1116)


Acknowledgments: This study was carried out in accordance with the provisions of the World Medical Association Declaration of Helsinki. Funding: This work was supported by a grant (Social_Robot_2017-2018) from the Maison des Sciences de l’Homme, Clermont-Ferrand, France. Data and materials availability: All data are publicly available via the Open Science Framework and can be accessed at

Stay Connected to Science Robotics

Navigate This Article