IEEE Robotics & Automation Magazine - September 2020 - 80

a1 =

1.19∗

Experimental
Condition

Warmth
Attribution

c ′ = -19.63
(c = -29.85)∗

b1 = -8.56∗

Selective
Attention
Performance

Figure 5. A diagram showing mediation of the robot-presence
effect on standard selective attention performance by warmth
inferences. *: p < 0.05.

those in the nonsocial-robot condition. After the social inter-
action, the more warmth they perceived the robot possessing,
the greater was the performance improvement generated by
the presence of the robot [b = -0.22, 95%  CI (-0.50;
-0.01)] (Figure 5).
The Best Predictor of Performance Improvement
We conducted a regression analysis on selective attention per-
formance in the second session, including the scores on the
three anthropomorphism dimensions and the four human-
ization dimensions. This analysis makes it possible to evaluate
the unitary effect of each variable to extract the best predictor,
while controlling for collinearity. When all of the other
dimensions were controlled for, the results showed that only
the attribution of human-uniqueness traits had a signifi-
cant predictive effect on selective attention performance
[b = -0.63, t (45) = -2.80, p = 0.008, h 2p = 0.17, 95% CI
(-25.29; - 4.07)] . The more uniquely human traits the par-
ticipants considered the robot to possess, the better they per-
formed in the second session. We also found a statistical trend
toward a significant effect of discomfort attribution: the more
the participants described the robot as having discomfort
traits (e.g., awkward), the lower the interference due to incon-
gruence in the EFT during the second session [b = -0.25,
t (45) = -1.75, p = 0.089, h 2p = 0.07, 95% CI (-23.75; 1.75)].
These results were confirmed by the mediation comparison
analysis, including all dimensions as independent mediators
in a global mediation model [R 2 = 0.47, F (2, 43) = 4.17,
p 1 0.01]. Again, only mediation by human uniqueness was
significant [b = -37, 95% CI (-0.50; -0.01)], thereby vali-
dating the third hypothesis concerning the modulation of the
robot-human conceptual distance as a mediator of SFI effects.
Discussion
In the future, robots are likely to be more common in the
everyday human environment. It seems crucial to understand
today how their presence may affect human cognitive pro-
cesses. The aim of this study was to evaluate how the passive
presence of a robot affects human cognitive performance.
First, we demonstrated that social interaction changed
how participants considered the robot in terms of humanuniqueness (e.g., cognitive openness), human-nature, and
warmth traits compared to a simple description. This result
80

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

SEPTEMBER 2020

confirmed that the nature of the HRI influences how
humans represent the nearby robot [64]. It is important to
note that social interaction only boosted positive attributions
(i.e., more human-uniqueness, human-nature, and warmth
traits), while negative attributions (i.e., animal-dehumaniza-
tion, mechanistic-dehumanization, and discomfort traits)
remained stable.
These results could point to two phenomena. First, negative
attribution does not change because, unlike humans, robots
seem to be considered objects. In contrast to a human being,
who can be dehumanized, a robot cannot be deanthropomor-
phized. One can only assign new human constructs [65]-[67]
or social traits [30] to the robot, as if this social agent were a
"blank page." However, it seems likely that this process applies
only to new HRI. In the same way as for humans, it is be possi-
ble that, once a robot has been anthropomorphized, subsequent
HRI might modulate these inferences. However, the question of
whether long-term HRI might increase, stabilize, or reduce this
type of anthropomorphism process remains unclear [68].
Second, these anthropomorphic inferences occur only in
specific contexts that require such inferences to be made.
The scientific literature shows that there are automatic neuro-
sociocognitive perceptual processes, such as face recognition
(automatically activated during simple perception, even with
nonhumans) [69], [70]. In the case of HRI, it seems likely that
the perception of and interaction with robots are based on
such processes. However, this "hijacking" of human-human
interactions, together with the automatic neurosocial process-
es they involve and that result from the social evolution of our
species, is not enough to cause the attribution of social con-
structs [44]. In other words, the strength of bottom-up inputs
is not enough in itself. To create anthropomorphic inferences,
it is necessary for the context to reinforce these bottom-up
inputs through top-down processes activated by the context
(e.g., the type of HRI, the robot's behavior). For example, in
both experimental conditions in this study, the robot was per-
ceived in the same way before the interaction. The robot was
not seen as more "mechanical" in one condition than in the
other as a function of the stability of the negative attributions.
In other words, the robot started with a certain quantity of
mechanical (e.g., superficial) and animal (e.g., amoral) traits,
and this quantity remained stable, even after the experimental
induction. However, we suggest that the activation of social
scripts during the social interaction would have led to the
attribution of a more anthropomorphic perception of the
robot [7], [13], [71]. The reason could be that when we inter-
act with creatures (even nonhumans), processes (e.g., mental-
ization) occur that allow us to attribute intentionality and
mental activity to the other to understand and predict that
other's behavior [44], [72].
We found the same effect for the "warmth" dimension of the
"warmth and competence" dyad, which is considered to repre-
sent universal dimensions of social perception [19]. This result
is interesting in that it suggests that, when people spontane-
ously interpret their impressions of others, these two dimen-
sions account for almost all of the variance (≈82%) [73]-[75].
IEEE Robotics & Automation Magazine - September 2020

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2020
