Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Accelerat ing t he world's research.

Ruijten, P. A. M., Bouten, D. H. L.,


Rouschop, D. C. J., Ham, J., & Midden,
C. J. H. (2014). Introducing a Rasch...
Peter Ruijten

Related papers Download a PDF Pack of t he best relat ed papers 

Too humanlike t o increase my appet it e: Disposit ion t o ant hropomorphize animals relat es t o …
Kat arzyna Byrka, Michał Bilewicz

Dimensions of Ant hropomorphism: From Humanness t o Humanlikeness


Jakub Złot owski

T he Robot Got t he Looks. Social Robot s Appearance as Predict or of Ant hrpomorphism and Int ent ion …
jean-christ ophe Giger, Nuno Piçarra
Introducing a Rasch-Type Anthropomorphism Scale
Peter A. M. Ruijten*, Diane H. L. Bouten, Dana C. J. Rouschop, Jaap Ham, Cees J. H. Midden
Eindhoven University of Technology
School of Innovation Sciences
Eindhoven, Netherlands, P.O. 513
p.a.m.ruijten[at]tue.nl

ABSTRACT differences in focus, a robot that is considered to be


In human-robot interaction research, much attention is given to anthropomorphic according to one definition might be seen as
the extent to which people perceive humanlike attributes in robots. non-anthropomorphic according to another one, and vice versa. In
Generally, the concept anthropomorphism is used to describe this other words, since in most of the current research only one of
process. Anthropomorphism is defined in different ways, with these different aspects of anthropomorphism is being measured,
much focus on either typical human attributes or uniquely human the results are non-generalizable and non-comparable.
attributes. This difference has caused different measurement tools We argue that the attribution of typically and uniquely human
to be developed. We argue that anthropomorphism can best be characteristics to robots reflect a single human predisposition, as
described as a continuum ranging from low to high human is indicated in Figure 1. In other words, we expect that
likeness, and should be measured accordingly. We found that anthropomorphic characteristics may be ordered according to the
anthropomorphic characteristics can be invariantly ordered ease to which they are ascribed to robots in a manner that is the
according to the ease with which these can be ascribed to robots. same for all individuals in their encounter with all robots. In
particular, we expect that typically human characteristics are more
General Terms easily ascribed to robots than uniquely ones, regardless of the
Design, Human Factors, Measurement, Standardization. individuals’ predisposition to attribute anthropomorphic
characteristics to robots. Inversely, the expected ordering of
Keywords anthropomorphic characteristics would prescribe that people who
Anthropomorphism, Rasch. attribute uniquely human characteristics to a robot must also
ascribe typical characteristics to it.
1. INTRODUCTION
As technology plays an increasingly prominent role in our lives, Figure 1: Conceptual representation of human characteristics
our perception of those technologies becomes more important. To on an anthropomorphism continuum
describe how people perceive robots, the concept of
anthropomorphism is often used. Generally, this involves the
tendency to attribute human characteristics to robots. However,
not all researchers use the exact same definition. For example, [1]
refer to anthropomorphism as “the attribution of a human form,
human characteristics, or human behavior to nonhuman things”
(p.74), while [2] define anthropomorphism as “a process of
inductive inference whereby people imbue the real or imagined
behavior of other agents with humanlike characteristics,
motivations, intentions, or underlying mental states” (p.411). A
closer look of these definitions shows us that differences arise due
to referring to different types of human characteristics. Human
characteristics are generally divided in typical and unique.
Typically human characteristics are generally associated with One model that mathematically describes these expectations is the
humans, but humans are not the only entities that have those traits Rasch model, which assumes that the trait to be measured is a
[3]. An example is being able to interact with a changing one-dimensional latent variable [5, 6, 7]. In the model, two
environment. Uniquely human characteristics refer to traits that parameters are estimated; one to distinguish between individuals,
are unique for the human species [3], like having a moral and another one to order the latent variable’s characteristics [6].
conscience. We note that [1] mainly refer to typically human Both of them are interesting, because they can tell us something
characteristics, while [2] mainly refer to uniquely human about the underlying process of anthropomorphism. We aim to
characteristics. Furthermore, [4] take on yet another approach. develop a measurement scale for anthropomorphism based on the
While they adopt [3]’s concepts of typically and uniquely human, Rasch model.
they only refer to cognitive traits, leaving appearance and physical
abilities that characterize humans aside. Because of these 2. METHOD & RESULTS
As a first step, we designed 37 yes/no items about
Permission to make digital or hard copies of all or part of this work for anthropomorphic qualities a robot can have. We subjected these
personal or classroom use is granted without fee provided that copies are items to a content validity test, in which we tested all items on
not made or distributed for profit or commercial advantage and that unique humanness (n=17, Mage=21.1, SDage=1.79), typical
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
humanness (n=20, Mage=19.9, SDage=1.98), sociality (n=18,
requires prior specific permission and/or a fee. Mage=19.9, SDage=1.79) and valence (n=20, Mage=20, SDage=2.50),
HRI'14, Mar 03-06 2014, Bielefeld, Germany where valence should not be correlated with anthropomorphism.
ACM 978-1-4503-2658-2/14/03.
http://dx.doi.org/10.1145/2559636.2559825 *Corresponding author
Hereafter, we tested the full measurement scale in an online the ordering of items on the anthropomorphism continuum is
sample with 124 participants (Mage=26.1 and SDage=8.82) and independent of the study.
analyzed the probability of each item to be answered with ‘yes’.
The items that were most likely to be answered with ‘yes’ have a 3. DISCUSSION
low so called ‘item difficulty’ and indicate low anthropomorphism In this paper we argued that anthropomorphism does not simply
whereas the items that were least likely to be answered with ‘yes’ consist of one single attribution, but covers a continuum of
have a high ‘item difficulty’ and are high anthropomorphic. These attributions ranging from typically to uniquely human
item difficulties thus tell us something about the underlying characteristics. Most of the previous research conducted on
process of attributing humanlike characteristics to robots. anthropomorphic robots only investigated parts of the continuum,
which limits the interpretation of the underlying process.
We used the Rasch model fit statistics (for an overview, see [6])
to determine which of the 37 items were fitting the model. During We argue that when a robot is perceived to have some uniquely
this process, we removed 12 items (about half of those because of human characteristics, this indicates that this robot must contain
misfit, half because they did not correlate well with the content both typically and uniquely human characteristics. An additional
validity results) to end up with a revised 25-item scale. Infit mean advantage of measuring the two aspects simultaneously is that it
squares for the 25 items ranged from .77 to 1.21; outfit mean shows that all characteristics belong to one continuum, and that
squares ranged from .50 to 1.85. these characteristics are hierarchically ordered in a way that is the
same for all individuals. This ordering tells us something about
Next, we correlated item difficulties of both the 37-item scale and the underlying process of anthropomorphism.
the 25-item scale with the results of the content validity questions.
A high correlation means that the items have similar scores on In sum, we have developed a measurement scale that consistently
anthropomorphism and the content validity measure. All measures anthropomorphism as a continuum over different
correlations confirmed our expectations and are shown in Table 1. studies. Furthermore, the ordering of the items seems to be stable
Table 1: Correlations between item difficulties and content over different samples. We have shown that the Rasch model can
validity measures be used to mathematically describe this invariant ordering of
anthropomorphic inferences. Follow-up research will be designed
37-item 25-item to further develop the scale and to investigate its robustness.
Uniquely human r = .434, p = .007 r = .551, p = .004
Typically human r = .603, p < .001 r = .652, p < .001 4. ACKNOWLEDGMENTS
Sociality r = .556, p < .001 r = .609, p = .001 We would like to thank Antal Haans and Maartje van ‘t Sant for
their helpful contributions to this project.
Valence r = -.098, p = .563 r = -.194, p = .352

2.1 Measurement verification 5. REFERENCES


[1] Bartneck, C., Kulic, D., Croft, E., & Zoghbi, S. (2009).
After having validated out measurement scale, it still had to be
Measurement Instruments for Anthropomorphism, Animacy,
verified. That is, we needed to know to what extent our
Likeability, Perceived Intelligence, and Perceived Safety of
measurement scale related to already existing anthropomorphism
Robots. International Journal for Social Robotics, 1 (1), 71-
measurements.
81. DOI: 10.1007/s12369-008-0001-3.
In an online study, 131 participants (Mage=34.86, SDage=17.592) [2] Waytz, A., Morewedge, C., Epley, N., Monteleone, G., Gao,
were divided over two conditions. In each condition, participants J., & Cacioppo, J. (2010). Making Sense by making Sentient:
watched a video of a robot that either only had typically human Effectance Motivation Increases Anthropomorphism. Journal
characteristics or that also had uniquely human characteristics. of Personality and Social Psychology, 99 (3), 410-435. DOI:
Next, participants completed multiple anthropomorphism 10.1037/a0020240.
questionnaires: our 25-item scale (α = .77), a 6-item questionnaire [3] Haslam, N., Bain, P., Loughnan, S., & Kashima, Y. (2008).
proposed by [2], referred to as the Waytz scale (α = .78), and the Attributing and denying humanness to others. European
5-item anthropomorphism part of the Godspeed questionnaire [1], Review of Social Psychology, 19, 55-85.
referred to as the Bartneck scale (α = .71).
[4] Eyssel, F., Kuchenbrandt, D., Bobinger, S. (2011). Effects of
When checking for differences between the videos, no significant Anticipated Human-Robot Interaction and Predictability of
effects were found on any of the anthropomorphism Robot Behavior on Perceptions of Anthropomorphism.
measurements (all p’s > 0.15). Nevertheless, significant Proceedings of the 6th international conference on Human-
correlations were found between our scale (M = 9.84, SD = 2.79) robot interaction, 61-68. DOI: 10.1145/1957656.1957673.
and both the Waytz scale (M = 1.23, SD = 0.40, r = .453, p < .001) [5] de Jong-Gierveld, J., & Kamphuis, F. (1985). The
and the Bartneck scale (M = 2.52, SD = 0.82, r = .232, p = .008). Development of a Rasch-Type Loneliness Scale. Applied
This suggests that our scale measures the same concept as the Psychological Measurement, 9, 289-299.
other scales. It appears that our scale measures personal [6] Bond, T. G., & Fox, C. M. (2007). Applying the Rasch
dispositions to attribute humanlike characteristics to robots and model: Fundamental measurement in the human sciences.
conjointly the ease with which various anthropomorphic
characteristics are attributed to robots. Furthermore, since both [7] Green, K.E., & Frantom, C.G. (2002). Survey Development
our studies applied the same scale, it is possible to make a and Validation with the Rasch Model. Paper presented at the
comparison between them. The item difficulties strongly correlate International Conference on Questionnaire Development,
between the two studies (r = .962, p < .001), which suggests that Evaluation, and Testing. Charleston, SC, November 14-17,
2002.

You might also like