Perception, 2013, volume 42, pages 998 – 1000

doi:10.1068/p7569

SHORT AND SWEET Can looking at a hand make your skin crawl? Peering into the uncanny valley for hands

Ellen Poliakoff 1, Natalie Beach1, Rebecca Best1, Toby Howard2, Emma Gowen3

1 School of Psychological Sciences, The University of Manchester, Oxford Road, Manchester M13 9PL, UK; e‑mail: ellen.poliakoff@manchester.ac.uk; 2 School of Computer Science, The University of Manchester; 3 Faculty of Life Sciences, The University of Manchester Received 22 July 2013, in revised form 24 September 2013

Abstract. It is postulated that there is an uncanny valley, whereby human-like stimuli such as robots or animated characters that fall short of being fully human are perceived as eerie or unsettling. Previous research has explored the existence of this effect for faces and whole bodies, while here we explore responses to photographs of real and artificial hands. In keeping with the notion of an uncanny valley, prosthetic hands that were of intermediate human-likeness were given the highest ratings of eeriness. However, within the categories of hands, ratings of eeriness reduced as human-likeness increased, suggesting a more complex pattern. Further investigation of this effect will be of relevance to the design of prosthetic limbs and could be used to test theories of the uncanny valley and social perception with simple stimuli.

Keywords: hands, social, humanness, uncanny valley, prosthetics

More than forty years ago, Mori (1970) hypothesized the existence of an uncanny valley whereby people would find robots that were close to being human more uncomfortable than those that are clearly not human. Mori conceived of the valley as being low in ‘shinwakan’, which has been translated as familiarity, but it is argued that ‘eerie’ is a better description of the effect (MacDorman and Ishiguro 2006). Empirical investigation has revealed a peak of ratings of eeriness as faces become more human-like, and various explanations have been proposed, such as fear of death or the violation of expectations produced by the mismatch between human and nonhuman features (eg Saygin et al 2012). This effect may therefore indicate how expectations evolve during human social interactions and also has implications for understanding how people will interact with robots and avatars, which are increasingly used in a wide number of settings. However, the research to date has used faces or whole bodies, and here we investigate for the first time an uncanny valley for hand stimuli. This is not only of theoretical interest, but also has implications for the design of prosthetics. Indeed, Mori (1970) predicted that a realistic, artificial limb would unsettle the viewer when it becomes apparent that it is not real. Although this claim is supported anecdotally, to our knowledge it has not been tested empirically. Therefore, we obtained ratings of photographic images of real, prosthetic, and mechanical hands. We predicted that as hands were rated as more human-like there would be a peak in the ratings of eeriness prior to the highest ratings of human-likeness.

Forty-three right-handed participants were recruited (thirty-six female; mean age = 20 years, SD = 2.4 years). The study had local research ethics committee approval, and participants gave written informed consent. Participants rated a series of photographic images of hands (see figure 1) viewed on a 15-inch laptop screen. In each half of the experiment, participants rated the hands for eeriness or human-likeness (in a counterbalanced order) using a 9-point scale from ‘not at all’ to ‘extremely’. Eerie was defined as: ‘mysterious, strange, or unexpected as to send a chill up the spine’ and human-likeness as ‘having human

Can looking at a hand make your skin crawl? 999

Figure 1. [In colour online, see http://dx.doi.org/10.1068/p7569] Twenty-two coloured photographic hand images were selected from Internet searches: mechanical (N = 6), prosthetic (N = 9), and human (N = 7). The hands were in a similar upright posture, cropped at the wrist, and presented against a neutral background at a resolution of approximately 640 × 480 pixels. The graph shows mean (standard error) ratings for eeriness plotted against ratings of human-likeness for individual hands and the significant quadratic relationship between human-likeness and eeriness.

mechanical

prosthetic

human

Mechanical Prosthetic Human

0 1 2 3 4 5 6 7 8 9Rating of human-likeness

Rat

ing

of e

erin

ess

7

6

5

4

3

2

1

0

1000 E Poliakoff, N Beach, R Best, T Howard, E Gowen

form or attributes’. The next image was presented 2 s after the participant made their rating. Each hand image was horizontally flipped and presented once as a left hand and once as a right hand, in a random order. The average ratings were calculated for each image, as initial analysis showed no difference between left and right hands.

Figure 1 shows the mean rating for eeriness against the rating for human-likeness for each hand image. It can be seen that the lowest eeriness ratings (M = 1.94) were given for the real human hands, which also received the highest ratings for human-likeness (M = 8.45). Meanwhile, the mechanical hands received intermediate ratings of eeriness (M = 4.18) coupled with the lowest ratings of human-likeness (M = 1.69). In keeping with the notion of an uncanny valley, the prosthetic hands generally received the highest eeriness ratings (M = 5.20) and were rated as more human-like than the mechanical hands (M = 4.78). It is interesting to note, however, that prosthetic hand 9 received a particularly low eeriness rating (3.86) coupled with a particularly high human-likeness rating (5.56), suggesting that this hand was often mistaken for a real human hand. Overall, there was a significant quadratic relationship between human-likeness and eeriness (r2 = 0.830, p < 0.0005), which was also significant for thirty-seven out of forty-three participants. Within categories, there was a significant negative correlation between human-likeness and eeriness for both prosthetic (r = –0.759, p = 0.018) and mechanical hands (r = –0.943, p = 0.005) and which approached significance for human hands (r = –0.736, p = 0.059). In other words, more human-like hands were rated as less eerie, as has been observed for whole body movements (Thompson et al 2011).

Overall, we have demonstrated that prosthetic hands that fall short of being human are rated as being eerie. This is in keeping with an uncanny valley for hand stimuli, although the negative correlations indicate a more complicated relationship. Further work parametrically manipulating attributes (eg colour, texture) will be required to test whether the findings are due to a violation of expectations (Saygin et al 2012) or can be accounted for by less human-like stimuli being perceived as eerie (Thompson et al 2011), as suggested by the within-category correlations. Importantly, as hands can be presented without the potentially confounding effect of emotional facial expressions, they provide a simpler way to probe the mechanisms of the uncanny valley (cf Thompson et al 2011). The combination of visual features with temperature and touch would also allow the mismatch between multisensory elements to be investigated. Moreover, a human imitating the movement of a robotic, virtual, or prosthetic hand could indicate a successful interaction (Kahn et al 2007). ReferencesKahn P H, Ishiguro H, Friedman B, Kanda T, Freier N G, Severson R L, Miller J, 2007 “What is a

human? Toward psychological benchmarks in the field of human–robot interaction” Interaction Studies 8 363–390

MacDorman K F, Ishiguro H, 2006 “The uncanny advantage of using androids in cognitive and social science research” Interaction Studies 7 297–337

Mori M, 1970 “Bukimi no tani” [The Uncanny Valley] Energy 7 33–35Saygin A P, Chaminade T, Ishiguro H, Driver J, Frith C, 2012, “The thing that should not be: predictive

coding and the uncanny valley in perceiving human and humanoid robot actions” Social Cognitive and Affective Neuroscience 7 413–422

Thompson J C, Trafton J G, McKnight P, 2011 “The perception of humanness from the movements of synthetic agents” Perception 40 695–704