doi: 10.1038/s41598-022-14615-x.
Monkey visual attention does not fall into the uncanny valley
Affiliations
- PMID: 35817791
- PMCID: PMC9273626
- DOI: 10.1038/s41598-022-14615-x
Item in Clipboard
Monkey visual attention does not fall into the uncanny valley
Sci Rep.
.
Abstract
Very humanlike artificial agents can induce feelings of uneasiness in human perceivers. Stimuli that generate this response are said to occupy "the uncanny valley". Given inconsistent findings in the literature, whether or not nonhuman animals experience the uncanny valley is unclear. Here, we recorded the visual attention of eleven male rhesus monkeys as they viewed faces varying in realness across five levels, with visual attention measured by both number and duration of visual fixations on faces as a whole and on areas of interest within the faces (e.g., eyes, mouth). Face stimuli varied in terms of the realism of the image and behavior depicted by the face (lipsmack, threat, bared teeth, and neutral). We largely found no support that rhesus monkeys perceive an uncanny valley when viewing our stimuli; however, monkeys did generally pay more attention to eyes and less attention to mouths in real images compared to less realistic images. Across all stimuli, monkeys' visual attention was drawn to the mouths of images when teeth were visible. These findings suggest that rhesus monkeys in our study did not display an uncanny valley effect when viewing realistic stimuli but did percieve affective information depicted by faces regardless of how real those faces appear.
© 2022. The Author(s).
Conflict of interest statement
The authors declare no competing interests.
Figures
Visual descriptions of the UV hypothesis and stimuli used in the present experiment. (a) The UV hypothesis developed by Mori (1970) suggests that as human likeness increases so does affinity, until a certain threshold is reached at which point affinity quickly drops (producing the UV) before spiking up again to its highest level for a real human image. Figure modified from,. (b) Examples of stimulus images we used for a threat face ranging from a Real image of a monkey through increased levels of manipulation (Faces 1 through 4, where Face 4 is the least realistic) to create a realness continuum for each stimulus.
Frequency of fixations on the eyes of the face stimulus as a function of stimulus realness separated by facial behavior of the stimulus. Each subject’s mean behavior is depicted by a gray line and the black line represents the average across all subjects. Each panel represents data from 1040 individual trials. (a) Visual attention to the eyes of faces displaying neutral (or no) behavior. Minimal evidence for a UV effect in which subjects fixate less frequently on the Face 1 stimulus than either the Real (z = − 4.102, p = < 0.001) or Face 2 (z = 2.75, p = 0.047) stimulus. (b) Visual attention to the eyes of faces displaying a bared teeth display revealed no evidence for a UV effect. Subjects looked at the eyes of the real image more frequently compared to the less realistic Face 3 (z = − 3.15, p = 0.014), but there were no other significant differences between stimulus realness. (c) Visual attention to the eyes of faces displaying a lipsmack display reveals no evidence for a UV effect. Subjects looked at the eyes less frequently in the least real image Face 4 compared to either the Real image (z = − 3.21, p = 0.012) or most real, but still manipulated image, Face 1 (z = − 2.92, p = 0.029), indicating reduced attention to less realistic eyes. (d) Visual attention to the eyes of faces displaying a threat display reveals no evidence for a UV effect. Subjects looked at the eyes fewer times in the least real image Face 4 compared to either the Real image (z = − 3.35, p = 0.007) or most realistic of the manipulated images, Face 1 (z = − 2.89, p = 0.032). Subjects also looked at the eyes fewer times in the Face 3 image compared to the Real image (z = − 2.86, p = 0.034). *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent mean ± 95% Confidence Interval.
Visual attention directed to the screen but not the face. Fixation count data are in top panels and fixation duration data in bottom panels. In plots on the left (a,c), gray lines represent average behavior of a given subject pooled across facial behaviors and the black line represents average behavior across all subjects. In plots on the right (b,d), individual data points are overlaid on boxplots with violin plot density functions on the right for each facial behavior pooled across face realness levels. (a) Subjects looked outside more times when the stimulus image was the most realistic of the manipulated images Face 1 compared to when it was the Real image (z = − 3.23, p = 0.011), Face 3 (z = 2.80, p = 0.041) or the least realistic Face 4 (z = 4.34, p < 0.001), suggesting that there may be some degree of aversion to the highly realistic Face 1 image. Monkeys also fixated outside the stimulus more times when the image was Face 2 than when it was the least realistic Face 4 (z = 3.06, p = 0.019). Data are summarized across 832 individual trials per face realness level. (b) Subjects looked outside the face stimulus more times when the facial behavior of the stimulus was neutral compared to when it was bared teeth (z = − 2.91, p = 0.019). Data represent 1040 trials per facial behavior. (c) Subjects did not vary the duration of their fixations on the outside of the stimulus based on stimulus realness. Data are averaged across 771–794 individual trials per level of face realness. (d) Subjects fixated on the outside for longer when the stimulus displayed a neutral facial behavior compared to when it displayed bared teeth (z = 2.70, p = 0.035). Data represent 964–988 trials per facial behavior. *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent mean ± 95% Confidence Interval.
Duration of fixations directed at the eyes of the face stimuli. (a) Gray lines represent average fixation duration of a given subject across all facial behaviors. The black line represents average fixation duration across all subjects. Data in this panel represent 514–573 individual trials per Face Realness level. Subjects looked longer at the eyes in the Real image compared to either the least realistic Face 4 (z = 4.31, p < 0.001) or Face 3 (z = 3.51, p = 0.004). Subjects also looked longer at the eyes in the most realistic manipulated Face 1 image compared to the least realistic Face 4 image (z = 3.41, p = 0.005), which contradicts what would be expected for a UV in which Face 1 is anticipated to receive the least amount of attention. (b) Plots represent individual data points (622–722 trials per facial behavior) overlaid on boxplots on the left with violin plot density functions on the right for each facial behavior. Monkeys altered how long they looked at the eyes based on facial behavior of the stimulus, such that monkeys spent less time looking at the eyes of a face with the bared teeth display than one displaying a neutral facial behavior (z = 2.40, p = 0.004). *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent mean ± 95% confidence Interval.
Visual attention directed toward the mouth AOI of the face stimuli. Fixation count data are in top panels and fixation duration data in bottom panels. In plots on the left (a,c), gray lines represent average behavior of a given subject pooled across facial behaviors and the black line represents average behavior across all subjects. In plots on the right (b,d), individual data points are overlaid on boxplots with violin plot density functions on the right for each facial behavior pooled across face realness levels. (a) Subjects looked less frequently at the mouth of Real images than Face 4 (z = 4.25, p < 0.001) or Face 2 (z = 3.16, p = 0.014). Data are summarized across 832 individual trials per face realness level. (b) Subjects looked at the mouth fewer times when the stimulus depicted a neutral facial behavior compared to any other facial behavior (neutral vs bared teeth: z = 13.77, p < 0.001; neutral vs lipsmack: z = 9.30, p < 0.001; neutral vs threat: z = − 10.46, p < 0.001) and more times when the facial behavior depicted was bared teeth compared to all other facial behaviors (bared teeth vs lipsmack: z = 6.06, p < 0.001; bared teeth vs threat: z = 3.65, p = 0.002). Data represent 1040 trials per facial behavior. (c) Subjects fixated on the mouth for less time when the image was real compared to any manipulated image (Face 1: z = − 3.00, p = 0.023; Face 2: z = − 4.96, p < 0.001; Face 3: z = − 3.75, p = 0.002; Face 4: z = − 5.16, p < 0.001). Data are averaged across 240–256 individual trials per level of face realness. (d) Subjects fixated on the mouth for less time when the facial behavior of the stimulus was neutral compared to either bared teeth (z = − 3.38, p = 0.004) or threat (z = 3.81, p = 0.001). Data represent 157–465 trials per facial behavior. *p < 0.05, **p < 0.01, ***p < 0.001. Error bars represent mean ± 95% confidence Interval.
References
-
- Mori M, MacDorman K, Kageki N. The uncanny valley. IEEE Robot. Automat. Mag. 2012;19:98–100. doi: 10.1109/MRA.2012.2192811. - DOI
-
- Ho CC, MacDorman KF. Measuring the uncanny valley effect: Refinements to indices for perceived humanness, attractiveness, and eeriness. Int. J. Soc. Robot. 2017;9:129–139. doi: 10.1007/s12369-016-0380-9. - DOI
