getting lost in the uncanny valley
TRANSCRIPT
Getting lost in the Uncanny Valley
Learning how to pack for eerie explorations
Dr Stephanie Lay, Psychology, The Open University
Welcome to the Uncanny Valley
The landscape circa 2006
That looks eerie
It’s because it’s almost
human
That looks very close to
human…
Therefore it must be uncanny
My research approach: plans and realityThe plan: my research proposal• Examine the uncanny
valley effect from cognitive, behavioural and emotional angles. • Isolate and test factors
causing the effect.• Find ‘the answer’ – sell it
to video game companies and Pixar!
The reality• Three research phases:
starting with a broad exploration of near-human entities. • Finding interesting
elements to follow up.• Learning how to ask
awkward questions.• Find ‘an answer’ – sell it to
examiners!
2006-2015 … Three research phases
Phase One• Small scale• Mainly
qualitative• Exploratory and
designed to generate more questions
Phase Two• Large scale• Quantitative
and experimental
• Following up on findings of Phase 1
Phase Three• Large scale• Quantitative
and experimental
• Following up on findings of Phase 1 and 21 and 2
Phase One: Exploring the UVE
Ratings and descriptions of a gallery of near-human images
Very little was known about the UVE at this time … • What might cause it?• Is it a real effect?• Can it reliably be elicited?• Is anyone susceptible? • Does it extinguish over
time?• Is it an emotion in its own
right?
These results were surprising. I had expected to see a clearer dip in the relationship between human-likeness, strangeness and eeriness.Instead, I found these 3 distinct clusters of images. My next study went on to explore this in more detail by selecting examples from each cluster.
Phase One: Exploring the UVE
Phase One: Exploring the UVE
Phase Two: A focus on faces• One thing that stood out about the faces found most eerie in Phase 1 was how
their eyes appeared. • I’d also noticed this anecdotally when looking at images in my gallery. • There had also been new research published in this period about the role of
exaggerated or unusual eyes in triggering the effect. • I decided to investigate this systematically, and test out a theory about how
near-human faces might be processed.
Phase Two: A focus on faces
The uncanny valley surprised me again here.Not only was the pattern of human likeness, strangeness and eeriness not as I expected, but the effect on recognition speed was not as predicted. The near-human faces ‘should’ have been recognised faster than the human and non-human faces. Interestingly, there were also clear differences between the four categories.
Phase Three: Towards an Explanation
Phase two had shown me that there were clear differences in eeriness for faces, but they weren’t as expected. The literature into the uncanny valley effect continued to grow, and one avenue for exploration was mismatched expressions. Tinwell et al (2009, 2010, 2011) had suggested dead eyes might cause the effect.But what if there was more to it than that?
Phase Three: Angry or scared eyes, warm smile
Circling around the uncanny valley
The results from my three phases were novel and interesting discoveries about face perception and the uncanny valley effect.However, the stages I went through made me realise that this is a complex area for research. I came up with a set of 5 principles for research in this area.If applied, these allow researchers to investigate the valley effect and be sure that their conclusions are based on evidence, not circularity.
1) Stimuli should cover a range of levels of human likeness, with a minimum of 5 points including human and nonhuman anchor points.2) To produce the graph’s X axis, the human likeness of the stimuli displaying the identified quality should be controlled or measurable. 3) The Y axis of the graph, labelled as familiarity, represents the emotional response or reaction to the stimuli. This is not as clear-cut as the human likeness dimension as there are many different interpretations of what could be meant by this axis. 4) When the two measurements of human likeness and emotional response are plotted against each other, the path described by the response measurement should display a single clear dip or deviation from a linear path, occurring somewhere between 50% and 100% on the human likeness axis. 5) A rating of eeriness should be collected from participants for each of the stimuli in addition to the familiarity or emotional response measure. Without it, any observed valley could be explained as a mere dip in response to the stimuli and it would not be possible to confidently assert that the valley was uncanny in its nature. Lay, Brace, Pike, and Pollick (2015) – under review.
Leaving the Uncanny Valley...In 2006, there were only a handful of academic publications about the uncanny valley. In 2015, there were over 300. Where would I take the research next? • Embodied agents – robotic
receptionists, real baby dolls and robots as companions.
• Virtual humans – will a digital actor ever be completely convincing?
…for now
Links and further readingMori, M., 1970. The Uncanny Valley. Energy, 7(4), pp. 33-35 translated as ‘The Valley of Familiarity’ in MacDorman, K., 2005. Androids as an experimental apparatus: why is there an uncanny valley and how can we exploit it? CogSci-2005 Workshop: Toward Social Mechanisms of Android Science, July 25 – 26 2005, pp4-8.Tinwell, A., Nabi, D. A., & Charlton, J. P. (2013). Perception of psychopathy and the Uncanny Valley in virtual characters. Computers in Human Behaviour, 29(4), 1617–1625.
My research website: http://uncanny-valley.open.ac.uk Curated ‘scoopit’ channel: http://www.scoop.it/t/uncanny-valley @uncannyvalley on Twitter & Flickr