Does a robot's creepiness change how we judge its moral choices?
When creepy-looking robots made the choice to sacrifice one person to save many, people judged that choice as less moral than when a human or a friendly-looking robot made it, but the pattern reversed for choices not to act.
Source
Moral Uncanny Valley revisited - how human expectations of robot morality based on robot appearance moderate the perceived morality of robot decisions in high conflict moral dilemmas
Study at a glance
- Design
- Human experiment — Study 1: preregistered 2 (utilitarian vs deontological decision) x 5 (four robots of varying uncanniness vs a human) between-subjects online experiment; Study 2: 2 x 4 between-subjects lab experiment with CGI human-robot morphs
- N
- N=624 · Study 1: 624 Prolific participants after exclusions; Study 2: 221 adults recruited at a public library in Espoo, Finland
- Population
- English-fluent online adults (Study 1) and Finnish public-library visitors aged 18-80 (Study 2)
- Outcome
- Rated morality of the agent's decision across high-conflict trolley-type dilemmas (1-7 scale); agent creepiness, mind perception and baseline morality as checks and covariates
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
In a preregistered online study, 624 adults saw a photo of either a human or one of four real robots that varied in how creepy they looked, rated the agent's mind and creepiness, and then read high-conflict trolley-type dilemmas in which that agent either acted to sacrifice one person (utilitarian) or refused to act (deontological). They rated how moral each decision was. A second lab study with 221 library visitors repeated the idea using computer-generated human-robot morph images.
What they found
In Study 1, the type of decision had a large effect, while agent appearance mattered only in interaction with it: utilitarian decisions formed a 'valley', rated least moral for the two creepiest robots, whereas deontological decisions formed an unexpected 'mountain', rated most moral for those same robots. Controlling for creepiness removed the valley for utilitarian decisions, but controlling for mind perception did not, suggesting the effect is not simply about how much mind the robot seems to have. Study 2's stylised CGI images produced no effects at all.
The limits
What it doesn't show
The appearance effects were small and several contrasts were only just significant, and the author states the studies are not highly powered one-off results needing independent replication. The mountain-shaped deontological effect was not preregistered and was driven mainly by one robot that was not the creepiest. Only four robot photos were used, so the results may reflect particular images rather than uncanniness in general, and Study 2 shows the effect disappears with non-photorealistic stimuli. Participants judged hypothetical vignettes, which may not reflect reactions to real robots.
Key terms
- Uncanny valley
- The finding that agents that look almost but not quite human feel creepier than clearly mechanical or fully human ones.
- Human-robot moral asymmetry
- The tendency to judge the same moral decision differently depending on whether a human or a robot made it.
- High-conflict moral dilemma
- A dilemma, like pushing someone off a footbridge, in which the decision-maker directly and personally causes harm to save others.
- Mind perception
- How much an agent is seen as able to feel, experience and act intentionally.
- Preregistration
- Publicly recording hypotheses and analyses before collecting data, to distinguish planned tests from exploratory findings.
Flashcards
0 of 9 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
In Study 1, whose utilitarian decisions were rated least moral?
Common questions
Did people simply dislike robots making moral choices?
No. There was no overall main effect of the agent; appearance only mattered in combination with whether the agent acted or refrained from acting.
Why did Study 2 find nothing?
Its images were stylised CGI morphs that do not reliably produce the uncanny valley, so the author argues photorealistic stimuli are needed to see the moral effect.
Is the effect explained by robots seeming to lack a mind?
Apparently not: adding mind perception as a covariate barely changed the results, whereas controlling for creepiness removed the utilitarian valley.
More on Moral psychology