The past nine weeks have been an exercise in frustration for Westworld fans struggling to figure out who’s ‘bot and who’s not. Even within the park’s walls, people can wind up up in bed with hosts without their knowledge. While this fictional scenario may stoke panic in the AI-shy among us, scientists want to assure you: Real humans will always be able to tell.
In a study published recently in the journal Nature Communications, a team of researchers from the University of California-Berkeley report that humans are really, really good at discerning between living and non-living things. Because we’ve evolved a unique visual strategy called “ensemble lifelikeness perception,” humans can “perceive what’s really alive and what’s simulated in just 250 milliseconds,” the study’s lead author Allison Yamanashi Leib, a postdoctoral scholar in psychology, said in a release.
Scientists used to think that it took people a certain level of active, cognitive deliberation to figure out if what they were looking at was lifelike or not. Turns out that’s far from the truth: By studying the snap judgments that 68 participants made about the “lifelikeness” of photos of random objects, both living and inanimate, Yamanashi Leib found that people can assess what’s real almost instantaneously by relying on their vision. This is because ensemble lifelikeness perception makes us focus on the broad elements of a scene, rather than the fine details. So, if you were to walk into Sweetwater’s Main Street, your visual system would take in the whole dusty scenario and instantaneously appraise the “overall level of activity in a scene,” as Yamanashi Leib put it, rather than ponder, long and hard, about whether individual hosts are real or not. We already use this system when we’re visiting new places, shopping, and attending parties in order to “gauge where the action is at,” she points out.
It makes sense that evolution would have favored people who could spot lifelikeness instantaneously. Just imagine the opposite — a world in which people constantly struggled to decide if a thing was alive or not. Socializing would be much harder; our ability to infer emotional and social information from an image, as the study authors point out, relies on an initial, more rapid assessment of what actually we’re looking at.
The elements of a scenario that lead us to decide whether it’s lifelike or not, however, remain elusive. Reality is more than just the sum of its parts, the researchers note:
[Our] impression of the liveliness and energy depicted in a photo of mounted animals at a natural history museum compared to a photo of those animals at the zoo is not well explained by ensemble perception of colour (similar), texture (similar) or even biological motion information (irrelevant in the case of static pictures). The context alone also does not give away the answer, as it depends on an interaction between the objects and context, among other factors. Museums, for example, can be more or less animate than zoos, depending on things like whether a busload of children is arriving.
So, in light of this study, Westworld’s conjecture that hosts with perfectly human-like skin, voice, and mannerisms should convince people that they’re real now seems a bit suspect. It’s good news for us, considering we’re moving toward a future where robots really are becoming more lifelike in their looks and behavior. It seems that for actual humans, the park will always just be a park.