In a 2013 episode of Black Mirror called “Be Right Back”, a woman interacts with an artificial intelligence based on her dead boyfriend – first by text and voice communication, and then by means of an advanced robot that fully resembles the human that the AI was trained on. Clips from this episode were recently used in a small study to assess how people perceived these different methods of communicating with an AI, and the voice-only version came out on top.
Looking too much like a human might not be ideal for AI companions, according to a new study.
getty
The volunteers didn’t get to see the full context of the episode, but enough to understand that the woman in the story (played by Hayley Atwell) was talking to the AI to feel less lonely. One half of the volunteers saw her interacting with the voice-only version and the other half watched the clips where she talks to the human-like robot.
All of the volunteers were then asked whether they thought an AI like the one they just saw could be helpful for lonely people, and if they would recommend it. Overall, the people who watched the video where the woman talks to a voice-only AI were more likely to recommend it than the group who saw the clip where she interacts with the humanoid robot.
The researchers did specifically ask the volunteers why they rated the version of the AI they saw as helpful or not, but they think it has something to do with the uncanny valley. Kelly Merrill Jr., a graduate student at Ohio State University who was the lead on this research, told Ohio State News, “We think it may seem a little too creepy to have these embodied robots that act and look almost human,” says Merrill.
The idea of the uncanny valley was first introduced in 1970 by roboticist Masahiro Mori, who suggested that people would generally feel more comfortable with objects when they start to look like humans a little bit, but once they look almost like a human the opposite happens and the objects suddenly become creepier.
Besides robotics, the uncanny valley is also seen in visual effects for film and TV. Realistic animations of humans reach a similar creepy level when they’re less cartoonish and more like a video of real people (but not quite enough). For many viewers, the 2004 film The Polar Express reached this uncanny valley, for example.
In the Black Mirror episode, robot Ash is not portrayed by an android or an animation, but by a human actor (Domhnall Gleeson) pretending to be a robot. Still, the deliberately robot-like acting from a very real human could invoke that same uncanny feeling.
But perhaps there is a simpler explanation for people’s preference for the voice-only AI: We already have systems like that in the form of Siri, Alexa, Google Nest and other voice-activated assistants. According to Merrill , “We already talk to disembodied AI through our smartphones and smart speakers, so we are used to that and comfortable with those kinds of interactions.”
Maybe a voice-based AI companion for lonely people just seemed more realistic than recreating a full life-sized human likeness in robot form.
Obviously this was just a small study based on a fictional show, but as AI research gets more advanced the question of what our future AI companions should look like becomes more relevant and will likely be the subject of more studies over the coming years. So far, people seem to find the idea of humanoid robot companions a bit creepy, and perhaps also just a bit over the top.
