What the uncanny valley offers that VR doesn't.
There's nothing like social pressure when it comes to getting people to do stuff. That has some obvious dark sides when it comes to, say, internet comment sections and Tide pods but it's also a force for good. We want approval. Approval releases dopamine, which feels good. This is a feature of humans that can be engineered in beneficial ways.
This is the work, generally, of Maja Mataric, a computer science and neuroscience researcher at the University of Southern California. Martaric is interested in the therapeutic uses of physical artificial intelligence: robots. More specifically, she's interested in "socially assistive robots." These are meatspace machines that can be used for "coaching, motivation, and companionship" in rehabilitative and caregiving contexts—not replacing humans in those roles, but augmenting them.
"Social creatures are driven by social interactions, mostly by embodied interaction," Mataric explained in a presentation at Re-Work's AI Assistant Summit this week in San Francisco. That is, there is a limit to the capabilities of screen-based AI. Neurologically, we are more activated when in the presence of a physical thing. "We are fundamentally embodied creatures."
A robot doesn't even have to really do anything to have an effect on social compliance, Mataric said. It can just sit there, though that would be a waste of potential.
An example she used is of a stroke patient regaining the use of a limb. This is exhausting and tedious. When we lose some use of a body part, our natural response is to not use it. The effects of a stroke, however, require the opposite. To regain use of a limb is largely a matter of forced usage. This rebuilds the parts of the brain responsible for using that limb.
A robot, in Mataric's example, can act as a cheerleader of sorts, validating and encouraging the desired behavior. Even just having the robot interaction has to the potential to limit the progression of dementia, she said.
The benefits of socially assistive robots have been validated in preclinical trials, but this is still a difficult problem. In particular, implementing personality in a meaningful way is a big challenge. Making a good robot face, one that really lands in the uncanny valley sweet spot, is hard.
Crucially, this is not something that can be replicated in virtual reality. "The problem with virtual reality is immersion," Mataric said. "You're not in this world."