What would make you trust a robot?
People don’t trust self-driving cars. They trust themselves to drive. Yes, a robot doesn’t fall asleep or get drunk or look at their phone. But people don’t care. They don’t trust self-driving cars. How much? Well, AAA’s annual automated vehicle survey from March 2019 found that 71 percent of people are “afraid” to ride in fully self-driving vehicles. Scientists are trying to get people to trust A.I robots more. How? By having them explain themselves.
“In the past, robotics wasn’t closely tied to psychology. This is changing,” says Dr. Hongjing Lu, a professor in the department of psychology and statistics at UCLA, speaking to Inverse during a press conference about her team’s recent study offering a potentially new way forward for A.I. “Trust is a central component of humans and robots working together.”
You interact with artificial intelligence everyday. There’s a strong chance that it’s part of the reason you’re reading this right now, and from here data points will be gathered and some algorithm will take this story into account as it decides what ads to present on social media. Apps like Google Maps and Uber use it all the time to find the best routes for cars.
These examples feel passive—anyone who has used social media or Google Maps can attest to their ease and quickness. Users can assume that some work is going on behind-the-scenes but can’t look underneath the hood. It’s enough to make their algorithms feel invisible, like the editing of a good movie. But just as editors never get their due, the UCLA team feels that the hidden nature of algorithms has made people suspicious of them in larger, more active scenarios.
But, of course, what is trust?
“Trust is a very complex psychological construct,” Lu says. “It can emerge in different ways at different stages of the relationship development. So therefore, we started from the very first set of trusted development. When they say I trust you to do X as we do the past acts, they basically have the believe that do the past acts with competence and we can predict that you will actually do the task right in the way, so conveying under the impression of competence and the predictability is important for fostering human trust and human robotics business.
So to agitate on such impression the machine not only and need to achieve past performance, but it also needs to provide good explanation for it’s behavior. That’s one of the core ideas in this paper.”
Scientists decided that opening a pill bottle would provide enough of a challenge for their robot to be a worthy test. It’s a complex task, as any human with a headache can attest. There’s the act of it requires “pushing on a lid or somehow subtly manipulating the lid using force, and the robot learns to mechanisms to complete the task,” says Mark Edmonds, a PhD Candidate in Computer Science at UCLA who also participated in the teleconference.
