He grabs the briefcase. Spindles of smoke unravel across the warehouse floor. Flames crackle and creep closer. Crumbling debris blocks an exit.
A metal robot, about the height and girth of a large garbage can, appears near the avatar.
“I know the way. Follow me.”
The scenario is part of a government-funded online experiment scheduled for June and run by David Atkinson of the Florida Institute for Human and Machine Cognition in Ocala.
The project, backed by the armed forces, has only a few final adjustments needed before Atkinson can test what it takes for a person to trust a robot in emergencies, a problem the Air Force is particularly interested in, the senior researcher said.
“The armed forces [want] to see software working alongside people rather than simply being used as tools,” said Atkinson, who’s directed the Lunar robotic missions at NASA and was a program manager at the Air Force Office of Scientific Research.
“But like any team member, people have to make decisions with trust.”
Ben Knotts, Atkinson’s Air Force sponsor, couldn’t get permission to speak about the Air Force’s interest.
However, according to an Air Force report, “Developing methods for establishing ‘certifiable trust in autonomous systems’ is the single greatest technical barrier that must be overcome to obtain the capability advantages that are achievable by increasing use of autonomous systems.”
Military commanders are uncomfortable using machines to interpret data, Atkinson said. The Air Force wants to know more about how people interact with machines as it takes more pilots out of planes.
“I’m most curious about the initial reaction,” Atkinson said about the experiment, which will be conducted twice a day for four months in SecondLife, an online world simulation.
He’ll ask users about 30 questions to gauge how anxious they are and their attitudes about the robot during the encounter. The results of his theoretical work will be used to program robots for emergency search–and-rescue situations.
“We want robots that are going places and doing things that are too dangerous for people,” he said.
People often rely on machines beyond their capabilities or not enough to the point of the machine being a waste of expenses, he said.
“You don’t want to use a broom to dig a hole in your yard,” Atkinson said. “The first things we look for in people doing work for us is what we look for in a machine: competence and predictability.”
Rose Anna Rutledge, a counselor and evaluator at the Intensive Treatment Modalities mental health provider in Gainesville, said people build trust on how others appear and behave, not on what they say.
“Following through over time is an essential feature of trust,” she said. “If a person is consistently late then we’re going to struggle with a level of distrust.”
When the human and robot encounter each other in the SecondLife warehouse, the robot will display 1 of 17 combinations of physical and interpersonal qualities.
Sometimes it’ll speak louder or softer. It might lean close or pull away. To gauge how people respond to a robot’s decor, it might appear in heroic red or as a humble janitor.
“There are cognitive aspects of trust that add up consciously. There’s also an emotional side,” Atkinson said. “For example, ‘Do I like the color red?’ We’re teasing apart all of these elements.”
Participants will be random people already using SecondLife. Atkinson may do an additional test in his laboratory with 3-D goggles, heat sensors and vibrations to make the experience more realistic.