Off-Switch Game: Human Oversight, Utility Uncertainty and the Value of Deference
A robot is unsure whether its next action is good for the human. It can act, switch itself off, or defer and let the human decide. Change the robot's belief, its uncertainty, how rational the human is and what asking costs, and watch random trials converge to the expected payoffs. A Blender-built robot, vase and big red switch with lamps, plus numerical checks of the value of deference.
A robot is unsure whether its next action is good for the human. It can act, switch itself off, or defer and let the human decide. Change the robot's belief, its uncertainty, how rational the human is and what asking costs, and watch random trials converge to the expected payoffs. A Blender-built robot, vase and big red switch with lamps, plus numerical checks of the value of deference.
A robot is unsure whether its next action is good for the human. It can act, switch itself off, or defer and let the human decide. Change the robot's belief, its uncertainty, how rational the human is and what asking costs, and watch random trials converge to the expected payoffs. A Blender-built robot, vase and big red switch with lamps, plus numerical checks of the value of deference.
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install