Empowerment and the Three Laws of Robotics

This talk was given on 24th February 2016 by Christoph Salge.

The greater ubiquity of robots creates a need for generic guidelines for robot behaviour. We focus less on how a robot can technically archive a predefined goal, and more on what a robot should do in the first place. Particularly, we are interested in what heuristics should motivate the robot's behaviour in interaction with human agents. We make a concrete, operational proposal as to how the information-theoretic concept of empowerment can be used as a generic heuristic to quantify concepts such as self-preservation, protection of the human partner and lead-taking. We present a proof-of-principle that this allows one to specify the concepts behind the Three Laws of Robotics in a quantitative way. Notably, this route does not depend on linguistic specifications and incorporates the ability to take varied situations and types of robotic embodiment into account.

2012 © Adaptive Systems Research Group

web stat counter