IPAB workshop - Michael Herrmann

The Feelings of Robots: A formal approach

Human-robot interaction is known to benefit from the recognition of emotional states of humans as well as from the display of artificial emotions by the robot. As machine intelligence is often considered as being implied by an optimisation process for a given cost function, and thus undisturbed by emotional effects, it is still doubtful whether robots can have or may need intrinsic feelings. We argue that the concept of a utility function is too narrow to represent the incentives for an autonomous acquisition of meaningful skills in a complex environment, and that artificial quasi-emotions can be derived from the identification of insufficiencies of a utility-based approach. In particular, robots can perform better in the Pareto sense in a situation where several goals are available. Through a series of scenarios, we demonstrate the usefulness of artificial emotions which may be critical for the organisation of behaviour in robots, for producing credible emotional expressions in human-robot interaction, and ultimately for the understanding of human emotions.