Explaining Sympathetic Actions
Timotheus Kampik
Juan Carlos Nieves
Helena Lindgren
Umeå University
Agenda
* Motivation
* Goals for Sympathetic Actions
* Explanation Types
* Towards Empirical Assessment
* Ongoing Work
* Future Work
Motivation
* *Bounded rationality* has strong influence on economics, psychology, and AI
* Empirical: Kahneman
* Theoretical: Simon, Rubinstein
* Design agents of bounded rationality in the context of explainability
* Focus on *empathy* (altruism)
Kahneman. Maps of bounded rationality: Psychology for behavioral economics. 2003.
Rubinstein. Modeling bounded rationality. 1998.
Simon. Models of bounded rationality: Empirically grounded economic reason. 1997.
Goals for Sympathetic Actions
* Altruistic/utilitarian preferences
* Establishing or following a norm/encouraging sympathetic actions from others
* Compromising in case no equilibrium strategy exists
Explanation Types
* No Explanation.
* Provide a *clue* (that hints at the favor).
* Explain *that* (a favor is provided).
* Explain *why* (a favor is provided, norm).
* Explain *why* (a favor is provided, consequence).
Towards Empirical Assessment I
* Human-robot ultimatum game
* 6 rounds
* Roles switch every round
* 100 coins to share per round
* Two modes, randomly selected per participant:
*explanations*, *no explanations*
* $n = 19$
Towards Empirical Assessment II
* Robot is rational acceptance, but *sympathetic* in proposals
* #Rejections per game
* #Coins optained by human and agent
* Reported *niceness* of robot (scale: 1-5)
Towards Empirical Assessment III
Towards Empirical Assessment IV
Towards Empirical Assessment V
Towards Empirical Assessment VI
Ongoing Work
* Scale study
=> Port study application to web
* Test different modes
Future Work
* Learning agents that facilitate sympathetic actions
* $n$-agent scenarios
Thank you!
Questions?
*This work was partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.*