T.S. Bhatia, S.A. Khan, and L. Bölöni

The education of a crook: reinforcement learning in social-cultural settings


Cite as:

T.S. Bhatia, S.A. Khan, and L. Bölöni. The education of a crook: reinforcement learning in social-cultural settings. In Proc. of Int'l Conf. on Autonomous Agents and Multi Agent Systems (AAMAS-2014), pp. 1397–1398, May 2014.

Download:

(unavailable)

Abstract:

The ability to manipulate social and cultural values in order to achieve one's own goals is a hard-to-teach but profitable skill. In this paper we represent a complex social scenario, the Spanish Steps flower selling scam, using a social calculus framework based on culture sanctioned social metrics (CSSMs) and concrete beliefs (CBs). Then, we show how a crooked seller can learn a profitable strategy through reinforcement learning. Although the search space defined by the social calculus is large, we found that function approximation based Q-learning allows us to successfully learn efficient strategies in a relatively small number of runs. The learned strategy allows the seller to manipulate an unprepared tourist's social values of politeness and dignity, as well as his perception of the peers and crowds opinion. This allows the seller to manipulate some of his opponents to act against their own interests by purchasing an overpriced flower while well-knowing that they are being cheated.

BibTeX:

@inproceedings{TSBhatia-2014-AAMAS,
   title = "The education of a crook: reinforcement learning in social-cultural settings",
   author = "T.S. Bhatia and S.A. Khan and L. B{\"o}l{\"o}ni",
   booktitle = "Proc. of Int'l Conf. on Autonomous Agents and Multi Agent Systems (AAMAS-2014)",
   year = "2014",
   pages = "1397-1398",
   month = "May",
   abstract = {
   The ability to manipulate social and cultural values in order to achieve one's own goals is a hard-to-teach but profitable skill. In this paper we represent a complex social scenario, the Spanish Steps flower selling scam, using a social calculus framework based on culture sanctioned social metrics (CSSMs) and concrete beliefs (CBs). Then, we show how a crooked seller can learn a profitable strategy through reinforcement learning. Although the search space defined by the social calculus is large, we found that function approximation based Q-learning allows us to successfully learn efficient strategies in a relatively small number of runs. The learned strategy allows the seller to manipulate an unprepared tourist's social values of politeness and dignity, as well as his perception of the peers and crowds opinion. This allows the seller to manipulate some of his opponents to act against their own interests by purchasing an overpriced flower while well-knowing that they are being cheated.
},
}

Generated by bib2html.pl (written by Patrick Riley, Lotzi Boloni ) on Fri Oct 06, 2017 18:15:24