Page 276 - Socially Intelligent Agents Creating Relationships with Computers and Robots

Chapter 32

ENABLING OPEN AGENT INSTITUTIONS

1
Juan A. Rodríguez-Aguilar and Carles Sierra 2
1
iSOCO (Intelligent Software Components)
2
Institut d’Investigació en Intel.ligència Artificial, Spanish Scientific Research Council (CSIC)

Abstract In this paper we argue that open multi-agent systems can be effectively designed
and implemented as electronic institutions composed of a vast number of het-
erogeneous (human and software) agents playing different roles and interacting
by means of speech acts. Thus taking inspiration from traditional human insti-
tutions, we offer a general agent-mediated computational model of institutions
that serves to realise an actual agent-mediated electronic auction house where
heterogeneous agents can trade.

1. Introduction

Up to date most of the work produced by multi-agent systems(MAS) re-
search has focused on systems developed and enacted under centralised con-
trol. Thus, MAS researchers have bargained for well-behaved agents immersed
in reliable infrastructures in relatively simple domains. Such assumptions are
not valid when considering open systems [3] whose components are unknown
beforehand, can change over time and can be both human and software agents
developed by different parties. Examples of open agent systems include open
electronic marketplaces and virtual supply chains, disaster recovery operations,
collaborative design and international coalition military forces.
Although open systems have recently started to be considered by MAS re-
searchers as one the most important application of multi-agent systems, their
inherent issues (agent heterogeneity, reliability, accountability, legitimacy, so-
cietal change, etc.) have not been conveniently addressed yet. And then how
to approach their design and construction? Although there has been a surge
of interest in agent-oriented methodologies and modelling techniques in the
last few years motivated and spurred by the first generation of agent develop-
ments [7, 8, 11], at present most agent applications lack a principled method-