Page 103 - Socially Intelligent Agents Creating Relationships with Computers and Robots

86 Socially Intelligent Agents

the human perception of intelligence. The comparison of computers’ sociality
to the human one is sharp - one can see that humans are able to get involved in
very complex behaviours when interacting, even when they have only a minimal
amount of conscious data about the others.
If the final aim is to engender some social intelligence in artificial agents, the
usual approaches (e.g. trying to limit misunderstandings in the interpretation of
messages exchanged or by reducing uncertainty in social organisation) might
not be the best way forward. These approaches do not contradict that aim, in
fact they can be useful, but they will not be sufficient.

2. Socially Intelligence In MAS

The problem of organising interactions among computers has grown dras-
tically since the spread of PCs to a wider public, and more precisely with the
ever-wider adoption of computer networks. The field of Multi-Agent Systems
that has developed in the last years was originally led by the idea of having
several artificial intelligences accomplish tasks together. This is often seen as
a continuation of both Artificial Intelligence and Distributed Artificial Intelli-
gence, and researchers usually look for ways of building societies of artificial
intelligences [12].
There exists a commonly accepted definition of a MAS: it is constituted of
an environment that evolves in time in which autonomous entities (agents) are
able to act; they can also interact; the overall organisation is effected by rules
that coordinate time evolution and all actions of the entities. Four axes can thus
be emphasised for such a systems, (called the AEIO model): the exploration
of what an Agent should be (mainly in terms of cognition and perception),
what characteristics the Environment has, which kind of Interactions can exist
between the entities, and what kind of Organisation is needed to put these
different elements together [2].
The reasons why people use MAS are diverse. Some need to coordinate
computers, as in networks, some need to coherently integrate different expert
systems, and some need to make robots act together. In most cases, it is common
to consider that the reliability of the system, its predictability and its efficiency
are what is most important. In order to coordinate agents so that they work
together without interfering with each other it is common to use a set of norms.
In that case agents have clear roles that are interdependent and a known set
of acquaintances with complementary roles (for example [5]. In this case,
communication is mainly used by agents to make requests for help and to
accept tasks that are proposed to them. Since tasks are necessarily thought as
collective, the social life of agents is very important for the global system. They
are thus designedwiththis ability tointeract, either bybeingcommittedtoothers