Page 104 - Socially Intelligent Agents Creating Relationships with Computers and Robots

Social Intelligence for Computers 87

[20], or by having intentions towards the others [11]. In such an organisation,
agents can choose whether or not to cooperate with the others but have no ability
to make choices about their social life by, for example, choosing with whom
they can interact, and on which topic. This limitation is often justified by the
need for stability.
However, for some people it is important that relations can evolve over time,
since not all agents in the network are reliable [18]. Some of these systems are
designed so that the relations between agents evolve and agents build represen-
tations about their network. Whenever an agent engages in some work for or
with another, both can change their point of view on the relation, and each one
can decide whether to stop the interaction or reinforce some weaker link.
This form of learning could help in addressing the question of social order
in a deeper way, but it is usually used in networks where the agents already
know the whole group at the beginning or can get in touch with any other.
Thus these systems don’t display one of the main characteristics of social life,
which is openness. This property is the ability to accept newcomers into a
group, to have them be integrated into the usual activities and judged as the
others are. This openness has often recognised as a very important question for
MAS [4]. Some systems were hence designed with the aim of dealing with that
openness. MadKit [13] is one of them: in this system agents are divided into
communication groups. Most of the agents belong to just one group and are not
necessarily aware that there exist other agents with whom they don’t interact.
Communication between groups is very important and is done by representative
agents; these also receive requests from agents who are outside the group and
ask for entry. If accepted, the requesting agent is allowed to be a normal agent
among the others.
Although being quite open, the organisation is necessarily based on strong
assumptions about the agents and their ability to interact. Agents still have
predefined goals, the idea that they can ask for help is already implemented,
and they express their needs to the others in a direct way (i.e they know with
whom they are to communicate and how to interpret the messages from them).
This implies at least a common language and similar cognitive processes. This
reduces the flexibility of the evolving social life.
There is a different approach in MAS - one that draws its inspiration from
societies of simple animals, rather than using linguistic metaphors of commu-
nication. This choice of bottom-up perspective is often referred to as "eco-
resolution" and can exhibit quite complex social patterns that can be seriously
validated [9]. It is possible to build systems where each agent has no global
knowledge, communication between agents is not direct but takes place through
the environment and the fulfilment of tasks mainly relies on self-organising
properties of the system. Each agent acts with reflexes that are provoked by the
discovery of stimuli in the environment, and the intensity and position of these