Page 107 - Socially Intelligent Agents Creating Relationships with Computers and Robots

90 Socially Intelligent Agents

social situation, which enable them to act coherently at any moment, in the face
of unexpected and complex situations [10].
The idea of creating new communication groups and norms from existing
ones has been used for quite a long time: for example, by [17] and finding ways
of having new institutions emerge is often emphasised (e.g. [14]. Although
successful, these attempts still require very little creativity from the agents in
their social life, since this ability would require the involvement of a meta-
language.
It is obvious that the creativity that is required is not easy to organise. Just one
element, inspiredbytheobservationof humansocieties, couldhelpinimproving
it. As we saw, humans use old rules to create new ways of communicating by
putting them in a new context. It is true in everyday life: some power or
trust relations that are expected to appear in intimate relations can sometimes
be recognised at work. Even without much invention, this overlap between
different fields of relation can eventually engender new institutions: the change
of context forces individuals to explain to the others the meaning of the new
rule they try to use and thus create a new relational system [3]. For the moment,
software systems are built in a way that makes them very specialised and quite
independent from each other. More generally, when agents communicate, it
is mainly about single topics, rarely dealing about different type of tasks with
different type of acquaintances. Maybe making agents more generic would
be the first step to enable them to transpose their knowledge of interactions
between different contexts so that the result would be more creative.

4. Conclusion
The idea of social intelligence can be summarised by two main requirements:
the ability to be able to exchange quite complex information to undertake so-
phisticated tasks in a reliable way, but also the ability to open the system so that
new rules of communication can be created with new type of agents when they
are recognised as valuable for these interactions.
For the moment these aspects are treated quite independently, and the reason
certainly lies in the final aims of the research. A technological system that has
to be reliable cannot afford to address the questions of the self-consciousness
of artificial agents: such approaches have not been proven to lead to predictable
results (and one could even anticipate the opposite effect). This is why no one
tries to design social intelligence of entities when these are used to solve precise
problems.
Lots of researchers agree on the fact that social intelligence can appear only
if agents can recognise who is valuable for interaction, and more precisely if
they are willing to communicate with others even if they receive messages that
are not clear right away. But in that case, one has to enable them to imagine