Page 424 -

17 Social Constraint 427

Transformation
There exist a wide range of varieties how agents change their behaviour. Behaviour
transformation is not as straightforward as in game theoretic models. However, it
has to be emphasised that the very first model of Conte and Castelfranchi did not
include any behaviour transformation at all. Agents have no individual freedom in
this model. As critics accuse the role theory, the action repertoire is also (depending
on conditions) deterministic. Thus, even though the authors succeed in ‘bringing
man back in’, the agents in the model are merely normative automata. Insofar as the
norms are a pregiven element in the model, the approach can also be regarded as an
‘over-socialised’ conception of man.
This limitation has been overcome by the subsequent developments. With regard
to the transformation problem, the agents have become more flexible than in the
very first model. However, a key difference to game theoretic models still remains:
while game theoretic models mostly concentrate on sanctioning, in models of
cognitive agents, sanctions are only employed by Flentge et al. as the transformation
mechanism.
However, while the norms in these models can be interpreted as internalised
properties of the agents, an investigation of the process of internalisation is only
in the beginning. So far no commonly accepted mechanism of internalisation has
been identified. Memetic contagion is a candidate. In Verhagen’s model, a quite
sophisticated account is undertaken, including a self-model, a group model and a
degree of autonomy. It is highly advanced in constructing a feedback loop between
individual and collective dynamics. By the combination of a self-model and a group
model, a representation of the (presumed) beliefs held in the society is integrated in
the belief system of individual agents. Conceptually, this is quite close to Mr. Smith.
However, it might be doubted whether the mechanisms applied are a theoretically
valid representation of real processes.

Transmission

Complementary to the wide range of different mechanisms of agent transformation,
also a variety of different transmission mechanisms are applied. Basically, agents
apply some kind of knowledge updating process, if agent transformation takes
place at all. Up to date, the transmission problem is no longer a blind spot of
cognitive agents as it was the case in the (Conte and Castelfranchi 1995a, b) model.
By comparison, communication plays a much more important role than in game
theoretic models and is much more explicitly modelled in models within the AI
tradition. The processes utilised are more realistic mechanisms than the replicator
dynamics of game theoretic models. However, no consensus has been reached, what
an appropriate mechanism would be. This is also due to the fact that a modelling
of agent transformation and norm transmission is computationally more demanding
than in game theoretic models. It has to be emphasised, however, that with regard to
the transformation and the transmission problem, the borderlines of both approaches
are no longer clear-cut. The models of Verhagen and Savarimuthu et al. include
elements of the other line of thought.