Page 421 -

424 M. Neumann

updated). Faced with this situation, agents choose the alternative that maximises
their expected utility. However, behaviour change goes not along with goal change.
Agents can do no more than react to different environmental conditions. The
agents’ behaviour is guided strategic adaptation. An active element of normative
orientation in the choice relating to the ends of action cannot be found in a game
theoretic approach. This is simply due to the fact that agents do not possess any
internal mechanism to reflect and eventually change their behaviour, other than
the desire to maximise utility. This point has already been highlighted in Parsons’
critique of ‘utilitarian theories’ of action (Parsons 1937), namely, that the ends
of individual actions are in some way arbitrary. Even though the modelling of
behaviour transformation is the strength of this kind of models, the ends of the
action remain unchanged: the goal is to maximise utility. In this respect, the relation
between the action and the ends of the action remains arbitrary.
However, the very idea of role theory is to provide an answer to the question:
where do ends come from? Parsons’ (and Durkheim’s) answer was the internali-
sation of norms. A corresponding answer to this problem is not supplied in game
theoretical models. This is due to the fact that agents do not act because they want
to obey (or deviate from) a norm. They do not even ‘know’ norms. Even though
the model provides a mechanism for the transformation of the agents, this is not
identical with norm internalisation. This remains beyond the scope of this account.
The agents’ behaviour can only be interpreted as normative from the perspective
of an external observer. Thus, transformation is not identical with internalisation.
While the model provides a mechanism for behaviour transformation, it cannot
capture the process of internalisation. Compared to the classical role theory, this
is a principle limitation of a game theoretical description of the problem situation.

17.2.2.2 Models Utilising Cognitive Agents

This shortcoming calls for cognitively richer agents. For this reason, a sample of
models in the AI tradition will be examined more closely.
Conte and Castelfranchi (1995b) investigate three different populations of food
gathering agents: aggressive, strategic and normative agent populations. Aggres-
sive agents attack ‘eating’ agents, strategic agents attack only weaker agents, and
normative agents obey a finder-keeper norm. The aggregated performance of the
normative population is the best with regard to the degree of aggression, welfare
and equality.
In an extension of the above model, Castelfranchi et al. (1998) study the interaction
of the different agent populations. Interaction leads to a breakdown of the
beneficent effects of norms, which can only be preserved with the introduction
of normative reputation and communication among agents. 1

1
For a more in-depth discussion of this model, the interested reader is referred to the chapter on
reputation (Giardini et al. 2013).