Page 433 -

436 M. Neumann

is different: these theories emphasise that a full internalisation of norms is only
realised when they have become part of one’s identity. Thus, internalised norms
form part of the person’s own goals. According to identity theories, agents of this
kind of architecture have not yet fully internalised norms. Norms, implemented in
an ‘obligations’ component, do not represent complete external regulation, but by
the same token, they are not part of the agent’s own desires. In fact, the dichotomy
between obligations and desires becomes only effective once conflicts between both
components arise. This is explicitly wanted: the ‘obligations’ component is added
to the architecture to enable norm compliance as well as violation. It is claimed that
this process preserves the agent’s autonomy. Hence, the dichotomy of obligations
and desires refers to an antagonism between an individual and society. To represent
the process of norm internalisation as described by modern theories, a dynamic
relation between the components ‘obligations’ and ‘desires’ would be required:
contingent on the salience of a norm, elements of the ‘obligations’ component
should be imported to the ‘desires’ component.

17.3.2.3 Abstract Concepts

Agents may face several obligations that may contradict one another. For this reason,
some authors differentiate between norms and obligations. Norms are regarded as
more stable and abstract concepts than mere obligations (Dignum et al. 2002; Conte
and Dignum 2001). One example of such an abstract norm is ‘being altruistic’:
further inference processes are needed for the formation of concrete action goals
from this abstract norm. The striking feature of this approach is to allow for
normative reasoning. This calls for an active role of the agent. This conception of
norms is a precondition for a modelling approach of social behaviour regulation
based on identity conceptions.
In particular, the cognitive capacity of role taking constitutes a crucially impor-
tant step in the development of goals from abstract concepts: that is, the ability to
regard oneself from another’s perspective. Interestingly, steps in this direction can
be found in the AI literature. In Boella and van der Torre’s architecture of a ‘norm-
governed system’ (Boella and van der Torre 2003), the agent’s decision-making
process is governed by the belief that they are observed by other agents and by
the belief that the other agents have expectancies with regard to how they ought to
behave. This can be regarded as a first step in simulating identity theory. However,
from the perspective of socio-psychological identity theories, it is a shortcoming
of this architecture that the agents regard themselves only in terms of the question
concerning whether they fulfil their—externally given—social role. Identity consists
of an inner and outer perspective. The inner perspective is dependent on one’s
personal decisions. This is not the case in this architecture, which consists solely
of an outer perspective. It can be questioned if and how an inner perspective can be
modelled: among other things, the development of an inner perspective is correlated
to a social identity. This social identity, however, is correlated to peer groups and
reference groups. Hence, it refers to social networks, which can be simulated. In
principle, a propensity to take over group norms could be simulated, dependent on