Page 430 -

17 Social Constraint 433

It will now be examined how current architectures can be assessed from an empir-
ical perspective. As the examination of current simulation models has revealed, a
comprehension of the effects of the social level on individual agents is far from being
sufficient so far. To provide an outlook of possible future modelling approaches
of the effects of social norms on individual agents, a brief sample of normative
agent architectures will be provided. In fact, the number of conceptually oriented
articles on the architecture of normative agents exceeds the number of existing
models. These architectures study how these processes could be modelled in
principle. Typically, norms in concrete models are less sophisticated than concepts
proposed in formal architectures (Conte and Dignum 2001). The development of
architectures is a kind of requirement analysis: it specifies the essential components
of normative agents. It can be expected that future implementations will be guided
by deliberations that can be found in these architectures. For this reason, a sample
of cases is selected (Neumann 2008a) for a closer examination with regard to
the question of what decisions are made on how to represent effects of norms on
individual agents.

Andrighetto et al. (2007) investigate the process of norm innovation. The behaviour
of an agent may be interpreted by an observing agent as normative if it is marked
as salient in the observer’s normative board. Thus, norm instantiation is regarded
as an inter-agent process.
Boella and van der Torre (2003) differentiate between three types of agents: agents
who are the subject of norms, so-called defender agents, who are responsible
for norm control and a normative authority that has legislative power and that
monitors defender agents.
Boella and van der Torre (2006) rely on John Searle’s notion of institutional
facts (so-called ‘counts-as’ conditionals) to represent social reality in the agent
architecture. A normative base and a ‘counts-as’ component transforms brute
facts into obligations and permissions.
The Belief-Obligation-Intentions-Desire (BOID) architecture (Broersen et al. 2001)
is the classical approach to represent norms in agent architectures. Obligations
are added to the BDI architecture to represent social norms while preserving the
agent’s autonomy. Principles of the resolution of conflicts between the different
components are investigated in the paper.
Boman (1999) proposes the use of supersoft decision theory to characterise real-
time decision-making in the presence of risk and uncertainty. Moreover, agents
can communicate with a normative decision module to act in accordance with
social demands. Norms act as global constraints on individual behaviour.
Castelfranchi et al. (2000) explore the principles of deliberative normative reason-
ing. Agents are able to receive information about norms and society. The data
is processed in a multi-level cognitive architecture. On this basis, norms can be
adopted and used as meta-goals in the agent decision process.
Conte and Castelfranchi (1999) distinguish between a conventionalist (in rational
philosophy) and a prescriptive (in philosophy of law) perspective on norms. A
logical framework is introduced to preserve a weak intuition of the prescriptive
perspective which is capable of integrating the conventionalist intuition.