Page 431 -

434 M. Neumann

Conte and Dignum (2001) argue that imitation is not sufficient to establish a
cognitive representation of norms in an agent. Agents infer abstract standards
from observed behaviour. This allows for normative reasoning and normative
influence in accepting (or defeating) and defending norms.
Dignum et al. (2002) investigate the relations and possible conflicts between differ-
ent components in an agent’s decision process. The decision-making process of
so-called B-doing agents is designed as a two-stage process, including norms as
desires of society. The authors differentiate between abstract norms and concrete
obligations.
Garcia-Camino et al. (2006) introduce norms as constraints for regulating the rules
of interaction between agents in situations such as a Dutch auction protocol.
These norms are regulated by an electronic institution (a virtual auctioneer) with
an explicitly represented normative layer.
Lopez and Marquez (2004) explore the process of adopting or rejecting a normative
goal in the BDI framework. Agents must recognise themselves as addressees of
norms and must evaluate whether a normative goal has a higher or lower priority
than those hindered by punishment for violating the norm.
Sadri et al. (2006) extend their concept of knowledge, goals and plan (KGP) agents
by including norms based on the roles played by the agents. For this reason, the
knowledge base KB of agents is upgraded by KB soc , which caters for normative
reasoning, and KB rev , which resolves conflicts between personal and social goals.
Shoham and Tennenholtz (1992) propose building social laws into the action
representation to guarantee the successful coexistence of multiple programmes
(i.e. agents) and programmers. Norms are constraints on individual freedom. The
authors investigate the problem of automatically deriving social laws that enable
the execution of each agent’s action plans in the agent system.
Vazquez-Salceda et al. (2005) provide a framework for the normative regulation
of electronic institutions. Norms are instantiated and controlled by a central
institution, which must consist of a means to detect norm violation and a means
to sanction norm violators and repair the system.

How can these examples be evaluated with regard to the design decision
suggested by socialisation research? The existing approaches can be regarded as
a hierarchy of increasingly sophisticated accounts, ranging from mere constraints to
abstract concepts. Broadly speaking, three concepts of norms can be differentiated:
norms as constraints (the simplest choice), as obligations or as abstract concepts
(the most sophisticated choice). This is summarised Table 17.3.

17.3.2.1 Constraints
The simplest and most straightforward way is to regard norms as mere constraints
on the behaviour of individual agents. For example, the norm to drive on the right-
hand side of the road restricts individual freedom. In this case, norms need not
necessarily be recognised as such. They can be implemented off-line or can emerge