Requirements as Goals and Commitments Too - Intentional Perspectives on Information Systems Engineering - page 138

2.1 Multiagent Systems
Multiagent Systems (MAS) are open systems : autonomous and heterogeneous enti-
ties known as agents participate in multiagent systems. An agent's autonomy means
that no agent has control over it. An agent's heterogeneity means that an agent's
internal construction is inaccessible to other agents. An agent may be a human,
organization, or some stakeholder projected into the system as software. It is
worth emphasizing that socio-technical systems are, first and foremost, multiagent
systems.
The purpose of the system , specifically, is to provide a basis for coherent interac-
tions among agents in spite of their autonomy. Indeed, the system may be specified
independently of the agents [37] . The system itself serves as the specification, from
a global perspective, of the legitimate expectations that agents adopting roles in the
system would have of each other. In other words, the system is the protocol (MAS
terminology), or specification (RE terminology).
We specify expectations in terms of commitments. An agent that does not ful-
fill its commitments to others is noncompliant. Compliance balances autonomy.
An agent may do as it pleases, but from the system's perspective it may be
noncompliant. Example 1 illustrates these concepts.
Example 1. A housing contract is a system that specifies the commitments that gov-
ern interaction between a tenant and the landlord, both agents. For example, the
contract may say that the tenant may not accommodate other persons on the property
unless he seeks permission from the landlord. However, the tenant, in noncompli-
ance with the clause, may on occasion host visiting family members. It does not
matter whether the landlord knows of the violation; what matters is that from the
system perspective, there is a violation.
The question of the basis of compliance goes to the heart of multiagent sys-
tems research. The answer lies in how systems (protocols) are specified. Systems
specified in terms of control and data flow impose strong ordering and synchroniza-
tion constraints on interaction; compliance for such specifications amounts to not
violating such constraints, as Example 2 shows.
Example 2. Consider a scenario where Alice wants to buy a book from the bookseller
EBook. The protocol (the system) they employ specifies that the delivery of the
book must precede payment. If Alice pays first, she would be noncompliant with
the protocol.
Systems specified in terms of intentional abstractions such as goals and beliefs
are brittle because they lead to strong assumptions about an agent's construction
[ 34] .
By contrast, system specification approaches based on commitments hit the right
balance between over-abstraction (exemplified by goal-oriented approaches) and
under-abstraction (exemplified by process-oriented ones). Goal-oriented approaches
model desired states of the world without saying who is responsible for doing