Requirements as Goals and Commitments Too - Intentional Perspectives on Information Systems Engineering - page 144

• Agents may be specified in terms of abstractions such as goals, capabilities, and

so on.

• Goals may also be supported in middleware: an agent can monitor its goals and

act in order to achieve them.

Goals and commitments are complementary. An agent has certain goals that it
wants to satisfy, and in doing so it typically must make (to others) or get (from oth-
ers) commitments about certain goals. Alternatively, an agent has commitments to
others (and a goal to comply), and it then adopts specific goals in order to discharge
its commitments.
Thus, there are two things that an agent designer or the agent itself, by

introspection at runtime, may do.
First, an agent may induce a protocol—the set of commitments—that are nec-
essary to supports the goals it wants to achieve. The agent would additionally
publish the protocol along with the role it has adopted in the protocol, and possibly
invite others to adopt the other roles in the protocol or just wait to be discovered.
Example 7 illustrates this method.

Example 7. Alice has the goal BNW . Alice figures that to get the topic, it
must interact with a bookseller and pay the bookseller for the topic. So Alice
induces a protocol with two roles, customer and merchant , with the commitment
C ( customer, merchant, BNW, payment). She adopts customer , and publishes the
protocol as her interface. Eventually, a seller may sell BNW to Alice by playing role
merchant.

Second, an agent may select a protocol from a repository. This recognizes the
fact that protocols are reusable specifications of interaction [ 14] . Indeed, this is the
case with many standardized protocols such as for financial transactions [ 12] . An
agent would naturally want to verify if a protocol selected from some repository
were suitable for the achievement of the agent's goals. The agent would also want
to verify that if he makes a certain commitment, then his goals support fulfillment
of the commitment.
The notion of compliance with a protocol helps decouple one agent's specifi-
cation from another agent's. For example, a merchant would only care (perhaps
modulo other properties deriving from interaction such as trust and reputation) that
Alice is committed to payment for the topic, irrespective of whether Alice actually
intends to pay. In other words, if an agent commits to another for something, from
the perspective of the latter, it does not matter much what the former's goals are or
how the former will act to bring about the goal he committed to.
We now sketch some elements of the reasoning one can perform with goals
and commitments. Given some role in a protocol and some goal that the agent
wants to achieve, goal support verifies whether an agent can potentially achieve
his goal by playing that role. Commitment support checks if an agent playing a
role is potentially able to honor the commitments he may make as part of playing
the role.