Requirements as Goals and Commitments Too - Intentional Perspectives on Information Systems Engineering - page 145

Note the usage of the words support and potentially . Goal (commitment) support
is a weaker notion than fulfillment; support gives no guarantee about fulfillment at
runtime. And yet, it is a more pragmatic notion for open systems, where it is not
possible to make such guarantees anyway. For instance, a commitment that an agent
depends upon to fulfill his goal may be violated.
Goal support We illustrate the basic intuitions with examples.
An agent's goal is supported if the agent has a capability for it (Example 8).

Example 8. Consider Alice's goal payment. Alice supports the goal if she has a
capability for it.

An agent's goal is supported if it can get an appropriate commitment from some
other agent about the state of affairs that the goal represents (Example 9).

Example 9. Consider Alice's goal BNW . The commitment C ( merchant, Alice,
payment, BNW ) from some merchant supports the goal, but only if Alice sup-
ports payment. The intuition is that Alice won't be able to exploit the merchant's
commitment unless she pays.

An agent's goal is supported if it can make a commitment to some other agent
for some state of affairs (presumably one that the latter would be interested in) if the
latter brings about the state of affairs that the goal represents (Example 10).

Example 10. Consider EBook's goal payment . He can support this goal by making
an offer to some customer, that is, by creating C(EBook, customer, payment,
BNW) .

The intuitions may be applied recursively for decomposition in goal trees. Thus
for example, if an agent wants to support g , and g is and-composed into g 0 and g 1 ,
then the agent would want to verify support for both g 0 and g 1 , and so on.
Commitment support It makes sense to check whether an agent will be able to
support the commitments it undertakes towards others.
Commitment support reduces to goal support for the commitment consequent
(Example 11).

Example 11. Consider that C(EBook, customer, payment, BNW) holds. EBook
supports his goal payment by the commitment; however, if he does not support
BNW , then if the customer pays, he risks being noncompliant.

We consider goal and commitment support as separate notions. A reckless or
malicious agent may only care that his goals are supported regardless of whether his
commitments are supported; a prudent agent on the other hand would ensure that
the commitments are also supported.
Reasoning for support as described above offers interesting possibilities. Some
examples: (i) [ Chaining ] x can reason that C ( x , y , g 0 , g 1 ) is supported by C ( z , x , g 2 ,
g 1 )if x supports g 2 ; (ii) [ Division of labor ] x can support a conjunctive goal g 0 ∧ g 1

by getting commitments for g 0 and g 1 from two different agents; (iii) [ Redundancy ]
to support g , x may get commitments for g from two different agents; and so on.