5.5.4 Generating Shared Team Beliefs
After the team-agent has added into its contents all the agents required
for the achievement of its initial commitment, the team-agent
may try to build the appropriate mutual beliefs among the team members.
The team-agent achieves this by sending appropriate messages to
its Contents, for example:
(5.9)
Bi
⎛
⎜⎜⎝
¬doing(i, book holiday)∧
capable(i, book hotel)∧
capable(i, book flight)∧
capable(i, rent car)
⎞
⎟⎟⎠
⇒
_
Tbelief (i, doing(i, book holiday))∧
Tbelief (i,♦done(book holiday))
_
(5.10)
Bidoing(i, book holiday)
⇒
⎛
⎝
Tbelief (i, doing(i, book hotel))∧
Tbelief (i, doing(i, book flight))∧
Tbelief (i, doing(i, rent car))
⎞
⎠
Now, when an agent receives a Tbelief message, it believes that the team
is undertaking the activity specified. This allows the agent to act, knowing
that certain activities are being performed elsewhere.
5.5.5 Team Computation
The team-members (the team-agent’s Contents), in order to be able to
start the execution of their particular part of the team’s joint task, need
to accommodate (something like) the following rule:
(5.11)
∀ x ∈ Contexts :
⎛
⎝
Tbelief (x, doing(σ))
∧ capable(i, σ)
∧ Bi(¬doing(i, σ)
⎞
⎠
⇒ request(i, σ) ∧ Bidoing(i, σ)
The team’s execution terminates either when all the individual tasks,
required for the achievement of the team-agent’s initial goal have been achieved, or when a particular team-member that plays a critical role
within the team drops its commitment to carry out the task that the team
has delegated to it.
In the first case, the appropriate rules that terminate the team are as
follows:
(5.12)
⎛
⎝
done(book flight)
∧ done(book hotel)
∧ done(rent car)
⎞
⎠ ∧ Bi¬done(book holiday)
⇒ Tbelief (i, done(book holiday))
In the second case we can either incorporate specific rules within the
team-agent’s description that will try to recover the situation or we can
simply terminate the team.
5.6 Summary
In this chapter, we have provided a review of work on executable specifications
based upon temporal and modal logics. In particular, we have
shown how such an approach might be productively used in the development
of simple organizational structures, such as teams.
5.6.1 Related Work
Teams are often viewed as groups of agents requiring their members
at least to have a joint goal and, possibly, to share some specific mental
properties such as intentions, beliefs and commitments, for example
[2, 21, 28]. Typically, an agent team is a group of agents working on
a common task having the ability to operate on dynamic environments
incorporating flexible coordination and communication facilities.
Simple groups of individual agents with prespecified coordination
plans in a very dynamic and continuously evolving domain will fail to
deliver due to inflexibility [28]. This is mainly because of the fact that it
is difficult to anticipate, and plan in advance for, all possible coordination
failures. Cohen and Levesque developed a framework for constructive
agent teamwork, taking into account the fact that teamwork is not
just the union of simultaneous actions carried out by the team’s members,
even if these actions are coordinated [2]. In their framework, they claim
that for flexible teamwork in real-time and highly dynamic domains, the
teams’ members have to share some certain mental features. They argue
that the mental notion of joint intention, which they formulate as a joint
commitment to perform a collective action while in a certain mental state,
serves as the glue that binds teams members together.
However, joint intention theory assumes that it is always possible to
attain mutual belief, and that once an agent comes to think the goal is
finished, it never changes its mind. Unfortunately, this may not always
be the case.
Another researcher who tried to addresses issues related with agent
teamwork is Tambe. Tambe’s model of teamwork is based on the notion
of hierarchical reactive plans and on a modified version of the joint intentions
theory [28] . The framework has been successfully applied in the
modeling of a soccer team [22], as well as in the modeling of intelligent pilots
in a battlefield simulator [28]. Tambe modifies joint intention theory
by: making communication conditional, rather than unconditional; and
enhancing mechanisms for dissolution of a joint commitment.
An attempt to model teamwork and, more generally, the cooperation
of agents for solving problems, is described by Wooldridge and Jennings
in [31]. The model represents the cooperation process from its beginning,
with some agent recognizing the potential for cooperation with respect to
one of its goals, through to team action that leads to the realization of this
goal.
It is important to note that we do not, in this chapter, attempt to model
the ideas of joint intentions/goals/beliefs in their full generality, but use a
combination of a logic of belief (TLBB), broadcast communication, and the
idea of representing groups as agents, in order to enforce common goals.
There is relatively little related work on executable agent specifications.
The main alternative approach is that of Golog [24] and Con-
Golog [17]. These are both based upon the Situation Calculus, in which
the preconditions for, and effects of, actions are specified. These actions
can be applied to situations (states) to generate new situations. In this
way, a sequence of actions can lead from an initial state to a goal state.
The typical way in which this is used in agents is to identify a goal state,
and then generate, via a planning process, a sequence of actions that will
transform the current/initial state into the goal state. This sequence of
actions is then executed by the agent. ConGolog is a multi-agent version
of Golog.
