Information Technology Reference
In-Depth Information
6.7
Agents and Groups
The aim of every agent is to raise its general confidence in some view of the world by
discovering facts about the world. A group of agents has a similar aim; to obtain new
information to minimize their collective uncertainty about the world. For a collection
of agents a similar group-confidence value can be derived from the Group Entropy:
A
H
n
(H/A)
Entropy(Grp)
=
∗{
Log
2
(E
n
(H
/
A))
−
Log(
|
A
|
)
}
/
|
A
|
This expression represents the expected result of a single sampling of the confidence
in any hypothesis of an agent randomly chosen from the group.
Similarly, the inverse of the group entropy represents the significance threshold
for the group of agents in terms of an expected probability. We can also calculate
the inverses of entropy for an agent I(A), a group of agents I(G) and for a set of
experiments I(E(A)) as perceived by a single agent or a group of agents I(E(G)),
where an experiment 'e' is defined in terms of the probabilities of results R that are
considered possible for a given range of hypotheses H.
These dynamic threshold values are independent of the number of hypotheses or
experiments, so we can define the indifference level of a group of actors indepen-
dently of particular hypotheses that happen to be in play. This is important for two
reasons. It allows us to use these dynamic values in agent-based decision-making
about the next action to take (see Sect. 6.6). It also allows us to represent the pro-
cesses whereby scientists respond to changing evidence by altering their view of the
world, e.g. a hypothesis changes from being a mere possibility, to being considered
plausible, to being generally accepted and, finally, coming to have the status of a
fact, law or principle. If all hypotheses were to be eliminated except one, both I(A)
and E(A) would equal unity (implying certainty). If this happens, it can be said of
the agents that they are both indifferent to the hypothesis as well as certain of it.
6.8
The Choice of an Action
6.8.1
Evaluating Actions
Agents decide whether to experiment or to consult by evaluating each of the possi-
bilities offered by each kind of action. We represent an experimental setup as a table
of real numbers that indicate the a priori probability of a result occurring, given that
a hypothesis (or model) is a correct description of the world's constraints. We refer
to the hypothesis that is active within a simulation run as the objective model. Sim-
ulated experiments have those outcomes that are most probable in a world in which
the objective model is true (see Gooding and Addis
1999
). This list of occurrence
probabilities defines each possible experiment. The list will sum to unity for each
hypothesis, since at least one of the results must occur in a world of which that model
is the best available description. Experiments differ in producing different sets of re-
sults, so where the set of occurrence probabilities for an experiment is the same for
any hypotheses, it is the same experiment, regardless of the physical apparatus used.
