Information Technology Reference
In-Depth Information
Fig. 7.3
Single agent,
flexibility 0.3; 3 hypotheses,
negative bias: sphere 0.45,
pyramid 0.45, cone 0.1;
object in box: cone
CONE
INDIFFERENCE
THRESHOLD
SPHERE
PYRAMID
Fig. 7.4
The effect on belief
of maximum responsiveness
to new data. Single agent, 2
hypotheses, negative bias:
sphere 0.8, cone 0.2;
flexibility 1.0, object in box:
cone
SPHERE
INDIFFERENCE
THRESHOLD
CONE
SPHERE
each combination of
<
hypothesis
procedure
>
. The 'belief' (action prob-
ability) for each hypothesis adds up to 1.0 since the agent must act even if the action
is 'do nothing'. This representation treats an experiment as mediating between four
sets of objects: hypotheses, procedures, physical setups and observable outcomes. In
selecting an experiment our agents will prefer those having more decisive outcomes.
However, such an agent does not 'know' the specific outcome of any experiment
beforehand. When the agent selects an experiment, its results in our simulation are
generated 'Monte Carlo' fashion according to the probability distribution of the out-
comes that are possible for each hypothesis (see, for example, Tables
7.7
and
7.8
).
+
setup
+
