Information Technology Reference
In-Depth Information
Fig. 6.4
Agent 'Jan' starts
with no knowledge of the
correct hypothesis. Two
major competing hypotheses
are shown (
medium red
and
medium dark blue
). Overall
confidence for agent 'Jan'
falls as new information is
obtained then rises. Notice
how the probability of
selecting more informative
experiments increases
steadily from cycle 10
actions (e.g. A, 7), reducing their probability to a very low value. There is a corre-
sponding increase in the probability of selecting the best moves (e.g. D, 4). Figure
6.4
also illustrates the effect of biasing an agent against the correct rule by initiating a
run with Jan's confidence in the correct rule set at zero. This makes a different set of
actions more probable, so that the agent Jan is less likely to discover the correct rule
in the early stages. However, agent 'Jan' gets to know about the 'correct' hypothesis
from fellow agents. As soon as this new hypothesis is recognised Jan's confidence
increases, different cards are selected and the 'new' idea takes rapid hold.
6.12
Belief and Truth
If agents are prevented from consulting, they take longer to arrive at a correct out-
come than agents that combine experiments with consultation. This endorses the
post-Kuhnian view that science is an inherently social process. Consultation also
