Design of Social Agents - Foundations on Natural and Artificial Computation - page 197

1.4

1.2

1

0.8

0.6

0.4

0.2

0

1 1.5 2 2.5 3 3.5 4

Score
Fig. 2. Distribution of the performances of the random agents

However, we have to note that the performances calculated in this way should
depend on the way the pool is populated. In our study we used a procedure in
which any two agents, that have the same number of the internal states, have
the same chance to be put into the pool. Moreover, the probability of an agent
to be added to the pool does not depend on the number of the internal states. In
other words we got a ”uniform” distribution of the agents in the pool and, as a
consequence, the performance of agents, calculated in this way, can be considered
as more ”representative” or ”universal”.
However, we need to remember that the assumption of the homogeneous dis-
tribution of the agents in the pool is not always valid. In particular, the pool
can be populated by artificial agents that were constructed with the intention to
increase the personal performance. Or, as another example, the content of the
pool can be a product of an evolution. In both cases the chances of an agent to
be in the pool are higher if its performance is better.
To overcome the problem of the dependency of the score on the way the pool
of partners was generated, we propose to compare performances of two agents by
letting them to interact with each other. In more details, we want to model perfor-
mances of two different agents in the pool that is equally populated by the agents
of the two considered types. To model this situation we do not need to use a larger
pool because the interactions of a given agent with the partners of the same type
will always give the same score per game. The agent of a chosen type needs to inter-
act with one partner of the same type and one partner of another type. Interactions
with partner of two different types of agents gives two values of the performance
that are then averaged to get the performance of the given agent in the considered
situation. The performances of the two agents, calculated in this way, can be in-
terpreted as their abilities to survive, in the evolutional sense, in the pool that is
equally populated by the agents of the two considered types. We would like to
emphasize that the performances of agents, calculated in the above described
way, are not transitive. If agent A outperforms agent B and agent B , in its turn,
outperforms agent C , it does not necessarily means that agent A outperforms