Design of Social Agents - Foundations on Natural and Artificial Computation - page 193

internal states. The previous move of the partner is considered as the input for
the algorithm. The input and the internal state of the agent determine its output
(the next move) that can be either cooperation or defection. The mapping of the
inputs and internal states into the output is given by a fixed matrix. Additionally
to the generation of the output the agent changes its internal states. The new
internal state, like the output, is determined by the input and previous internal
state. The mapping of the input and the current internal states into the new
internal states is also given by a fixed matrix.
2.3 Design Principles
The maximization of the personal benefit of an agent in a sequence of interac-
tions with a random partner is the main design criterion that we used for the
construction of the social computer agents. In other words, our intention is to
create an egoistic agent whose the only goal is to maximize its own benefit in the
environment populated by other agents whose playing strategies are not known.
In the prisoners dilemma the defection is always the most beneficial choice
independently on what choice, defection or cooperation, is made by the partner.
As a consequence, the agents, that are rational in the classical game theory sense,
always choose to defect. In other words, the rational agents demonstrate mutual
defection that is known as Nash equilibrium for the prisoners dilemma. Further
on we will call the agents of this kind as ”defective” agent.
The permanent defection is one of the two simplest possible strategies and the
permanent cooperation is the second one. We will call the agents that perma-
nently cooperate as ”cooperative” agents. This strategy can also be considered
as a rational one since the mutual cooperation is more beneficial than the mutual
defection.
However, it has to be emphasized that the cooperation is beneficial only if
it is reciprocal. The cooperation with an agent that defects can be considered
as an exploitation of the cooperative agent by the defective one. This kind of
interaction is very beneficial to the exploiting agent and very unbeneficial to the
exploited one. Thus the cooperative agent cannot be considered as a successful
solution for the environment populated by agents of different kinds.
To prevent an exploitation by a partner the agent should exhibit a reciprocal
behavior (defect with the defective agents and cooperate with the cooperative
agent). We will call the agents of this type as ”reciprocal” agents. The reciprocal
agents can be considered as more successful than the above introduced defec-
tive and cooperative agents. In contrast to the cooperative agents, the reciprocal
agents cannot be exploited by the defective agents since they will behave defec-
tively with the defective agents. In contrast to the defective agents, the reciprocal
agents will establish and maintain cooperation with each other if they are con-
structed to start from the cooperative move. The weak point of the reciprocal
agents is that they are not able to exploit cooperative agents. In other words, the
reciprocal agent cooperates with its partner even if it is not a necessary condition
for the cooperation of the partner. This kind of behavior is in a disagreement