Design of Social Agents - Foundations on Natural and Artificial Computation - page 194

with the above introduced goal to find a behavior that maximizes the personal
benefit of the agent.
The above described requirements can be summarized in the following list of
the social skills that have to be embedded into the social agents:
- To avoid the exploitation by the partner, the social agent should be able to
reciprocate defection of the partner (react defectively on the defection).
- To maintain the cooperation, the social agents should be able to reciprocate
the cooperation of the partner (react cooperatively on the cooperation).
- To establish the cooperation, the social agents should be able to propose the
cooperation in the phase of the mutual defection (react cooperatively on the
defection).
- To exploit the partner, the social agents should be able to try defection in
the phase of the mutual cooperation (react defectively on the cooperation).
As we can see, the above described requirements can be competing. So, the
agent needs to be able to decide what logic to apply at the current stage of
the interaction. Moreover, the agent needs to remember the responses of the
partner. In more details, it should not keep trying to establish a cooperation if
all these attempts fail. Similarly, the agent should not continuously try to exploit
its partner if it sees that these attempts are not successful. On the other hand,
the agent should not cooperate if it is not required for the cooperation of the
partner.
Fig. 1. Schema of the constructed social agent
We implement the above described design principles by the following algo-
rithm. The agents start cooperation by proposing cooperative move as a response
on the defective move of the partner. Doing that agent switches to the state of
cooperation ( CN in the Fig.1). If partner answers defectively on cooperation ini-
tiated by the agent, the agents switches to the cooperation-failed ( CF n ) states.
In this state the agents never initiates a new cooperation. If partner starts coop-
eration, the agent will defect, immediately or after several steps of cooperation,
to check if its partner can be exploited. In more details, for several steps agent