Information Technology Reference
In-Depth Information
diculties to constrain the experiment by having a control over one or several
individuals.
We propose to use a multi-agent system approach for the purpose of creating
an agent-based support tool for team collaboration as a first step towards pre-
diction of conflicts in small teams of people. We design the agent-players that in
controlled manner can express collaborative or egoistic behavior. We hypothe-
size that the agent with persistent collaborative behavior will be more successful
in collaboration game. We also expect that the outcome of the experiments will
give us better insights of how to improve the agents so they can be used as a
controlled substitute of a human player.
Moreover, we want to understand the mechanism of collaborative behavior,
i.e. how collaboration is maintained, how to avoid exploitation of collaborative
agents. This understanding will help us interpret the data from MARS 500 ex-
periment [12,20,10].
2 Methods
2.1 Settings of the Interactions
In our study we use the prisoner dilemma, a classical problem in the game
theory, which used to study collaborative behavior. Every player can chose either
cooperation or defection. If both players choose to cooperate, than they get 3
points each. In contrast, every player gets only 1 point if both of them choose
to defect. In other words, the mutual cooperation is more beneficial than the
mutual defection. On the other hand, the defection is always more beneficial
than then cooperation under any given strategy of the partner. If one player
defects and another one cooperates than the defecting and cooperating player
get 5 and 0 points, respectively. Because of the above mentioned properties of
the game, players face a dilemma: to cooperate or to defect.
In our study the interaction between two players (computer agents) is orga-
nized in the following way. One of the agents starts the game by choosing one
of the two available moves (cooperation or defection). Another agent ”sees” the
move chosen by the first agent and uses this information to make its own move.
After two agents made their moves, each of them gets points determined by the
above given payoff matrix. The third move is made by the first agent. The third
and the second moves together determine another portion of points that will be
delivered to the two interacting agents. In other words, the second move deter-
mines the payoffs two times (first, in combination with the first move and then
in combination with the third move). The agents make subsequent moves in the
above described way until a predefined number of moves is made.
2.2 Model of the Agents
To formalize algorithms imbedded into the computer agents we consider a model
similar to the Turing machine. Every agent has a fixed and limited number of
Search WWH ::




Custom Search