CollaWorld

Click on the "Add Agent" button in the applet several times to add agents to the pool. Watch them interact through status shown in bars of four colors.

• DARK GREEN: Investment chunk. During each encounter, the agent devotes this much effort into the collaboration, if it decides to cooperate. Otherwise it turns away from the collaboration and devotes zero. The scale of this bar is the same as the accumulated wealth bar.
• OLIVE GREEN: Collaboration effectiveness. For each collaboration, the partner's investment is effectively amplified by this factor. An empty bar indicates an effectiveness of 1.0, while a full bar indicates 2.0.
• BLACK: Engagement counter. The total length of the bar (hollow plus solid) is the number of encounters the agent has gone through with others. The solid part indicates the number of times either agent actually decides to cooperate, i.e. dedicating a non-zero effort.
• GRASS GREEN: Accumulated wealth. This bar accumulates investment chunks and is displayed in the same scale. The profit made by an agent during a collaboration in this environment is formulated as explained later.

The flashing red bars indicate engagement between pairs of agents. The numbers underneath the names are the current accumulated wealth of the agent. When you want more agents into the pool, click on Add Agent. Clicking on Remove Stagnant Agent kicks out the oldest agent that has turned down the most number of collaboration opportunities. The Controlled Environment option requires each addition of agents be designated the same investment chunks and collaboration efficiencies, as well as a regulated encounter permutation.

Currently there are eight classes of agent strategies implemented. Program listings in Java for each strategy are available for the interested reader.

• TIT FOR TAT: Made famous in Axelrod's work, this strategy calls for defection only if the partner made its last collaborator suffer a loss. Otherwise the strategy suggests cooperation. Agents of this strategy have names with "T" initials.
• BLIND: This strategy simply suggests unconditional cooperation. Agents of this strategy have names with "B" initials.
• CROOK: This strategy simply suggests unconditional defection. Agents of this strategy have names with "C" initials.
• TIT FOR 2 TATS: Similar to TIT FOR TAT except that it only defects after two consecutive losses. Agents of this strategy have names with "2" initials.
• RANDOM: Random choice. Initialed "R".
• JOSS: Like TIT FOR TAT, except there is a 10% chance when TIT FOR TAT cooperates, JOSS would defect. Initialed "J".
• DOWNING: A statistical strategy that cooperates if partner is responsive to my cooperations and defections. In the first two moves the strategy gives away two cooperations, then it starts counting and challenges the partner by defecting until it finds evidence to believe the partner is responsive. Initialed "D".
• FRIEDMAN: Cooperate until partner defects, then defect unconditionally from there on. Initialed "F".

During each round of encounter, each of the two agents receive a history of the other. The history of an agent records whether its previous partner made a profit. Based on this history of its partner, an agent must make a decision to devote to the collaboration (by spending its investment chunk) or to hold back. If an agent's wealth changes after an encounter, this is put in the record as one engagement.

CollaCalc

The profit or loss to each agent during a collaboration is computed according to

profit = 1/2 * ( 1 + q * (R - R') / (R + R') ) * ( c'R + cR' )

For interesting behavior that calls for detailed analysis, you may use the CollaCalc applet to compute the detailed profits to each agent during a collaboration. CollaCalc computes cases more general than those used in CollaWorld. For example, besides spending just zero when defecting, CollaCalc considers cases where there may be more serious results. An agent can cause a negative investment when it defects (the value of negligence). Of course, CollaCalc also computes the case where the quality of share is anywhere between 0.0 and 1.0.

The displayed results are the best calculated strategies for each agent (I or my partner) to choose under the situation. When short term choices conflict with that for the long term, there is a dilemma and agents must make a wise choice to profit from the encounters. That is precisely what CollaWorld had set out to investigate.

Created: June 6, 1997