1 / 21

Cooperation in Anonymous Dynamic Social Networks

Cooperation in Anonymous Dynamic Social Networks. Brendan Lucier University of Toronto. Nicole Immorlica Northwestern University. Brian Rogers Northwestern University. $. $. $. $. $. $. $. $. $. $. $. $. iLikeDogs. Sephiroth97. XcutieX. Jake93726. trustw0rthy.

mireya
Download Presentation

Cooperation in Anonymous Dynamic Social Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Cooperation in Anonymous Dynamic Social Networks Brendan Lucier University of Toronto Nicole Immorlica Northwestern University Brian Rogers Northwestern University

  2. $ $ $ $ $ $ $ $ $ $ $ $

  3. iLikeDogs Sephiroth97 XcutieX Jake93726 trustw0rthy Nice_guy_21 CatLover jdhxy John_Doe

  4. iLikeDogs Sephiroth97 XcutieX Jake93726 trustw0rthy Nice_guy_21 CatLover jdhxy John_Doe

  5. Explain cooperative behaviour Goal:

  6. Prisoner’s Dilemma on an Anonymous Network

  7. Prior Work • [Datta’93], [Ellison’94], [Ghosh-Ray’96], [Kranton’96], [Friedman-Resnick’01], ... • Common idea: Agents apply strategies that indirectly punish defecting behaviour. • E.g. contagion of defection, trust-building phase for new links. • Generally supports only fully cooperative behaviour at equilibrium.

  8. This Work • We consider a network formation process that runs simultaneously with the prisoner’s dilemma game. • There are equilibria in which cooperation arises naturally from the interplay between the dilemma game and the network formation game. • Strategies at equilibrium are extremely simple. • Supports stable co-existence of cooperation and defection. Main idea: defection stunts neighbourhood growth, while cooperation leads to more opportunities to interact with other cooperators.

  9. The Model • Directed graph. Each node sponsors up to K links. • Game proceeds in discrete rounds. On each round: • Link Proposal:Nodes sponsoring fewer than K links can propose new links to partners chosen randomly from the population. • Link Acceptance:Nodes can accept or reject each proposed link. • Stage Game: Each node chooses a single action: cooperate or defect. Prisoner’s Dilemma game played with each neighbour and payoffs are realized. • Link Severance: Nodes choose whether or not to sever any incident links. • Player Exit: Each node exits the system with probability 1-δ, to be replaced by a new player.

  10. The Model • Agents optimize for present value of expected lifetime payoffs. Time discounting is due to exit probability. • Anonymity: agents are unaware of their partner’s identities, and in particular whether a new partner has been encountered previously. • Quantity of interest: fraction q ϵ [0,1] of agents who choose cooperation in each round.

  11. Properties of Strategies A strategy that is... • Unforgiving: immediately breaks the link to any partner that defects. • Trusting: always accepts proposed links, and keeps the link with any partner that cooperates. • Consistent: either cooperates every turn (cooperator), or defects every turn (defector).

  12. Main Results: • There are (sequential) equilibria in which all agents apply UTC strategies, and which support either full or partial cooperation. • We determine the fraction of agents that will choose cooperation at equilibrium, as a function of the model parameters.

  13. Characterizing UTC Equilibria • In a UTC strategy, the only choice is between cooperation and defection, made at birth. • What is the expected utility of each choice? • Idea: Assume q is fixed for all time. • Implies a steady-state of system properties, such as the number of links proposed each round. • Compute lifetime utilities in this steady-state as a function of q: uC(q) and uD(q).

  14. Characterizing UTC Equilibria • Utility of choosing cooperation: • Utility of choosing defection: uD(q) uC(q) q q

  15. Analysis Case 1 Case 2 Case 3 • All-Defect (q=0) is an equilibrium in each case. • All-Cooperate (q=1) is an equilibrium in case 2. • Where the curves cross, there is an equilibrium in which defectors and cooperators coexist. • An equilibrium at q+ is stable; an equilibrium at q- is not. q q q q- q- q+

  16. Main Results: • There are (sequential) equilibria in which all agents apply UTC strategies, and which support either full or partial cooperation. • We determine the fraction of agents that will choose cooperation at equilibrium, as a function of the model parameters.

  17. UTC Strategies Revisited Why is it an equilibrium for agents to play unforgiving, trusting, and consistent strategies? ? Defectors will always defect, so it is optimal to break links. Cooperators will always cooperate, so it is optimal to keep links. Accept proposed links if expected utility is positive. Claim: always if q > 2/3 Potential deviation: an agent cooperates to build up social capital, then “cashes in” by defecting and starting over.

  18. Consistency Revisited Game utilities Coop rate Survival rate Intuition: a cooperator generates utility by building a network of partners (at a cost), then gaining benefit from those links. If an agent could improve utility by switching to defection, it must have been even better to defect every turn. • Lemma: If other agents play UTC strategies, and • (1+b) / (1+a) ≥ 1 – (1-q)δ2, then a consistent strategy maximizes expected utility. Note: sufficient to have b ≥ a.

  19. Conclusions • We proposed a model for repeated Prisoner’s Dilemma in an anonymous social network with sustained interactions. • Our model supports (full and partial) cooperation at equilibrium with simple strategies. • Key insight: interaction between network formation and behaviour choices. Open Questions: • What if agents are heterogeneous? • Different network formation models (preferential attachment, local search)? • Other applications of the interaction between network games and network formation.

  20. Thank You

  21. The Model • Directed graph. Each node sponsors up to K links. • Game proceeds in discrete rounds. On each round: • Link Proposal:Nodes sponsoring fewer than K links can propose new links to partners chosen randomly from the population. • Link Acceptance:Nodes can accept or reject each proposed link. • Action: Each node chooses an action: cooperate or defect. • Payoffs:Prisoner’s Dilemma game played with each neighbour. Payoffs awarded. • Link Severance: Nodes choose whether or not to sever any incident links. • Player Exit: Each node exits the system with probability 1-δ, to be replaced by a new player.

More Related