Moduleco 2 - Concepts of the multi-agent's dynamics 2.3. Collective Learning : collective intelligence and intelligence of the collective phenomenon (a)

Stochastic learning with global self-reinforcement (involve SELECTION rather than EMERGENCE of a collective structure or "order").
Agents Choose between known alternatives, given public information about average population choices, or global externality. Polymorphism is often residual or non-existent (absorbing states)

• Aggregate stochastic process : Master equation, Fokker Plank approximation, Field effect... (Weidlich & Haag [1983],Topol [1991],Aoki [1996],Orléan [1995],[1996]).
• Pair wise random matching  evolutionary games (i.e. Coordination games with and without stochastic mutations (Mailath [1993] )

This models exhibit multiple equilibria and equilibrium shift. Its depend mainly on the externality and/or structure of the revision process and the nature of the "noise" (i.e. volatility)

• Voter and Sequential models :  (Arthur, [1989], [1994]) ; informational cascades and mimetic chains : (Bikhchandani, Hirshleifer, Welch [1992])

The result depends mainly on the order of the sequential process (path dependence)

• Moving population : the "ant optimal path discovery" : the result drive to the faster path

Denis.Phan@enst-bretagne.fr ;   Antoine Beugnard@enst-bretagne.fr