Social Learning Dynamics and Norm Formation
- Dynamic Decision Theory of Learning Agents for POST Game Theory
-
In this paper we introduced the concept of social
learning dynamics and apply it to norm formation and collapsing processes.
Learning dynamics is a decision dynamics which is
introduced by H. Deguchi [1,2].The dynamics is derived from Markov
decision process which is induced by evaluation model for alternatives of
each agents.
In the special cases SLD induce the dynamics of the similar types as repricator
dynamics of evolutionary game theory.
We apply SLD to norm formation problem. Norm formation
is a most basic process of social interaction. To clarify this process we
introduce the mutual commitmentand meta commitment processes among agents.
R. Axelrod analyzed the commitment structure of norm formation and collapsing
process in his norm and meta norm game [3] . He analyzed the games
by using genetic algorithm. We try to deal with the modelfrom
theoretical point of view by using SLD. For the purpose we reformulate
the norm game as a coupling model between the social learning dynamics of
alternatives {C,D} and the dynamics ofmutual commitment as normative attitude.
We also extend this model and introduce several types of centralized and
decentralized commitment mechanism on the agent society.
Game theory is an excellent framework for analyzing rational
and dyadic interaction between agents. Unfortunately it is very rare to find
rational & dyadic interaction process in real societies or organizations.
Agents usually change its attitude after referring the actions of other agents
and itsresults. It is a social learning process. Thus we insist that SLD
gives mathematical foundation for agent based modeling with social learning.
[1] http://www.iss.uw.edu.pl/osrodki/obuz/simsoc5/contrib/HiroshiDeguchi.pdf
[2] Economics as Complex Systems, Hiroshi Deguchi, JUSE, Tokyo, 2000
(Japanese)
[3] Robert Axelrod, The Complexity of Cooperation - Agent-Based Models of
Competition and Collaboration, Princeton University Press , 1997
Hiroshi Deguchi
Tokyo Institute of Technology
Department of Computational Intelligence and Systems Science
Interdisciplinary Graduate School of Science and Engineering
http://www.dis.titech.ac.jp/
deguchi@dis.titech.ac.jp
Technological Innovation and Industrial Policies
- Multi Agent Simulation with Classifier System and Double Loop Learning
-
In this paper, we formulate a multi-agent model of high-tech
industry by agent-based simulation with double loop learning. Computer simulation
is becoming popular in economical and organizational research. Many economists
and organizational scientists use simulation for their research.Agent based
simulation is one modern methods in computer simulation. It is a useful method
to generate new theories and also to verify effects of policies. We formulate
a simulated society of firms in a virtual industry and analyze the society
by agent-based simulation.
We use a classifier system as a decision-making tool
of an agent who makes its decision depending on rules in the classifier system.
Firm agents determine how much R&D investment and production investment
they will spend. Technological innovation of high-tech industry causes the
increase of quality but not quantity in our model. R&D investment affects
the quality of goods; capital investment affects the quantity of goods. The
resource allocation of investment is very important for agents in this vitural
industry.
We assumed three types of agent in our virtual high-tech
industry, in which each agent has a different goal depending on the type.
The first type of firm agents seek profit maximization. The second type of
firm agents will maximize their market share. The goal of last type is technological
maximization. Agents of different types have different evaluation functions.
Agents may change their goals (evaluation functions) when they can not survive
well in industry.
We analyze that what kind of strategy is powerful in
agent society. Also we verify the effect of industrial policy in high-tech
industry. Our simulations are performed under the several types of initial
and boundary conditions.
Hiroshi Deguchi
Tokyo Institute of Technology
Department of Computational Intelligence and Systems Science
Interdisciplinary Graduate School of Science and Engineering
deguchi@dis.titech.ac.jp
Hao Lee
Kyoto University
Graduate School of Economics
ibuki@eco.mbox.media.kyoto-u.ac.jp
Sakyouku Takano