We proposeseveralhe-tic approachesto the development of pricing algorithms for softwareagents that incorporate foresight,i.e., an ability to model and predictresponsesby competitors. In the absence of foresight, prior work has shown that, in an economy of myopic softwareagents, undesirablesystem behaviors such as endlessprice wars can &equently occur (Kephart et al., 1998). We show how the introduction of even the smallest amount of lookaheadin the agents’ pricing algorithms can significantlyreduce or eliminatethe occurrenceof price wars. We also investigate two approachesto developingalgorithmsthat are capableof deep locbhead, while avoidingthe classicproblem of iniinite recursionof opponentmodels. The two approachesare based on adaptationsOE(i) the classicminirnaxfixed-depth search algorithmsused in two-playergames such as chess; (ii) dynamicpro~g (DP)-swle ~gorithmsathat have recentlybeen extendedto the domainof two-playerzer~sum Markov games (Elttman, 1994).
[1]
Paul R. Milgrom,et al.
Adaptive and sophisticated learning in normal form games
,
1991
.
[2]
Michael L. Littman,et al.
Markov Games as a Framework for Multi-Agent Reinforcement Learning
,
1994,
ICML.
[3]
Dimitri P. Bertsekas,et al.
Dynamic Programming and Optimal Control, Two Volume Set
,
1995
.
[4]
Junling Hu,et al.
Self-fulfilling Bias in Multiagent Learning
,
1996
.
[5]
James E. Hanson,et al.
Price-war dynamics in a free-market economy of software agents
,
1998
.
[6]
Edmund H. Durfee,et al.
Learning nested agent models in an information economy
,
1998,
J. Exp. Theor. Artif. Intell..
[7]
Dean P. Foster,et al.
Regret in the On-Line Decision Problem
,
1999
.