Agents in Economic Markets and Games
173
6.7 Multi-Agent Systems and Games
Modeling complex system behavior is an emergent science which demonstrates the complex social behavior of different communities working together.
Multi-agent systems can be used effectively to model this. These systems are
inherently distributed where agents are either spatially spread heterogeneous
in nature and have limited information available to them. Multi-agent systems
are essentially players involved in a non-cooperative game scenario. If all the
individuals tried to optimize their behavior, globally the system may optimize
as well. But there are problems analyzing these optimizations.
1. It is not possible to have a payoff matrix for models which are not games
to begin with. The measurement of the payoff will have to be associated
with a fitness variable U as part of the agent memory.
2. Evolutionary algorithms are used within agents, primarily to allow them
to evolve. Most of these algorithms use a comparison technique to calculate the difference between the actual fitness collected and the expected
fitness of a particular strategy. In multi-agent systems, the expected fitness is difficult to predict until the agent has tried the strategy in the
simulation.
3. Multi-agent systems are sometimes deterministic or stochastic. Agents
can use their memories to save good strategies, making the model deterministic by knowing what to play next. However, there is unpredictable behavior in complex systems and agents struggle to find the
better strategies. The system then takes longer to reach global maxima.
4. Agents need to calculate when to change their behavior to reduce complexities in their code.
5. The strategies are sometimes a continuous variable and not a set of
discrete strategies as in traditional game theory.
Importing the principles from game theory into multi-agent systems would
thus require a number of assumptions to be embedded into the agents to test
the theory. The world economy is often referred to as an “evolving game with
nearly four billion players” [49].
Multi-agent systems have a very close relation to the principles of the
games. The works of [13] and [73] are a few examples where game theory
has been used to develop agent-based models of players playing games and
evolving characteristics [132].
The behavior of an agent is programmed by the way it should respond
to signals using rules embedded in the program. Signals are messages coming
from other agents or can be the changes in the environment which affect the
agent’s behavior. The agent’s behavior is termed as the strategies or functions
173
6.7 Multi-Agent Systems and Games
Modeling complex system behavior is an emergent science which demonstrates the complex social behavior of different communities working together.
Multi-agent systems can be used effectively to model this. These systems are
inherently distributed where agents are either spatially spread heterogeneous
in nature and have limited information available to them. Multi-agent systems
are essentially players involved in a non-cooperative game scenario. If all the
individuals tried to optimize their behavior, globally the system may optimize
as well. But there are problems analyzing these optimizations.
1. It is not possible to have a payoff matrix for models which are not games
to begin with. The measurement of the payoff will have to be associated
with a fitness variable U as part of the agent memory.
2. Evolutionary algorithms are used within agents, primarily to allow them
to evolve. Most of these algorithms use a comparison technique to calculate the difference between the actual fitness collected and the expected
fitness of a particular strategy. In multi-agent systems, the expected fitness is difficult to predict until the agent has tried the strategy in the
simulation.
3. Multi-agent systems are sometimes deterministic or stochastic. Agents
can use their memories to save good strategies, making the model deterministic by knowing what to play next. However, there is unpredictable behavior in complex systems and agents struggle to find the
better strategies. The system then takes longer to reach global maxima.
4. Agents need to calculate when to change their behavior to reduce complexities in their code.
5. The strategies are sometimes a continuous variable and not a set of
discrete strategies as in traditional game theory.
Importing the principles from game theory into multi-agent systems would
thus require a number of assumptions to be embedded into the agents to test
the theory. The world economy is often referred to as an “evolving game with
nearly four billion players” [49].
Multi-agent systems have a very close relation to the principles of the
games. The works of [13] and [73] are a few examples where game theory
has been used to develop agent-based models of players playing games and
evolving characteristics [132].
The behavior of an agent is programmed by the way it should respond
to signals using rules embedded in the program. Signals are messages coming
from other agents or can be the changes in the environment which affect the
agent’s behavior. The agent’s behavior is termed as the strategies or functions