Multiagent Systems

1
Multiagent Systems

• For Artificial Intelligence Course

2
Rational Agents

• Selfishly Maximizing Utility
• Do the right thing but that idea is limited to
  one agent information
  
• Goals of individual agents can conflict.
• Utility of an agent can depends on what others
  are doing too.

3
Issues

• Selfish Agents can be worse than having just one
  agent.
  
• Examples One CPU or Multiple CPUs?
• Can we all get along?
• Develop mechanisms and methods that enable agents
  interact optimally.

4
Collaboration and Cooperation

• Decompose tasks
• Communicate
• Conflict Resolution
• Local computation vs Global computation
• Etc.

5
Applications

• E-commerce (Auction)
• Resource management in Distributed Systems
• Modeling and Optimization
• Traffic Flow Optimization
• Internet Information Filtering
• Etc.

6
Studying Multiagent interaction?

• Micromotives and Macrobehavior
• Thomas Schelling
• W.W. Norton and Co. Inc

7
A Formalism

• Prisoners Dilemma Game Effectively describes the
  situation that rational agents go through
  
• Albert Tucker introduced it.
• Von Neuman studied Game Theory
• RAND Corp. War game

8
Classic Prisoners Dilemma Game

• Two accomplices in bank robberies caught by the
  Police Interrogated separately The police are
  bargaining
  
• Choice of a prisoner Confess (Defect) or Do not
  confess (Cooperate) to the other prisoner
  
• Dominant Strategy is to defect

Prisoner One
Prisoner Two
9
Payoff Matrix for 2-Person PD
Prisoner 1
Prisoner 2
T R P S and 2R T S,
Typically, T 5, R 3, P 1, and S 0
10
Examples of PD

• Building a bridge participate or not
• Two competing companies setting the price for a
  product high price or low price
  
• Nuclear Arms race (Mutually Assured Destruction)
  build more bombs or not
  
• Many many others

How about repeating the PD game?
11
Some terminologies

• A strategy is dominant if, regardless of what any
  other players do, the strategy earns a player a
  larger payoff than any other.
  
• Nash Equilibrium If there is a set of strategies
  with the property that no players can benefit by
  changing its strategy while the other players
  keep their strategies unchanged, then that set of
  strategies and corresponding payoffs constitute
  the Nash Equilibrium.
• Pareto Optimum a situation in which it is not
  possible to make any one person better off
  without making someone else worst off.

12
What happens if we repeat?

• Repeated Prisoners Dilemma
• What will you do?
• CAN WE PROMOTE COOPERATION?

13
Evolutionary IPD Games Axelrod 1996

• Use Evolutionary Algorithms to evolve cooperative
  individuals or to study emergent behavior
  
• Each agent carries genes responsible for the
  game-playing strategy (among other components)
  
• The game is played and best performing agents are
  allowed to survive in the next round and even to
  produce offspring
  
• The process is repeated

14
What about Multiple Players?

• Many real world problems have more than two
  players
  
• Can we model this?
• The Tragedy of Commons - N-Person PD - IAP

15
Tragedy of the Commons

• Commons is Resource Shared
• Agents are selfishly rational

16
Tragedy of Commons

• Introduced by a biologist, Garrett Hardin, in
  1968
  
• Commons is a resource shared by a group of
  people
  
• Logic of Commons Each agent has the right to use
  the resource
  
• Agents are rational Tries to maximize its
  utility
  
• Overusing the commons will make everyone
  dissatisfied in the long run
  
• Internet is a common resource

17
An Example (Internet Information Search)

• Many internet search companies (hotbot, lycos,
  yahoo, inktomi, etc.), information archival
  companies (Alexa.com, etc) send out softbots to
  many, many, many.. sites
• Sending queries is almost free (The companies pay
  for the bandwidth)
  
• Data mining get all the data (even snap-shot of
  most of the web sites around world to local hard
  disks then sort out needed data)
  
• The resource is overused slow responses, server
  down times, etc.

18
N-Person PD

• N-Person PD formalism can be used to model the
  TOC.
  
• What to know more?
• See you in my Advanced AI class next semester!