Multiagent Systems

1
Multiagent Systems Societies of Agents (II)

• Authors Michael N. Huns Larry M. Stephens
• Speaker Shabbir Ali Syed
• CSCE 976, April 8th 2002

2
Agent Interaction Protocols

• Govern the exchange of a series of messages among
  agents
• Case 1 Agents have conflicting goals
• Case 2 Agents have similar goals

3
Agents with conflicting goals

• Conflicting goals or simply self-interested
• Maximize payoff (utility functions)

4
Agents with similar goals

• Objective maintain globally coherent performance
  without violating autonomous behavior of agents
• Determine shared goals
• Determine common tasks
• Avoid unnecessary conflicts
• Pool knowledge and evidence

5
Some Interaction Protocols

1. Coordination Protocols
2. Cooperation Protocols
3. Contract Net
4. Blackboard Systems
5. Negotiation
6. Multi-Agent Belief Maintenance
7. Market Mechanisms

6
1. Coordination Protocols

• Done between multiple agents to satisfy
  individual or group goals
• Why coordination is needed
• Maintain dependencies between actions
• Meet global constraints
• When no one agent has sufficient competence,
  resources, or information to achieve system goals

7
Distributed AI (DAI)

• Distributing data and control
• Agents have autonomy to generate new actions and
  to decide which goals to pursue next
• Disadvantages
• KB is distributed, so each agent has only a
  partial and imprecise perspective of KB
• Degree of uncertainty in actions
• Difficult to attain coherent global behavior

8
Goal Graph

• It is a AND/OR graph with the leaves representing
  the goals
• Activities are
• Defining Goal Graph, including identification and
  classification of dependencies.
• Assigning particular regions of the graph to
  appropriate agents
• Controlling decisions about which areas of the
  graph to explore
• Traversing the graph
• Ensuring that successful traversal is reported

9
Agent Structures

• Commitment
• Pledges to undertake a specified course of action
• As situation changes agents must evaluate whether
  existing commitments are still valid
• Internal and Belief consistent
• Convention
• Provide a means to manage commitments in changing
  circumstances
• Provides degree of predictability Agents can
  take into consideration future conflicts,
  dependencies, and activities of other agents

10
Limited B/W Social Convention

• INVOKE WHEN
• Local commitment dropped
• Local commitment satisfied
• ACTIONS
• Rule1 IF Local commitment satisfied
• THEN inform all related commitments
• Rule2 IF Local commitments dropped because
  unattainable or motivation not present
• THEN inform all strongly related commitments.
• Rule3 IF Local commitments dropped because
  unattainable or motivation not present
• AND communication resources not overburdened
• THEN inform all weakly related commitments.

11
Basic Joint Action Convention

• INVOKE WHEN
• Status of commitment to joint action changes
• Status of commitment to attaining joint action in
  present tem context changes
• Status of joint commitment of a team member
  changes.
• ACTIONS
• Rule1 IF status of commitment to joint action
  changes
• OR
• IF status of commitment to present team context
  changes
• THEN inform all other team members of these
  changes
• Rule 2 IF status of joint commitment of a team
  member changes
• THEN determine whether joint commitment still
  viable.

12
Cooperating Agents

• Each agent should share status of its commitment
  to
• the shared objective
• the given team framework
• If belief changes, should inform all agents

13
2. Cooperation Protocols

• Divide and Conquer Approach
• Smaller sub-tasks require less capable agents
• Fewer resources
• Distributing Criteria
• Avoid overloading critical resources
• Assign tasks to agents with matching capabilities
• Make an agent with wide view assign tasks to
  other agents
• Assign overlapping responsibilities to agents to
  achieve coherence
• Assign highly independent tasks to agents in
  spatial or semantic proximity-minimizes
  communication and synchronization costs
• Reassign tasks if necessary for completing urgent
  tasks

14
Methods for task distribution

15
3. Contract Net

• Best know and widely applied to distribute tasks.
• Connection problem finding an appropriate agent
  to work on a given task
• Manager
• Announces a task that needs to be performed
• Receives and evaluates bids from potential
  contractors.
• Award a contract to a suitable contractor.
• Receive and synthesize results.
• Contactor
• Receive task announcement
• Evaluate my capability to respond
• Respond (decline, bid)
• Perform the task if my bid is accepted
• Report my results.

16
Contract Net

17
Task Announcement

• Addressee Contractor
• Eligibility Specification Contractors should
  meet certain criteria to make bids.
• Task Abstraction A brief description of the
  task, is used by contractors to rank tasks from
  several task announcements.
• Bid Specification Tells contractors , what info.
  must be provide with the bids.Manager compares
  different contractors on basis of bids.
• Expiration Time Deadline for receiving bids.

18
Limitations

• Task must be awarded anyway, even if a better
  contractor is busy.
• Manager is under no obligation to inform other
  customers that an award has already been made.
• All potential contractors can be busy and so not
  send bids to the manager
• A potential contractor ranks the proposed task
  below other tasks, and so may not send bids.
• No contractor even if idle is able to handle the
  task.

19
Proposed solution

• Manager Requests immediate response bids.
• Contractors eligible but busy.
• ineligible
• uninterested.
• Manager directed contracts
• Contractorsacceptance
• refusal

20
4. Blackboard Systems

• Characteristics of BB systems
• Independence of expertise
• Diversity of problem-solving techniques
• Flexible representation of blackboard information
• Common interaction language
• Event based activation
• Need for control
• Incremental solution generation
• Knowledge source KS

21
BB Systems characteristics

• Independence of expertise
• A specialist (KS) can act independently of the
  other
• Diversity of problem-solving techniques
• Internal representation of each KS is hidden
  from others
• Flexible representation of blackboard
  information
• No restriction as to what can be placed on the
  blackboard
• Common interaction language
• KSs should be able to correctly interpret
  information posted by other KSS

22
BB Systems characteristics

• Event based activation
• KSs give their preferences and blackboard
  triggers them whenever it occurs
• Need for control
• Triggered KS, evaluates quality of its
  contribution?informs Control Component about the
  cost?estimates benefits and decides how to
  trigger for better problem solving
• Incremental solution generation
• KS contributes as needed (refining,
  contradicting, initiating)

23
Diagram for blackboard

24
5. Negotiation

• Joint decision reached by two or more agents,
  each trying to reach an individual goal
• Features
• Language used by participating agents
• Protocols followed by agents as they negotiate
• Decision process used for concession, criteria
  for agreement and to determine position

25
Attributes of negotiation

• Efficiency agents should not waste resources in
  coming to an agreement
• Stability no agent should have an incentive to
  deviate from agreed upon strategies
• Simplicity the negotiation mechanism should
  impose low computational and bandwidth demands on
  the agents
• Distribution no central decision maker
• Symmetry should not be biased against any agent
  for arbitrary reasons

26
Systems for Negotiation

• Two types
• Environment centered
• Agent centered

27
Environment Centered

• Rules by which agents can interact productively
  and fairly irrespective of their capabilities or
  intentions
• Task-oriented domain
• State-oriented domain
• Worth-oriented domain

28
Agent Centered

• Best strategy for an agent to follow in a given
  environment
• Task-oriented domain

29
Task-Oriented Domain

• Agents have set of tasks
• Resources needed are available
• Agents can achieve tasks without help nor
  interference
• Agents can benefit by sharing some tasks Example
  Internet downloading

30
Example Internet downloading Constraints

• Each agent declares documents it wants
• Common documents are assigned by the toss of
  coin
• Agents pay for the documents they download
• Agents are granted access to all documents in
  common set
• Mechanism is
• simple, accurate, systematic, and distributed
• (no document downloaded twice)

31
Agent Centered Approaches

• Speech act classifiers together with a possible
  world semantics
• used to formalize negotiation protocols and their
  components
• Unified Negotiation Protocol
• Assumption agents are economically rational
• Set of agents must be small
• Must have a common language
• Must have a common problem abstraction
• Must reach a common solution

32
Speech Act

• An agent forms and maintains its commitments to
  achieve a task individually iff
• It has not pre-committed itself to another agent
  to adopt and achieve a task
• It has a goal to achieve the task individually
• It is willing to achieve the task individually

33
Unified Negotiation Protocol

1. Deal joint plan between agents that would
   satisfy all their goals
2. Utility amount agent is willing to pay minus
   cost of deal (to be maximized)
3. Negotiation set set of all deals that have a
   positive utility for all agents
4. Conflict negotiation set is empty
5. Compromise agents agree to negotiate
6. Co-operative all deals in negotiation set are
   preferred by both agents over achieving their
   goals

34
IN
IN
Human(x)gtMortal(x)
Human(socrate)
fact
fact


Justification node
IN
Mortal(socrate)
Derived fact
35
IN
IN
IN
OUT


-

IN
36
Drank-Fountain-of-youth(socrate)
Human(x)gtMortal(x)
Human(socrate)
IN
IN

OUT
-

MORTAL(SOCRATE)
IN
37
6. Multiagent Belief Maintenance

• High level interaction among agents
• Relies on Truth Maintenance Systems (TMS)
• Data structure (AI) that keeps track of the truth
  of a fact in a KB given the truth of the facts it
  is derived from (which constitute its support or
  justification)

38
TMS (or Reason Maintenance System, RMS)

• Ensure integrity of agents knowledge
• Ensure that its stable
• Datum that has a valid justification is believed
• Datum that lacks a valid justification is
  disbelieved
• Well founded
• Permits no set of its beliefs to be mutually
  dependent
• Logically consistent
• Stable at the time consistency is determined and
  has no logical contradiction
• No datum is both believed and disbelieved at same
  time

39
Justification based TMS (JTMS)

• Datum
• Set of justification.
• Associated status.
• INTERNALBelieved because of a valid local
  justification.
• EXTERNALBelieved because another agent asserted
  it.
• OUT Disbelieved.
• A communicated Datum must be
• INTERNAL to at least one of the agents that
  believes it.
• Either INTERNAL or EXTERNAL to the rest.

40
TMS before Justification
41
Resultant Network
42
Multi-agent TMS

• Invoked by addition or removal of justifications
• Belief changes should be resolved with as few
  agents as possible
• Changing as few beliefs as possible
• When invoked
• Unlabels data
• Chooses labeling for unlabelled shared data
• Initiates labeling

43
7. Market Mechanisms

• For large or unknown of agents
• The goods being traded
• Consumer agents that are trading the goods
• Producer agents, with their technology for
  transforming some goods into others
• Bidding trading behaviors of agents

44
Competitive Equilibrium

• Consumers bid to maximize their utility, subject
  to budget constraints
• Producers bid to maximize their profits, subject
  to technological capabilities
• Net demand is zero for all goods
• Rational action
• maximizes preferences for an agent
• (including past commitments)

45
Societies of agents

• Intelligent agents work well in groups
  (societies) not in isolation
• Distributed system is a better solution
• Peer to peer better than client server
• Social commitments commitments of an agent to
  another

46
Social dependence

• Social dependence (x y a p)
• Agent x depends on agent y with regard to act a
  for realizing state p, when p is a goal of x and
  x is unable to realize p while y is able to do so.

47
Types of dependencies

• Voluntary agents adopt roles that bind them to
  certain commitments
• Compound mutual dependence occurs when x and y
  depend on each other for realizing a common goal
  p
• Reciprocal x and y depend on each other for
  realizing different goals

48
Co-operation

• Form of mutual dependence
• Agents form a co-operative team when
• All agents share a common goal
• Each agent is required to do its share to achieve
  the common goal by the group itself or a subgroup
• Each agent adopts a request to do its share

49
Conclusion

• Characteristics of multi-agent systems.
• Mechanisms for agents communication.
• High level agent interaction protocols.
• Societies of agents

50
Future work

• To develop protocols or societies in which
• the effects of deception and misinformation
• can be constrained

51

• QUESTIONS DISCUSSION