Agent Modeling Language: Mental States and Ontologies

1
Agent Modeling LanguageMental States and
Ontologies

• Marcelo Siqueira
• Edilson Leite
• Jacques Robin

2
Outline

1. Mental states package structure
2. Beliefs package
3. Goals package
4. Plans package
5. Mental Relationships package
6. Ontologies package

3
Context

• Mental States are cognitive components needed to
  the realization of inteligent tasks in MAS.
• Examples beliefs, plans, intentions,
  commitments, etc. (Ferber 95)
• Scenario
• Ronaldo believes that the ball is near to him.
• Ronaldo aims to kick the ball into the goal.
• Roberto Carlos commits to help Ronaldo (as soon
  as possible!).

In a Cooperatve MAS, before and after each speech
act a lot of reasoning about mental states is
necessary.
4
ReminderOverall AML Package Structure
5
ReminderOverall AML Package Structure
Mental
6
ReminderOverall AML Package Structure
Mental
Ontologies
7
Mental Package Structure

• Defines the metaclasses which can be used to
• Support analysis of complex problem/systems,
  particularly by
• Modeling intentionality in use case models,
• Goal-based requirements modeling
• Problem decomposition, etc.
• Model mental attitudes of autonomous entities,
  which represent their informational, motivational
  and deliberative states.

Mental
8
Mental States Package

• Comprises common fundamental classes used to
  define concrete metaclasses contained within the
  rest of the Mental sub-packages
• Mental States
• Mental Constraints
• Mental Associations
• Mental Relationships
• etc.

9
MentalState

• Models mental attitudes, which represents their
  informational, motivational and deliberative
  states
• Support for the human mental process of
  requirements specification and analysis of
  complex problems/systems, by
• Expressing intentionality in use case models
• Goal-based requirements modeling
• Problem decomposition

10
MentalClass

• Used to specify mental attitudes of
  MentalSemiEntityTypes, and meta-associations of
  Responsibility.

11
ConstrainedMentalClass

• Allows its concrete subclasses to specify
  MentalConstraints.

12
MentalConstraint

• Specifiy properties of ConstrainedMentalClassses
  which can be used within mental (reasoning)
  processes of owning MentalSemiEntityTypes.

13
MentalConstraintKind

• Specifies kinds of constraints specified for
  contributor and beneficiary in the Contribution
  relationship.

MentalConstraint
kind MentalConstraintKind
14
MentalRelationship

• A superclass to all metaclasses defining the
  relationships between MentalStates

MentalRelationship
15
MentalSemiEntityType

• Allows the classes (Agents and Environments) own
  mental states.
• A MentalClass instance can decide
• Which Goal is to be achieved and which not,
• When and how the particular Goal instance is to
  be achieved,
• Whether the particular Goal instance is already
  achieved or not,
• Which Plan to execute, etc.

16
MentalAssociation

• Defines an Association between a
  MentalSemiEntityType and a MentalClass to specify
  a MentalProperty.

17
MentalAssociation
Association (from UML)

• Defines an Association between a
  MentalSemiEntityType and a MentalClass to specify
  a MentalProperty.

0..1
MentalAssociation
2
association redefines association
0..1
Property (from UML)
subsets memberEnd mental MemberEnd
subsets ownedElement MentalConstraint
1
MentalConstraint
MentalProperty
0..1

degree ValueSpecification 0..1


/mentalAttribute ordered, subsets ownedAttribute

redefines type type
redefines classifier classifier
0..1
0..1
MentalSemiEntityType
MentalClass
MentalSemiEntityType
MentalClass
18
MentalAssociation

• Notation

ltltmentalgtgt
Name
MentalClass
MentalSemiEntityType
1 role
0..8 mentalRole degreevalue

• Example

degree1.0 scoreGoal
canScoreGoal
0..1
0..1
ScoreGoal
degree1.0 attack
canAttack
0..1
0..1
Attack
19
MentalProperty

• Specify the instances of MentaSemiEntityType
  which have instances of the MentalClasses in
  order
• To believe (or not) in a Belief (and to what
  extent)
• (Wheter and when) to commit to a Goal
• etc.
• Can be owned only by
• MentalSemiEntityTypes as attributes
• MentalAssociations as member ends.

20
MentalProperty

• Notation

MentalSemiEntityType
attribute list
MentalSemiEntityType
operation list
parts
...
ltltbeliefgtgt
behaviors
belief list
ltltmentalgtgt mental property 1 mental property
2 ...
ltltgoalgtgt
goal list
21
MentalProperty

• Example An STRIKER agent

22
Responsibility

• Used to model a relation between MentalClass
  (responsibility classes) and NamedElement
  (responsibility subjects) that are obligated to
  accomplish MentalClasses by means of
• Modification of beliefs
• Achievement or maintenance of goals
• Realization of plans

Realization (from UML)
Responsibility
/isResponsibilityOf subsets supplierDependency


redefines client subject
redefines supplier object
1..
1..
NamedElement (from UML)
MentalClass
23
Responsibility

• Notation

• Example

24
Beliefs Package

• The Beliefs package defines metaclasses used to
  model beliefs such as
• The ball is nearby to a player in a soccer
  matching
• Exists a wumpus in the cave ahead
• Its raining outside here
• The goalkeeper goes to the right side when its
  penalty

25
Belief

• Used to model a state of affairs, proposition or
  other information relevant to the system and its
  mental model.
• It represents the information which a mental
  semi-entity believes (it does not need to be
  objectively true!)
• The ability of a MentalSemiEntityType to believe
  in beliefs is modeled by the ownership of a
  MentalProperty

26
Belief

• Belief can be depicted as
• (I) an UML class with the stereotype ltltbeliefgtgt
  and/or a rectangle with beveled top-left and
  bottom-right corners in top-right corner
• (II) a rectangle with beveled top-left and
  bottom-right corners
• (III) is equals (II), but only with constraints.

(III)
(II)
27
Belief

• Example
• NearBall represents the belief that a ball is
  nearby.
• The attribute near specifies the distance between
  the agent and the ball (default value 1.5).
• The constraint specifies that the actual distance
  from the agent and the ball does not must exceed
  the value of near

28
Goals

• The Goals package defines metaclasses used to
  model goals, such as
• To arrive in an endpoint in a driverless desert
  car racing
• To catch the opponents king in a chessgame
• To score a goal in a soccer matching
• To destroy the red airplane in a warcraft
  simulation game

29
Goal

• Used to model goals, i.e. conditions or states of
  affairs, with which the main concern is their
  achievement or maintenance.
• Goals can be used to represent objectives, needs,
  motivations, desires, etc.

30
DecidableGoal

• Used to model goals for which the goal-holder can
  decide clearly whether it can be achieved or not.

31
UndecidableGoal

• Used to model goals for which the goal-holder can
  not decide clearly whether it can be achieved or
  not.

32
Plans Package

• The Plans Package defines metaclasses devoted to
  modeling plans, such as
• The (travel) planning to arrive in other city
• The strategy to win a soccer matching
• The plan to react to a nuclear accident

33
Plan

• Used to model capabilities of MentalSemiEntityType
  s which represents
• Predefined plans (i.e. kinds of activities a
  mental semi-entitys reasoning mechanism can
  manipulate to achieve Goals)
• Fragments of behavior from which the plans can be
  composed (also called plan fragments).

34
Plan
Plan
EntityRoleType

• Example

ltltpregtgt self.haveBall() ltltpostgtgt not
self.haveBall()
StrikerKickingPlan TeammateSoccerRobot
ltltdecisionInputgtgt teammateself.team.players-gt fir
st(playerplayerltgtself and self.pitch.isNear(self
,player) and player.notOffside())
ltltdecisionInputgtgt self.pitch.isNear(self,otherGoal
)
false
true
teammate-gtnotEmpty()
teammate-gtisEmpty()
KickGoal (self.kickGoal)
PassBall (self.passBall)
KickBallAway (self.kickBallAway)
KickGoal
PassBall
KickBallAway
35
CommitGoalAction

• Used to model the action of commitment to a Goal.
• This action binds the Goal instance to the
  MentalProperty of a MentalSemiEntityType.

36
CommitGoalAction

• Example

ltltpregtgt self.haveBall() ltltpostgtgt not
self.haveBall()
StrikerKickingPlan TeammateSoccerRobot
ltltdecisionInputgtgt teammateself.team.players-gt fir
st(playerplayerltgtself and self.pitch.isNear(self
,player) and player.notOffside())
ltltdecisionInputgtgt self.pitch.isNear(self,otherGoal
)
false
true
teammate-gtnotEmpty()
teammate-gtisEmpty()
PassBall (self.passBall)
PassBall (self.passBall)
KickGoal (self.kickGoal)
KickBallAway (self.kickBallAway)
KickGoal (self.kickGoal)
KickBallAway (self.kickBallAway)
KickGoal
PassBall
KickBallAway
Goal
Goal
Goal
37
CommitGoalAction

• Example

ltltpregtgt self.haveBall() ltltpostgtgt not
self.haveBall()
StrikerKickingPlan TeammateSoccerRobot
ltltdecisionInputgtgt teammateself.team.players-gt fir
st(playerplayerltgtself and self.pitch.isNear(self
,player) and player.notOffside())
ltltdecisionInputgtgt self.pitch.isNear(self,otherGoal
)
false
true
teammate-gtnotEmpty()
teammate-gtisEmpty()
PassBall (self.passBall)
PassBall (self.passBall)
KickGoal (self.kickGoal)
KickBallAway (self.kickBallAway)
KickGoal (self.kickGoal)
KickBallAway (self.kickBallAway)
KickGoal
PassBall
KickBallAway
38
CancelGoalAction

• Used to model de-commitment from goals.
• To cancel an commitment is realized by
  destruction of the Goal instance.
• In that cases which is not necessary to destroy
  the Goal instance, de-commitment can be modeled
  by RemoveStructuralFeatureValueAction or
  DestroyLinkAction (both of UML)

39
CancelGoalAction

• Example

40
Mental Relationships Package

• Defines metaclasses used to model relationships
  between MentalStates which can support reasoning
  processes, such as
• A Goal Get Rich can be supported by a Belief
  Win the Prize in a game
• A Goal ScoreGoal can be support by other Goal
  Attack

41
Contribution

• Used to model logical relationships between
  MentalStates and their MentalConstraints.

42
ContributionKind

• Specifies possible kinds of Contributions
• Modeled by means of Non-Classical Logic (Modal
  Logical)
• There are three kinds
• Sufficient - The contributor is a sufficient
  condition for the beneficiary
• Necessary - The contributor is a necessary
  condition for the beneficiary
• Iff - The contributor and beneficiary are
  equivalent, i.e. the contributor is a sufficient
  and necessary condition for the beneficiary.
• Notation

Iff
43
Mental States in AMLa Critical Appraisal

• Strengths
• Simple way to model mental states capabilities in
  MAS.
• Improves the learning curve of non-speacilist
  professionals or IA gurus to develop multi-agent
  based systems.
• Allows sophisticated forms of relationships by
  means of Modal Logic.
• Weaknesses
• Approach (almost) purely object-oriented
• Implementation issues How to map Modal Logic in
  OO?
• Doesnt support reasoning mechanisms (induction,
  abduction, analogy, etc.)
• Mental model is sufficient, but its not complete
  (see Ferber99).
• In pratice, Mental States are not very used to
  implement MAS.

44
Ontologies Package

• The Ontologies package defines the metaclasses
  used to model ontologies.
• Allows the specification of class-level as well
  as instance-level ontologies.

Ontologies
Basic Ontologies
45
Basic Ontologies Package

• Defines the generic means for modelling of
  ontologies in AML
• Classes
• Instances
• Relatioships
• Constraints
• Utilities

46
Ontology

• A specialized package (from UML) used to specify
  a single ontology

47
OntologyClass

• Used to represent an ontology class (also called
  ontology concept or frame).
• Slots are represented by attributes and
  operations.
• Functions, actions, and predicates belonging are
  modeled by its operations.
• OntologyClass can use all types of relationships
  allowed for UML Class (Association,
  Generalization, Dependency, etc.) with their
  standard UML semantics.
• Facets for attributes and operations can be
  extended by adding specific tagged values.
• OntologyClass can also be used as an
  AssociationClass.

48
OntologyClass

• Example

49
OntologyUtility

• A specialized Class used to cluster
• global ontology constants,
• ontology variables,
• ontology functions/actions/predicates
• OntologyUtility classes can be used by elements
  from the same ontology and others importing
  ontologies.
• There can be more than one OntologyUtility
  within the same ontology.

50
Ontologies in AMLa Critical Appraisal

• Strengths
• Support the necessary elements to model a simple
  ontology (classes, instances, relationships,
  etc.) by means of clear and graphical elements.
• Weaknesses
• No perceptible value aggregation (Just a poor
  extension of UML)
• Lack of formal semantics
• Not enough expressivity for complex domains
• Poor model of aximatization by means (only) of
  UML constraints, not being possible to incluide
  CFOL, or alike.
• Lack of integration with Knowledge Representation
  languages like RDF/OWL
• Lack of adherence to existing OMG efforts to
  standardize an MOF-based ontology metamodel
  (Calero et al., 2006)
• Questionable necessity of existence (disposable?)