Agents

1

KIMAS 2003 Tutorial
The Craft of Building Social Agents
Henry Hexmoor University of Arkansas Engineering
Hall, Room 328 Fayetteville, AR 72701
2
Content Outline

• I. Introduction
• 1. History and perspectives on MultiAgent
  Systems
• 2. Architectural theories
• 3. Agent Oriented Software Engineering
• II. Social agents
• 4. Sociality and social models
• 5. Dimensions for Developing a Social Agent
• Examples in Autonomy, Trust, Social Ties,
  Control, Team, Roles, Trust, and Norms
• 6. Agent as a member of a group...
• Values, Obligations, Dependence, Responsibility,
  Emotions
• III. Closing
• 7. Trends and open questions
• 8. Concluding Remarks

3
Definitions

1. An agent is an entity whose state is viewed as
   consisting of mental components such as beliefs,
   capabilities, choices, and commitments. Yoav
   Shoham, 1993
2. An entity is a software agent if and only if it
   communicates correctly in an agent communication
   language. Genesereth and Ketchpel, 1994
3. Intelligent agents continuously perform three
   functions perception of dynamic conditions in
   the environment action to affect conditions in
   the environment and reasoning to interpret
   perceptions, solve problems, draw inferences, and
   determine actions. Hayes-Roth, 1995
4. An agent is anything that can be viewed as
   (a)Perceiving its environment, and (b) Acting
   upon that environment Russell and Norvig, 1995
5. A computer system that is situated in some
   environment and is capable of autonomous action
   in its environment to meet its design objectives.
   Wooldridge, 1999

4
Agents A working definition

• An agent is a computational system that interacts
  with one or more counterparts or real-world
  systems with the following key features to
  varying degrees
• Autonomy
• Reactiveness
• Pro-activeness
• Social abilities
• e.g., autonomous robots, human assistants,
  service agents

5
The need for agents

1. Automation of dirty, dull, and dangerous as well
   as tedious, boring, and routine tasks to relieve
   humans of such duties.
   E.g., desktop assistants or intelligent in
   service of humans.
2. An improved human sense of presence for humans
   collaborating in physically disparate locations.
   
   E.g., knowledge management tasks like war-rooms
   and human users benefit from agents who proxy for
   their human counterparts.
3. Democratization of computing, services, and
   support. E.g., functions such
   as the department of motor vehicles or public
   libraries and virtual museums.
4. Reduction of redundancy and overlap due to
   competition. E.g., tracking and
   sharing power or telecommunication services.

6
Agent Typology

• Person, Employee, Student, Nurse, or Patient
• Artificial agents owned and run by a legal
  entity
• Institutional agents a bank or a hospital
• Software agents Agents designed with software
• Information agent Data bases and the internet
• Autonomous agents Non-trivial independence
• Interactive/Interface agents Designed for
  interaction
• Adaptive agents Non-trivial ability for change
• Mobile agents code and logic mobility

7
Agent Typology

• Collaborative/Coordinative agents Non-trivial
  ability for coordination, autonomy, and
  sociability
• Reactive agents No internal state and shallow
  reasoning
• Hybrid agents a combination of deliberative and
  reactive components
• Heterogenous agents A system with various agent
  sub-components
• Intelligent/smart agents Reasoning and
  intentional notions
• Wrapper agents Facility for interaction with
  non-agents

8
Falacies What Agent-based Systems are not

• Computational X where X is from the social
  sciences such as the economics
• Agents are not middleware components
• Agents are not Grid Services
• Agents are not Internet software
• Agents need not dwell online
• Agent-based Systems are not necessarily
  decision-support systems
• Agent-based Systems do not necessarily employ AI
  methods
• Agents need not be implemented in specific
  programming languages or paradigms

9
Multi-agency

• A multi-agent system is a system that is made up
  of multiple agents with the following key
  features among agents to varying degrees of
  commonality and adaptation
• Social rationality
• Normative patterns
• System of Values
• e.g., eCommerce, space missions, Intelligent
  Homes
• The motivation is coherence and distribution of
  resources.

10
Summary of Business Benefits

• Modeling existing organizations and dynamics
• Modeling and Engineering E-societies
• New tools for distributed knowledge-ware

11
Two views of Multi-agency

• Constructivist Agents are rational in the sense
  of Newells principle of individual rationality.
  They only perform goals which bring them a
  positive net benefit without regard to other
  agents. These are self-interested agents.
• Sociality Agents are rational in the Jennings
  principle of social rationality. They perform
  actions whose joint benefit is greater than its
  joint loss. These are self-less, responsible
  agents.

12
Multi-agent assumptions and goals

• Agents have their own intentions and the system
  has distributed intentionality
• Agents model other agents mental states in their
  own decision making
• Agent internals are of less central than agents
  interactions
• Agents deliberate over their interactions
• Emergence at the agent level and at the
  interaction level are desirable
• The goals is to find some principles-for or
  principled ways to explore interactions

13
Abstract Architecture
action
action
actions
states
Environment
14
Architectures

• Deduction/logic-based
• Reactive
• BDI
• Layered (hybrid)

15
Abstract Architectures

• An abstract model ltStates, Action, S?Agt
• An abstract view
• S s1, s2, environment states
• A a1, a2, set of possible actions
• This allows us to view an agent as a function
• action S ? A

16
Logic-Based Architectures

• These agents have internal state
• See and next functions and model decision making
  by a set of deduction rules for inference
• see S ? P
• next D x P ? D
• action D ? A
• Use logical deduction to try to prove the next
  action to take
• Advantages
• Simple, elegant, logical semantics
• Disadvatages
• Computational complexity
• Representing the real world

17
Reactive Architectures

• Reactive Architectures do not use
• symbolic world model
• symbolic reasoning
• An example is Rod Brookss subsumption
  architecture
• Advantages
• Simplicity, computationally tractable, robust,
  elegance
• Disadvantages
• Modeling limitations, correctness, realism

18
BDI a Formal Method

• Belief states, facts, knowledge, data
• Desire wish, goal, motivation (these might
  conflict)
• Intention a) select actions, b) performs
  actions, c) explain choices of action (no
  conflicts)
• Commitment persistence of intentions and trials
• Know-how having the procedural knowledge for
  carrying out a task

19
Belief-Desire-Intention
Environment
belief revision
act
sense
Beliefs
generate options
filter
Desires
Intentions
20
A simplified BDI agent algorithm

• 1. B B0
• 2. I I0
• 3. while true do
• 4. get next percept r
• 5. B brf(B, r) //
  belief revision
• 6. Doptions(B,D,I) //
  determination of desires
• 7. I filter(B, D, I) //
  determination of intentions
• 8. p plan(B, I) //
  plan generation
• 9. execute p
• 10. end while

21
Correspondences

• Belief-Goal compatibility
• Des ? Bel
• Goal-Intention Compatibility
• Int ? Des
• Volitional Commitment
• Int Do ? Do
• Awareness of Goals and Intentions
• Des ? BelDes
• Int ? BelInt

22
Layered Architectures

• Layering is based on division of behaviors into
  automatic and controlled.
• Layering might be Horizontal (I.e., I/O at each
  layer) or Vertical (I.e., I/O is dealt with by
  single layer)
• Advantages are that these are popular and fairly
  intuitive modeling of behavior
• Dis-advantages are that these are too complex and
  non-uniform representations

23
Agent-Oriented Software Engineering

• AOSE is an approach to developing software using
  agent-oriented abstractions that models high
  level interactions and relationships.
• Agents are used to model run-time decisions about
  the nature and scope of interactions that are not
  known ahead of time.

24
AOSE Considerations Track 1

• Programming platforms (e.g., JACK) to support not
  just programming and design
• What, how many, structure of agent?
• Model of the environment?
• Communication? Protocols? Relationships?
  Coordination?

25
AOSE Considerations Track 2

• Extending UML to support agent communication,
  negotiation etc.
• Communication? Protocols? Relationships?
  Coordination?

26
Gaia- Wooldridge, et al

• The Analysis phase
• Roles model
• - Permissions (resources)
• - Responsibilities (Safety properties and
  Liveliness properties)
• - Protocols
• Interactions model purpose, initiator,
  responder, inputs, outputs, and processing of the
  conversation
• The Design phase
• Agent model
• Services model
• Acquaintance model

27
Scott DeLoachs MaSE
Roles
Agent Class Diagram
Conversation Diagram
Internal Agent Diagram
Deployment Diagram
28
Break 5 minutes
29
Content Outline

• I. Introduction
• 1. History and perspectives on MultiAgent
  Systems
• 2. Architectural theories
• 3. Agent Oriented Software Engineering
• Break 5 minutes
• II. Social agents
• 4. Sociality and social models
• 5. Dimensions for Developing a Social Agent
• Examples in Autonomy, Trust, Social Ties,
  Control, Team, Roles, Trust, and Norms
• Break 5 minutes
• 6. Agent as a member of a group...
• Values, Obligations, Dependence, Responsibility,
  Emotions
• III. Closing
• 7. Trends and open questions
• 8. Concluding Remarks

30
A Multiagent System Top level loop
Initialize Groups, Interconnections For agents 1-
n While (1) Sense (self, world,
others) Reason (self, others) Act (physical,
speech, social)
31
Inside an agent
While (1) Sense (self, world,
others) Determine attitude (self,
others) Reason (self, others) Act (physical,
speech, social)
32
What is Sociality?

• In interactions one individuals thinking,
  feeling, and/or doing affects another
  individual.
• ? may involve a social action, a social
  convention, and a personal rationality.

Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
33
What is Sociality?

• An individual may engage collectives in
  interaction of thinking, feeling, and/or doing.
• ? may involve a social action, a social
  convention, and a unit rationality.

Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
34
What is Sociality?

• An agent may engage a human in interaction of
  thinking, feeling, and/or doing.
• ? may involve a social action, a social
  convention, and a personal rationality.

Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
Think, Feel, Do
35
What is Social Action?

• Social actions produce different kinds of
  influences.
• For example actions involving Resources,
  Delegation, Permission, Help, and Service.

36
What is Social Convention?

• Social conventions prescribe transformations of
  social influences as well as shifts and changes
  in the transformations.
• Examples
• Interpersonal tactics such as reciprocity,
  scarcity, and politeness.
• Use of norms, values, plans, policies, protocols,
  and roles.
• Following a conversational policy.
• Emotional reactive responses
• Cooperation logics
• Adaptations and emergence rules

37
What is Personal/Unit Rationality?

• Personal/unit Rationality prescribes stance of an
  individual or a collective toward social
  conventions with respect to others.
• An agent/collective might choose to follow or
  abandon social conventions either with all agents
  or selectively.
• Social Rationality versus Individual Rationality

38
Putting it together (CEBACR) A social model of
interaction

• ltCognition,
• Emotions,
• Behaviors,
• Social Actions,
• Social Conventions,
• Personal/Unit Rationality,
• Embodimentgt

39
A Special Case of Do ? Do Sociality

• Do ? Do
• Actions are buy and sell
• Social Conventions are conventions of bartering.
• Personal/Unit Rationality is accounting for
  utilities of self or others. This can be simple
  or extend to issues of reciprocity and goodwill.

40
A Social Agent

• An agents that has to interact with people, other
  agent(s), where it is affected and can affect
  others cognitive states, emotions, and/or
  behavior via social actions, social conventions,
  a personal rationality.
• Generally, such agents are more complex than
  reactive agents and must include social
  perception in their deliberation.

41
A Social Agent

• We cannot merely add social modules to
  prefabricated agents. Social makeup of such
  agents are found in all aspects of their
  architecture and must be designed from the start.
• We must at least have access to an agents social
  model
• ltCognition, Emotions, Behaviors, Social actions,
  Social Conventions, Personal Rationalitygt

42
A Social Agent

• Socially intelligent agents are biological or
  artificial agents that show elements of
  (humanstyle) social intelligence. The term
  artificial social intelligence refers then to an
  instantiation of human-style social intelligence
  in artificial agents. (Dautehahn 1998)

43
Social Inference
Cognitive
Emotions in communication
Illocution in communication
Observing Interpersonal Exchanges
Goals and plans
Gesture
Body Language
Attitude
Capability
Commonalities in goals and plans
Inferred Attitudes and Relationships
Social ties
Psychological states
Benevolence
Dependence
Inferred Social Import
Trust Autonomy Power
Coherence Norms Values Team
Control
Sub-cognitive
44
Situatedness

• Physically situatedness promotes frequent
  sampling of physical environment, feedback via
  physical environment as in the Subsumption
  architecture
• Socially situatedness promotes frequent sampling
  of environment (gossip), feedback via social
  interaction to new agent architectures

45
Levels of Sociality

• There are many MAS or HAI problems that are
  deterministic and would not require social
  reasoning. I.e., agents actions would not depend
  on others and if so it is pre-determined. At
  best, sociality is a luxury.
• There are scenarios where sociality, explicit
  reasoning about other agents or human actions
  are critical and it is not all predetermined.
  This requires high level of sociality.

46
Social delegation

• E.g., X gives Y permission and authority to make
  decisions for their organization
• Social delegation differs from physical
  delegation in that agents will have a cognitive
  exchange in stead of a physical one.
• Models of social delegation might be economic
  (utilitarian), dependency (in-debtedness),
  power-based (authority), or democratic.

47
Dimensions for Developing a Social Agent
Social Environment
Culture
Multi-Agent
Emotions
Social and collaborative notions
Cultural shifts in institutions organizations
Public skills
Planning and learning abilities
Modeling other agents
Tasks Resources Ontologies
Adherence to norms, values, obligations, power,
org rules
Communication and exchange
Community
Communication and exchange
Awareness
Initiative, Autonomy, Power, Control,
Emergent Norms and roles
Anthropomorphism Language realism
Collaboration Trust, safety, flexible roles,
policies, preferences
Adaptation and changes in reasoning about basic
social notions
Emotions
Organization
Human
Team
48
Dimensions for Developing a Social Agent
Culture
Social Environment
Multi-Agent
Asynchronous Sit Aware Real-time Communication
Info sharing Coordination
Community
Organization
Human
Team
49
Social Environment

• Agents that are embedded in social environments
  must be designed to account for the following
  needs
• Social tasks
• Shared Resources
• Ontologies
• Public skills related to tasks and resources
  such as requesting and delegating

50
Agents in Public Service

• Interactions with the public beyond individuals
• Public libraries
• Museums
• Shopping malls
• Transportation stations
• Billboards and road signs

51
Multiagent

• Agents that can relate to other agents must be
  designed to account for the following needs
• Communication and exchange of information,
• Modeling other agents and rationality altruism
  and benevolence,
• Planning and learning abilities,
• Social and collaborative notions Autonomy,
  Values, Norms, Obligations, Dependence, Control,
  Responsibility, Roles, Preference, Power, Trust,
  Teaming, Persona.
• Emotional communication.

52
Agents in automation of dirty, dull, and
dangerous tasks

• Intelligent homes
• Factories
• Telecommunications
• Power Plants
• Investment
• Transportation
• Electronic Customer Relations Management
• Cross Organizational Relations

53
Multiagent Shared Autonomy Among Personal
Satellite Assistants
PSAs reason about commitments to teaming to
respond to alarms
54
Autonomy Sources

• Capability
• Social ties benevolence, permissions, peer
  pressure (autonomy norm), reciprocity, norm
  sanctions

55
Trust Can mean different things

• Expectation of partners competence- Cristiano
  Castelfranchi
• Expectation of partners benign intent- Diego
  Gambetti
• Trust as a reputation and a recommendation- Mike
  Schillo
• Correct Expectations about partners actions-
  Patha Dasgupta
• Trust as reliable contract- Svet Brainov

56
Social Ties

• Social ties between agents affects social
  relationships.
• Trust and autonomy are increased with stronger
  ties.
• Communities are more robust with ties.
• Network structures embody collective properties
  of their community.

Performance
Number of ties
57
Models of Trust and Autonomy 2002

• Trusting value(A, B, t) Capability(B, t)
  Benevolence(B, A, t) Delegation harmony(A,B)
• Autonomy value (A, t) Capability(A, t)
  Average Trust (A) Balance of reciprocity()

58
Terraforming Mars 2002

• Trust(Aj, Ak, t) Trust(Aj, Ak, t-1)
• (rate Trust(Aj,
  Ak, t-1)
• (rate (gain -
  investment))

59
Human

• Agents that can relate to humans socially must be
  designed to account for the following needs
• Communication and exchange of information,
• Human intent and preferences,
• Human need for anthropomorphic appeal,
• Nested representations of humans and agents,
• Human policies for interaction and guidance,
• Collaborative requirements, and
• Emotional communication

60
Trust

• Reasons for trust in agents
• Preference to delegate an human operator might
  want another agent who has more time or resources
  to carry out a task
• Human-agent relations Agents can use human their
  understanding of human models of trust to
  interact with humans

61
Autonomy

• Human-Agent Interaction
• Adjustable Autonomy

62
HAI Shared Autonomy between an Air Traffic
Control assistant agent and the human operator
ATC agent and human operators learn to share and
trade autonomies
63
HAI UCAV formations

• UCAVs reason about helping in attack situations
• HA power

64
Organization

• Agents that must operate in organizations must be
  designed to account for the following needs
• Awareness of organizational rules, and structure,
  
• Ability to evolve and recognize emergent norms
  and roles, and
• Adaptation and changes in reasoning about basic
  social notions.

65
Knowledge Management

• Data storage and retrieval functions
• Indigenous ontologies
• Norms and Policies
• Institutions

66
Norms

• Involve two or more agents. Each agent
  understands and shares them.
• Agents have power to not choose them.
• There is no direct rational account of them
  available to the agents.
• The bearer experiences an implicit or an
  explicit sanction or rewards for adoption.

67
City grid - 2003

• Collisions cost agents time and intersections
  are out for a period.
• Agents must reason about norms of stopping for
  traffic lights or not based on comparisons of
  their gains and losses relative to the society
• Adaptive norm revision outperforms prescriptive
  norm assignment

68
Multiagent Shared Autonomy Among Low-orbit
Satellites
Satellites learn to recruit and form teams for
collaborative image gathering
69
Roles

• Several agents can adopt it individually,
  independently, and concurrently. One agent may
  adopt several simultaneously. Several agents may
  adopt it as a group. In general we will call this
  the adopter.
• It is meaningful in the social context of other
  agents including (a) the adopters relationship
  to other agents and groups, (b) the agents
  mental attitudes about the social relationships,
  and (b) the available norms including obligations
  and responsibilities.
• There are typical capabilities associated with
  the adopter. If the adopter loses these abilities
  then the efficacy of the role is jeopardized.

70
Roles

• Networks of roles are more clearly seen in
  role-based access control.
• Role hierarchy and role grouping are useful for
  selecting subsequent roles Moffett and Lupu,
  1999, Na and Cheon, 2000.

71
Culture

• Agents that are culturally embedded must be
  designed to account for the following needs
• Ability to reason about adherence to norms,
  values, obligations, organizational rules, etc.,
  and
• Ability to recognize shifts in culture of their
  organizations and institutions

72
Agent Historians and Dictionaries

• Nuances of cultural shifts
• Norms
• Laws
• Institutions
• Collaborative filtering

73
Break 5 minutes
74
Content Outline

• I. Introduction
• 1. History and perspectives on MultiAgent
  Systems
• 2. Architectural theories
• 3. Agent Oriented Software Engineering
• Break 5 minutes
• II. Social agents
• 4. Sociality and social models
• 5. Dimensions for Developing a Social Agent
• Examples in Autonomy, Trust, Social Ties,
  Control, Team, Roles, Trust, and Norms
• Break 5 minutes
• 6. Agent as a member of a group...
• Values, Obligations, Dependence, Responsibility,
  Emotions
• III. Closing
• 7. Trends and open questions
• 8. Concluding Remarks

75
Agent as a member of a group...
agent
honors
handles
obligations
roles
partakes
goals
specifies
plans
member of
norms
institution
shares
relies on
inherits
partakes
set/ borrow
values (terminal goals)
contains
organization
group
partakes
76
Values

• "value" might mean
• assessment of usefulness of an object or action
  relative to a purpose, I.e., "(instrumental)
  evaluations", E.g., "this knife is good for chip
  carving ",
• (b) absolute assessment of desirability of
  something, I.e, principles, E.g., "honesty is
  good"
• Adding value to an agent enables it to generate
  internal desires as well as adds a level of
  behavior predictability for other agents.

77
Obligations

• Obligations capture all forms of social
  influence.
• Obligations have a strong deontological and
  motivational senses (more so than norms)
• Obligations are frequently assumed to have
  penalties associated with the failure to meet the
  obligation. We make no such assumption some
  obligations may have sanctions and some may not.

78
Responsibilities

• There are several types of responsibility
• Responsibility to concerns an agents obligation
  to perform an action.
• Responsibility for concerns an agents obligation
  to see that a state of affairs obtains.
• Responsibility about is the agents obligation
  to behave in accordance with its principles,
  which is general, abstract, and typically with
  respect to an agents immutable values.

79
Responsibilities, CAST project Yen, et al. 2001

• Agents are represented as nodes of a graph.
• One type of labeled directed edge is between two
  agents (A t? B), and it represents that A
  delegates t to B or conversely B is responsible
  to A with respect to t.
• The delegation relationships is non-reflexive,
  anti-symmetric, and transitive. The transitive
  property can be used to establish implied
  relationships.

80
The big picture
Values
Norms
Obligationsab (i.e., responsibility)
Autonomy
Dependenceba
Delegationba
Emerson, 1962 Tuomela, 2000
Trustba
Mayer, et al 1995
, Controlab)
(Powerab
Tuomela, 2000
81
Emotions

• Emotions provide possibilities for bypassing
  chains of reasoning to protect the agent in
  dangerous situations or to enable it to work with
  agents that have not been beneficial in the past.
  
• HAI quick feedback by human or agent human
  appeal
• Multiagent Appraisal of situations

82
Emotions

• Emotions Theories correspondence between
  emotions and behavioral situations. Feeling good
  a or bad into emotions
• Personality Theories Individual differences that
  affect emotional relationships

83
Content Outline

• I. Introduction
• 1. History and perspectives on multiagents
• 2. Architectural theories
• 3. Agent Oriented Software Engineering
• Break 5 minutes
• II. Social agents
• 4. Sociality and social models
• 5. Dimensions for Developing a Social Agent
• Examples in Autonomy, Trust, Social Ties,
  Control, Team, Roles, Trust, and Norms
• Break 5 minutes
• 6. Agent as a member of a group...
• Values, Obligations, Dependence, Responsibility,
  Emotions
• III. Closing
• 7. Trends and open questions
• 8. Concluding Remarks

84
Content Outline

• I. Introduction
• 1. History and perspectives on multiagents
• 2. Architectural theories
• 3. Agent Oriented Software Engineering
• II. Social agents
• 4. Sociality and social models
• 5. Autonomy, Team, Values, Norms, Obligations,
  Dependence, Control, Responsibility, Roles,
  Trust, Emotions
• III. Closing
• 6. Trends and open questions
• 7. Concluding Remarks

85
Current Trends

• Pervasive and emerging agent applications agent
  mediated e-commerce, emotional agents, embodied
  agents, virtual characters, conversational
  agents, etc.
• Standardization efforts FIPA.
• New Initiatives semantic web initiative.
• Agent tournaments RoboCup, Trading Agent
  Competition.

86
Concluding Remarks

• There are many uses for
• Agents These are highly intuitive and expressive
• Multiagent Systems These provide methods for
  defining institutions and working models of
  sociological theories
• Many open problems area available
• Theoretical issues for modeling social elements
  such as autonomy, power, trust, dependency,
  norms, preference, responsibilities, security,
• Adaptation and learning issues
• Communication and conversation issues

87
Further Explorations

• DAI-List_at_engr.sc.edu
• Agents.umbc.edu
• http//www.AgentLink.org/
• http//www.multiagent.com/
• http//homepages.feis.herts.ac.uk/comqkd/aaai-soc
  ial.html
• http//jasss.soc.surrey.ac.uk/5/4/4.html
• http//jom-emit.cfpm.org/
• http//www.stephenmarsh.ca/
• http//www.iiia.csic.es/
• http//www.salford.ac.uk/cve/va99/on-line99.htm

88
Further Explorations

• http//orgwis.gmd.de/projects/SocialWeb/
• http//bruce.edmonds.name/ssi/
• http//www.casos.ece.cmu.edu/home_frame.html
• http//bruce.edmonds.name/sfs/
• http//jasss.soc.surrey.ac.uk/4/1/contents.html
• http//www.isi.edu/teamcore/
• http//www.ecs.soton.ac.uk/nrj/soc-rat.html