2. Outline
1. History and perspectives on
1. History and perspectives on
multiagents
2. Agent Architecture
2. Agent Architecture
3. Agent Oriented Software Engineering
4. Mobility
4. Mobility
5. Autonomy and Teaming
Multiagent Systems: R. Akerkar 2
3. Definitions
An agent is an entity whose state is viewed as consisting
of mental components such as beliefs, capabilities,
p , p ,
choices, and commitments. [Yoav Shoham, 1993]
. An entity is a software agent if and only if it
communicates correctly in an agent communication
language. [Genesereth and Ketchpel, 1994]
language [Genesereth and Ketchpel 1994]
. 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
l bl d
inferences, and determine actions. [Hayes‐Roth, 1995]
Multiagent Systems: R. Akerkar 3
4. Definitions
. An agent is anything that can be viewed as
An agent is anything that can be viewed as
(a)Perceiving its environment, and (b) Acting upon
that environment [Russell and Norvig, 1995]
. A computer system that is situated in some
environment and is capable of autonomous action in
its environment to meet its design objectives.
i i i d i bj i
[Wooldridge, 1999]
Multiagent Systems: R. Akerkar 4
6. Test of Agenthood [Huhns and Singh, 1998]
“A system of distinguished agents should
“A t f di ti i h d t h ld
substantially change semantically if a distinguished
agent is added.”
t i dd d ”
Multiagent Systems: R. Akerkar 6
7. Agents vs. Objects
g j
“Objects with attitude” [Bradshaw, 1997]
“Obj i h i d ” [B d h ]
Agents are similar to objects since they are
i il bj i h
computational units that encapsulate a state
and communicate via message passing
d i t i i
Agents differ from objects since they have a
strong sense of autonomy and are active
versus passive.
i
Multiagent Systems: R. Akerkar 7
8. Agent Oriented Programming, Yoav
Shoham
AOP principles:
1. The state of an object in OO p g
j programming has no g
g generic
structure. The state of an agent has a “mentalistic”
structure: it consists of mental components such as beliefs
and commitments
commitments.
2. Messages in object-oriented programming are coded in an
g j p g g
application-specific ad-hoc manner. A message in AOP is
coded as a “speech act” according to a standard agent
communication language that is application independent
application-independent.
Multiagent Systems: R. Akerkar 8
9. Agent Oriented Programming
Extends Peter Chen’s ER model,
E t d P t Ch ’ ER d l
Gerd Wagner
• Different entities may belong to different epistemic categories. There are
agents, events, actions, commitments, claims, and objects.
• We distinguish between physical and communicative actions/events.
Actions create events, but not all events are created by actions.
• Some of these modeling concepts are indexical, that is, they depend on
the perspective chosen: in the perspective of a particular agent, actions
of other agents are viewed as events, and commitments of other agents
are viewed as claims against them.
Multiagent Systems: R. Akerkar 9
10. Agent Oriented Programming
Extends Peter Chen’s ER model,
Gerd Wagner
g
• In the internal perspective of an agent, a commitment refers to a specific action
to be performed in due time, while a claim refers to a specific event that is
created by an action of another agent, and has to occur in due time.
• Communication is viewed as asynchronous point‐to‐point message passing.
We take the expressions receiving a message and sending a message as
synonyms of perceiving a communication event and performing a
communication act.
• There are six designated relationships in which specifically agents, but not
objects, participate: only an agent perceives environment events, receives and
sends messages, does physical actions, has Commitment to perform some
action in due time, and has Claim that some action event will happen in due
time.
Multiagent Systems: R. Akerkar 10
11. Agent Oriented Programming
Extends Peter Chen’s ER model,
E t d P t Ch ’ ER d l
Gerd Wagner
An institutional agent consists of a certain number of (institutional, artificial and
human) internal agents acting on behalf of it. An institutional agent can only
perceive and act through its internal agents.
Within an institutional agent, each internal agent has certain rights and duties.
There are three kinds of duties: an internal agent may have the duty to full
commitments of a certain type, the duty to monitor claims of a certain type, or
yp , y yp ,
the duty to react to events of a certain type on behalf of the organization.
A right refers to an action type such that the internal agent is permitted to
p
perform actions of that type on behalf of the organization.
yp g
Multiagent Systems: R. Akerkar 11
13. 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
b
Intelligent/smart agents: Reasoning and intentional
notions
Wrapper agents: Facility for interaction with non-
agents
Multiagent Systems: R. Akerkar 13
14. Multi‐agency
A multi‐agent system is a system that is made up of
multiple agents with the following key features among
p g g y g
agents to varying degrees of commonality and
adaptation:
• S i l ti
Social rationality
lit
• Normative patterns
• System of Values
e.g., HVAC, eCommerce, space missions, Soccer, Intelligent Home,
e g HVAC eCommerce space missions Soccer Intelligent Home
“talk” monitor
The motivation is coherence and distribution of resources.
Multiagent Systems: R. Akerkar 14
15. Applications of Multiagent Systems
Electronic commerce: B2B, InfoFlow, eCRM
N t
Network and system management agents: E.g., The
k d t t t E Th
telecommunications companies
Real‐time monitoring and control of networks: ATM
Real time monitoring and control of networks: ATM
Modeling and control of transportation systems: Delivery
Information retrieval: online search
Automatic meeting scheduling
Electronic entertainment: eDog
Multiagent Systems: R. Akerkar 15
16. Applications of Multiagent Systems (cont.)
Decision and logistic support agents:Military and Utility
Companies
Interest matching agents: Commercial sites like Amazon.com
User assistance agents: E.g., MS office assistant
Organizational structure agents: Supply‐chain ops
Industrial manufacturing and production: manufacturing cells
Personal agents: emails
Investigation of complex social phenomena such as evolution of
roles, norms, and organizational structures
Multiagent Systems: R. Akerkar 16
19. Multi‐agency: allied fields
DAI
MAS: (1) online social laws, (2) agents may adopt goals and adapt beyond any problem
laws
DPS: offline social laws
CPS: (1) agents are a ‘team’, (2) agents ‘know’ the shared goal
• In DAI, a problem is being automatically decomposed among
distributed nodes, whereas in multi‐agents, each agent chooses to
, g , g
whether to participate.
• Distributed planning is distributed and decentralized action
selection whereas in multi‐agents, agents keep their own copies a
selection whereas in multi agents agents keep their own copies a
plan that might include others.
Multiagent Systems: R. Akerkar 19
20. Multi‐agent assumptions and goals
• Agents have their own intentions and the system
has distributed intentionality
y
• Agents model other agents mental states in their
own decision making g
• 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
g p p p p
• The goals is to find some principles‐for or principled
ways to explore interactions
Multiagent Systems: R. Akerkar 20
22. MAS Orientations
Computational
Organization
Theory Databases
Sociology
Formal AI
Economics Distributed
Problem
Solving Cognitive
Psychology
Science
Systems Distributed
Theory Computing
Multiagent Systems: R. Akerkar 22
28. Logic‐Based Architectures
g
These agents have internal state
See and next functions and model decision making by a set of
g y
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
p , g , g
Disadvatages
Computational complexity
Representing the real world
Multiagent Systems: R. Akerkar 28
29. Reactive Architectures
Reactive Architectures do not use
h d
symbolic world model
symbolic reasoning
An example is Rod Brooks’s subsumption architecture
Advantages
Simplicity, computationally tractable, robust,
elegance
g
Disadvantages
Modeling limitations, correctness, realism
Multiagent Systems: R. Akerkar 29
30. Reflexive Architectures:
simplest type of reactive
architecture
Reflexive agents decide what to do without
regard to history –
regard to history purely reflexive
action : P A
Example ‐ thermostat
ction(s) = { off
on
if temp = OK
otherwise
Multiagent Systems: R. Akerkar 30
31. Reflex agent without state
(Russell and Norvig, 1995)
Multiagent Systems: R. Akerkar 31
36. Belief-Desire-Intention
Environment
belief act
sense
revision
Beliefs
generate
options
filter
Desires Intentions
Multiagent Systems: R. Akerkar 36
37. Why is BDI a Formal Method?
• BDI is typically specified in the language of modal logic with
p
possible world semantics.
• Possible worlds capture the various ways the world might develop.
Since the formalism in [Wooldridge 2000] assumes at least a KD
axiomatization f each of B D and I each of th sets of possible
i ti ti for h f B, D, d I, h f the t f ibl
worlds representing B, D and I must be consistent.
• A KD45 logic with the following axioms:
• K: BDI(a, , t) (BDI(a, , t) BDI(a, , t))
• D: BDI(a, t) not BDI(a, not , t)
• 4: B(a, , t) B( B(a, , t) )
• 5: (not B(a, , t)) B( not B(a, , t))
• K&D is the normal modal system
Multiagent Systems: R. Akerkar 37
38. A simplified BDI agent algorithm
1. B = B0;
2.
2 I := I0;
3. while true do
4.
4 get next percept ;
5. B := brf(B, ); // belief revision
6.
6 D:=options(B,D,I);
D:=options(B D I); // determination of desires
7. I := filter(B, D, I); // determination of intentions
8.
8 := plan(B I);
plan(B, // plan generation
9. execute
10.
10 end while
Multiagent Systems: R. Akerkar 38
39. Correspondences
• Belief-Goal compatibility:
D B l
Des Bel
• Goal-Intention Compatibility:
Int Des
• Volitional Commitment:
Int Do Do
• Awareness of Goals and Intentions:
Des BelDes
Int BelInt
Multiagent Systems: R. Akerkar 39
40. Layered Architectures
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
Multiagent Systems: R. Akerkar 40
41. Outline
1. History and perspectives on
1. History and perspectives on
multiagents
2. Agent Architecture
2. Agent Architecture
3. Agent Oriented Software Engineering
4. Mobility
4. Mobility
5. Autonomy and Teaming
Multiagent Systems: R. Akerkar 41
42. Agent‐Oriented Software
Engineering
AOSE is an approach to developing software using
agent‐oriented abstractions that models high level
interactions and relationships.
p
Agents are used to model run‐time decisions about
g
the nature and scope of interactions that are not
known ahead of time.
Multiagent Systems: R. Akerkar 42
43. Designing Agents:
Recommendations from H. Van Dyke Parunak’s (1996) “Go to the Ant”: Engineering Principles from Natural Multi-
Agent Systems, Annals of Operations Research, special issue on AI and Management Science.
1. Agents should correspond to things in the problem domain rather than to
h ld d h h bl d h h
abstract functions.
2. Agents should be small in mass (a small fraction of the total system), time (able
to forget), scope (avoiding global knowledge and action).
g ), p ( gg g )
3. The agent community should be decentralized, without a single point of control
or failure.
4. Agents should be neither homogeneous nor incompatible, but diverse.
Randomness and repulsion are important tools for establishing and
maintaining this diversity.
5. Agent communities should include a dissipative mechanism to whose flow they
can orient themselves, thus leaking entropy away from the macro level at
which they do useful work.
hi h h d f l k
6. Agents should have ways of caching and sharing what they learn about their
environment, whether at the level of the individual, the generational chain, or
y g
the overall community organization.
7. Agents should plan and execute concurrently rather than sequentially.
Multiagent Systems: R. Akerkar 43
44. Organizations
Human organizations are several agents, engaged in multiple
g g , g g p
goal‐directed tasks, with distinct knowledge, culture,
memories, history, and capabilities, and separate legal
, y, p , p g
standing from that of individual agents
Computational Organization Theory (COT) models information
production and manipulation in organizations of human and
computational agents
Multiagent Systems: R. Akerkar 44
45. Management of Organizational
Structure
O
Organizational constructs are modeled as
i ti l t t d l d
entities in multiagent systems
Multiagent systems have built in mechanisms
for flexibly forming, maintaining, and
for flexibly forming maintaining and
abandoning organizations
Multiagent systems can provide a variety of
stable intermediary forms in rapid systems
development
Multiagent Systems: R. Akerkar 45
48. Stages of Agent‐Oriented
Software Engineering
A Requirements: provided by user
A.
B. Analysis: objectives and invariants
B A l i bj ti d i i t
C. Design: Agents and Interactions
D. Implementation: Tools and techniques
Multiagent Systems: R. Akerkar 48
49. KoAS‐ Bradshaw, et al
Knowledge (Facts) represent Beliefs in which the agent has
confidence about
F t and Beliefs may b h ld privately or b shared.
Facts d B li f be held i t l be h d
Desires represent goals and preferences that motivate the agent to
act
Intentions represent a commitment to perform an action.
There is no exact description of capabilities
Life cycle: birth, life and death (also a Cryogenic state)
birth life,
Agent Types: KaOS, Mediation (KaOS and outside) , Proxy
(mediator between two KAOS agents), Domain Manager (agent
registration),
registration) and Matchmaker (mediator of services)
Omitted: Emotions, Learning, agent relationships, Fraud, Trust,
Security
Multiagent Systems: R. Akerkar 49
50. Gaia‐ Wooldridge, et al
g ,
The Analysis phase:
Roles model:
-PPermissions (
i i (resources))
- Responsibilities (Safety properties and Liveliness
properties)
-P t
Protocols
l
Interactions model: purpose, initiator, responder, inputs,
outputs, and processing of the conversation
The D i
Th Design phase:
h
Agent model
Services model
Acquaintance model
Omitted: Trust Fraud Commitment and Security
Trust, Fraud, Commitment, Security.
Multiagent Systems: R. Akerkar 50
51. TAEMS: Keith Decker and Victor Lesser
The agents are simple processors.
Internal structure of agents include (a) beliefs
(
(knowledge) about task structure, (b) states, (c) actions,
g ) ,( ) ,( ) ,
(d) a strategy which is constantly being updated, of what
methods the agent intends to execute at what time.
Omitted: Roles, Skills or Resources.
Multiagent Systems: R. Akerkar 51
52. BDI based Agent-Oriented Methodology
(KGR) Kinny Georgeff and Rao
Kinny,
External viewpoint: the social system structure
and dynamics.
Agent Model + Interaction Model.
g
Independent of agent cognitive model and
communication
Internal viewpoint: the Belief Model the Goal
Model,
Model, and the Plan Model.
Beliefs: the environment, internal state, the actions
, ,
repertoire
Goals: possible goals, desired events
Plans: state charts
Multiagent Systems: R. Akerkar 52
53. MaSE – Multi-agent Systems Engineering, DeLoach
Domain Level Design (Use AgML for Agent type
Diagram,
Diagram Communication Hierarchy Diagram and
Diagram,
Communication class Diagrams.)
Agent Level Design (Use AgDL for agent
conversation)
Component Design AgDL
System Design AgML
y g g
Languages:
AgML (Agent Modeling Language- a graphical
language)
AgDL (Agent Definition Language- the system level
behavior and the internal behavior of the agent)
Rich in communication, poor in social structures
communication
Multiagent Systems: R. Akerkar 53
54. Scott DeLoach’s MaSE
Sequence
Roles Tasks
Diagrams
Agent Class Conversation
Diagram Diagram
Internal Agent
Diagram
g
Deployment
Diagram
Multiagent Systems: R. Akerkar 54
55. The TOVE Project (1998) ; Mark Fox, et al.
• Organizational hierarchy: Divisions and sub-divisions
• Goals, sub-goals, their hierarchy (using AND & OR)
• Roles, their relations to skills, goals, authority, processes, policies
• Skills, and their link to roles
• Agents, their affiliation with teams and divisions Commitment,
Empowerment
• Communication links between agents: sending and receiving information.
information
Communication at three levels: information, intentions (ask, tell, deny…), and
conventions
(semantics). Levels 2 & 3 are designed using speech act.
• Teams as temporary group of agents
• Activities and their states, the connection to resources and the constraints.
• Resources and their relation to activities and activities states
• Constraints on activities (what activities can occur at a specific situation and
a specific time)
• Time and the duration of activities. Actions occur at a point in time and they
have duration.
• Situation
Shortcomings: central d i i making
Sh t i t l decision ki
Multiagent Systems: R. Akerkar 55
56. Agent-Oriented Programming (AOP): Yoav Shoham
• AGENT0 is the first AOP and the logical component of this
language is a quantified multi-modal logic.
• M t l state: beliefs, capabilities, and commitments (
Mental t t b li f biliti d it t (or
obligations).
• Communication: ‘request’ (to perform an action), ‘unrequest’
(to refrain from action), and ‘inform’ (to pass information).
Multiagent Systems: R. Akerkar 56
58. Outline
1. History and perspectives on
1 History and perspectives on
multiagents
2. Agent Architecture
hi
3. Agent Oriented Software
Engineering
4. Mobility
4. Mobility
5. Autonomy and Teaming
Multiagent Systems: R. Akerkar 58
59. Mobile Agents
g
[Singh, 1999] A computation that can change its location of execution (given a
suitable underlying execution environment), both
code
d
program state
[Papaioannou, 1999] A software agent that is able to migrate from one host to
[P i ] A f h i bl i f h
another in a computer network is a mobile agent.
[IBM] Mobile network agents are programs that can be dispatched from one
computer and transported to a remote computer for execution. Arriving at the
remote computer, they present their credentials and obtain access to local
p y y g g
services and data. The remote computer may also serve as a broker by bringing
together agents with similar interests and compatible goals, thus providing a
meeting place at which agents can interact.
Multiagent Systems: R. Akerkar 59
61. A paradigm shift:
Distributed Systems versus mobile code
Instead of masking the physical location of a component, mobile code
infrastructures make it evident.
Code mobility is geared for Internet‐scale systems ... unreliable
Programming is location aware ...location is available to the programmer
g g
Mobility is a choice ...migration is controlled by the programmer or at runtime by the
agent
Load balancing is not the driving force ...instead flexibility, autonomy and
disconnected operations are key factors
Multiagent Systems: R. Akerkar 61
62. A paradigm comparison:
2 Components, 2 Hosts, a Logic, a Resource, Messages, a Task
2 Components 2 Hosts a Logic a Resource Messages a Task
Remote Computation
In remote computation, components in the system are static,
In remote computation components in the system are static
whereas logic can be mobile. For example, component A, at Host
HA, contains the required logic L to perform a particular task T, but
does not have access to the required resources R to complete the
q p
task. R can be found at HB, so A forwards the logic to component B,
k b f d f d h l
which also resides at HB. B then executes the logic before returning
the result to A. E.g., batch entries.
HA HB
L, T R
HA L HB Compute
L R
result
Multiagent Systems: R. Akerkar 62
63. A paradigm comparison:
2 Components, 2 Hosts, a Logic, a Resource, Messages, a Task
2 Components 2 Hosts a Logic a Resource Messages a Task
Code on Demand
In Code on Demand, component A already has access to resource R.
However, A (or any other components at Host A) has no idea of the
logic required to perform task T. Thus, A sends a request to B for it to
forward the logic L. Upon receipt, A is then able to perform T. An
example of this abstraction is a Java applet, in which a piece of code
example of this abstraction is a Java applet in which a piece of code
is downloaded from a web server by a web browser and then
executed.
HA HB
R L
HA Send L HB
Compute R L
L
Multiagent Systems: R. Akerkar 63
64. A paradigm comparison:
2 Components, 2 Hosts, a Logic, a Resource, Messages, a Task
2 Components 2 Hosts a Logic a Resource Messages a Task
Mobile Agents
With the mobile agent paradigm, component A already has the logic L required
to perform task T, but again does not have access to resource R. This resource can
t f t k T b t i d t h t R Thi
be found at HB. This time however, instead of forwarding/requesting L to/from
another component, component A itself is able to migrate to the new host and
interact locally with R to perform T. This method is quite different to the previous
two examples, in this instance an entire component is migrating, along with its
two examples in this instance an entire component is migrating along with its
associated data and logic. This is potentially the most interesting example of all
the mobile code abstractions. There are currently no contemporary examples of
this approach, but we examine its capabilities in the next section.
HA HB
L R
HA A moves HB Compute
L R
A returns
Multiagent Systems: R. Akerkar 64
65. A paradigm comparison:
2 Components, 2 Hosts, a Logic, a Resource, Messages, a Task
2 Components 2 Hosts a Logic a Resource Messages a Task
Client/Server
Client/Server is a well known architectural abstraction that has been
employed since the first computers began to communicate. In this
example, B has the logic L to carry out Task T, and has access to
resource R. Component A has none of these, and is unable to
transport itself. Therefore, for A to obtain the result of T, it must
t t it lf Th f f A t bt i th lt f T it t
resort to sending a request to B, prompting B to carry out Task T.
The result is then communicated back to A when completed.
HA HB
L, R
HA request HB
L, R Compute
result
Multiagent Systems: R. Akerkar 65
66. Problems in distributed
Systems: J. Waldo
Latency: Most obvious, Least worrisome
y
Memory: Access, Unable to use pointers, Because memory is both
local and remote, call types have to differ, No possibility of shared
memory
Partial Failure: Is a defining problem of distributed computing, Not
possible in local computing,
Concurrency: Adds significant overhead to programming model,
No programmer control of method invocation order
we should treat local and remote objects differently.
Waldo, J., Wyant, G., Wollrath, A., Kendall, S., “A note on distributed
computing”, Sun Microsystems Technical Report SML 94‐29, 1994.
i ” S Mi T h i l R SML
Multiagent Systems: R. Akerkar 66
67. Mobile Agent Toolkit from IBM:
Basic concepts
Aglet. An aglet is a mobile Java object that visits aglet‐enabled hosts in a computer
network. It is autonomous, since it runs in its own thread of execution after arriving at
, , y p g
a host, and reactive, because of its ability to respond to incoming messages. g
Proxy. A proxy is a representative of an aglet. It serves as a shield for the aglet that
protects the aglet from direct access to its public methods. The proxy also provides
location transparency for the aglet; that is, it can hide the aglet’s real location of the
g
aglet.
Context. A context is an aglet's workplace. It is a stationary object that provides a
means for maintaining and managing running aglets in a uniform execution
environment where the host system is secured against malicious aglets. One node in a
computer network may run multiple servers and each server may host multiple
contexts. Contexts are named and can thus be located by the combination of their
C d d h b l d b h bi i f h i
server's address and their name.
Message. A message is an object exchanged between aglets. It allows for
synchronous as well as asynchronous message passing between aglets. Message
passing can be used by aglets to collaborate and exchange information in a loosely
i b d b l t t ll b t d h i f ti i l l
coupled fashion.
Future reply. A future reply is used in asynchronous message‐sending as a handler to
receive a result later asynchronously.
Identifier. An identifier is bound to each aglet. This identifier is globally unique and
immutable throughout the lifetime of the aglet.
Multiagent Systems: R. Akerkar 67
68. Mobile Agent Toolkit from
IBM: Basic operations
Creation. The creation of an aglet takes place in a context. The new aglet is assigned
an identifier, inserted into the context, and initialized. The aglet starts executing as
soon as it has been successfully initialized.
it h b f ll i iti li d
Cloning. The cloning of an aglet produces an almost identical copy of the original
aglet in the same context. The only differences are the assigned identifier and the fact
that execution restarts in the new aglet. Note that execution threads are not cloned.
Dispatching. Dispatching an aglet from one context to another will remove it from its
current context and insert it into the destination context, where it will restart
execution (execution threads do not migrate). We say that the aglet has been “pushed”
to its new context.
Retraction. The retraction of an aglet will pull (remove) it from its current context and
insert it into the context from which the retraction was requested.
Activation and deactivation. The deactivation of an aglet is the ability to temporarily
halt its execution and store its state in secondary storage. Activation of an aglet will
restore it in a context.
i i
Disposal. The disposal of an aglet will halt its current execution and remove it from its
current context.
Messaging. Messaging between aglets involves sending, receiving, and handling
messages synchronously as well as asynchronously.
Multiagent Systems: R. Akerkar 68
69. Outline
1. History and perspectives on
1 History and perspectives on
multiagents
2. Agent Architecture
hi
3. Agent Oriented Software
Engineering
4. Mobility
4. Mobility
5. Autonomy and Teaming
Multiagent Systems: R. Akerkar 69
70. Autonomy
•Target and Context: Autonomy is only meaningful in terms of
specific targets and within given contexts.
•Capability: Autonomy only makes sense if an agent has a capability
oward a target. E.g, a rock is not autonomous
•Sources of Autonomy:
Endogenous: Self liberty, Desire, Experience, Motivations
Exogenous: Social, Deontic liberty, Environments
•Implementations: Off-line and by design, Online with fixed cost
analysis,
anal sis Online learning
Multiagent Systems: R. Akerkar 70
71. Perspectives on Autonomy
Communication
Cognitive Science and AI
Organizational Science
Software Engineering
g g
Multiagent Systems: R. Akerkar 71
73. Situated Autonomy and Action Selection
enablers sensory communications
data
beliefs
situated
autonomy
communication
physical goal
goal
physical act communication
i ti
intention intention
Multiagent Systems: R. Akerkar 73
74. Shared Autonomy between an Air Traffic Control assistant
agent and the human operator- 1999
g p
Multiagent Systems: R. Akerkar 74
76. Team- Building Intuition
•Drivers on the road are generally not a team
•Race driving in a “draft” is a team
•11 soccer players declaring to be a team are a
team
•Herding sheep is generally a team
Agents change their autonomy, roles, coordination strategies
•A String Quartet is a team
Well organized and practiced Multiagent Systems: R. Akerkar 76
77. Team- Phil Cohen, et al
Phil Cohen, et al:
Shared goal and shared mental states
Communication in the form of Speech Acts is required for team formation
p q
Steps to become a team:
1.
1 Weak Achievement Goal (WAG) relative to q and with respect to a team
to bring about p if either of these conditions holds:
•The agent has a normal achievement goal to bring about p; that is, the agent
does
not yet believe that p is true and has p eventually being true as a goal.
•The agent believes that p is true, will never be true, or is irrelevant (that is, q is
false), but has as a goal that the status of p be mutually believed by all the team
members.
2. Joint Persistent Goal (or JPG) relative to q to achieve p just in case
1. They mutually believe that p is currently false;
2. They mutually know they all want p to eventually be true;
y y y y
3. It is true (and mutual knowledge) that until they come to mutually believe either
that
p is true, that p will never be true, or that q isSystems: R. Akerkar will continue to mutually
Multiagent false, they 77
78. Team- Phil Cohen, et al
•Requiring Speech Act Communication is too strong
•Requiring Mutual Knowledge is too strong
•Requiring agents to remain in a team until everyone knows
about the team qualifying condition is too strong
team-qualifying
Multiagent Systems: R. Akerkar 78
79. Team- Michael Wooldridge
With respect to agent i’s desires there is potential for
cooperation iff:
1. th i
1 there is some group g such th t i b li
h that believes that g can j i tl
th t jointly
achieve ; and either
2. i can’t achieve in isolation; or
3. i believes that for every action that it can perform that
achieves , it has a desire of not performing .
i performs speech act FormTeam to form a team iff:
1. i informs team g that the team J-can ; and
2 i requests team g t perform
2. t t to f
Team g is a PreTeam iff:
1. g mutually believe that it J-can
2. g mutually intends
Multiagent Systems: R. Akerkar 79
80. Team- Michael Wooldridge
•Onset of cooperative attitude is independent of knowing about
specific individuals
•Assuming agent knows about g is hard too simplistic
Assuming
•Requiring Speech Act Communication is too strong
•Requiring Mutual Knowledge is too strong
Multiagent Systems: R. Akerkar 80
81. Team- Munindar Singh
g
<agents, social commitments, coordination relationships>
Social commitments: <debtor, creditor, context, discharge
condition>
Operators: Create, Discharge, Cancel, Release, Delegate, Assign
Coordination relationships about events:
e is required by f
e disables f
e feeds or enables f
e conditionally feeds f
…
Multiagent Systems: R. Akerkar 81
82. Agent as a member of a group...
g g p
agent
honors
handles
roles obligations
partakes
specifies goals
plans member of
institution norms
s a s
shares
relies on
partakes inherits set/
values borrow contains
(terminal
goals) organization
group
partakes
Multiagent Systems: R. Akerkar 82
83. The big picture
Norms Values
Obligationsab (i.e., responsibility)
consent perfect
f t
agreement
Autonomyb
Dependenceba Autonomyb + Autonomya
Delegationba
D l ti
coordnation weak
agreement
coordnation Controlab
Trustba
definciency
d fi i
Powerab
Multiagent Systems: R. Akerkar 83
84. Concluding Remarks
Concluding Remarks
Th
There are many uses for
f
Agents
Agent‐based Systems
Agent Frameworks
Many open problems area available
Theoretical issues for modeling social elements
such as autonomy, power, trust, dependency,
norms, preference, responsibilities, security, …
f ibili i i
Adaptation and learning issues
Communication and conversation issues
Multiagent Systems: R. Akerkar 84