Foundations of Multi-Agent Systems

Foundations of Multi-Agent Systems
Andrea Omicini
andrea.omicini@unibo.it
Dipartimento di Informatica – Scienza e Ingegneria (DISI)
Alma Mater Studiorum – Universit`a di Bologna a Cesena
18th European Agent Systems Summer School (EASSS 2016)
Catania, Italy, 25 July 2016
Andrea Omicini (DISI, Univ. Bologna) Foundations of MAS EASSS 2016, 25/7/2016 1 / 171

Outline
1 Complex Software Systems
2 Towards Agents
3 Autonomy
4 Agents
5 Building Agents & MAS
6 Hot Topics in MAS

Complex Software Systems
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Complex Software Systems Toward a Paradigm Change
Outline
Toward a Paradigm Change
Away from Objects
2 Towards Agents
Moving Toward Agent Technologies
The Many Agents Around
3 Autonomy
Autonomy in Dictionaries
Autonomy in Philosophy
Autonomy in Military
Autonomy in Social Sciences & AI
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming Agent Interaction
Programming MAS
6 Hot Topics in MAS

The Change is Widespread
software systems (present and forthcoming ones) are essentially
diﬀerent from “traditional” ones
the diﬀerence is widespread, and not limited to some application
scenarios
Computer science & software engineering are changing
dramatically
complexity is too huge for traditional CS & SE abstractions
like object-oriented technologies, or component-based methodologies
new computational paradigm? [Kuh62]

The Impending Crisis of Software
The Scenario of the (Ongoing) Crisis
Computing systems
are (going to be) anywere
are (going to be) embedded in every environment item/ object
are (going to be) always connected
wireless technologies are making interconnection pervasive
are (going to be) always active
to perform tasks on our behalf
more and more autonomously

Impact on Software Engineering
Which impact on the design & development of software systems?
quantitative in terms of computational units, software components,
number & size of interconnections, people involved,
time required, . . .
current processes, methods and technologies do not
scale up
qualitative new software systems are diﬀerent in kind
new features never experimented before

Novel Features of Complex Software Systems
situatedness computations occur as immersed within an environment
computations and environment mutually aﬀect each
other, and cannot be understood separately
openness systems are permeable and subject to change in size
and structure
locality in control components of a system are autonomous and
proactive loci of control
locality in interaction components of a system interact based on some
notion of spatio-temporal compresence on a local basis

This does to hold for MAS only
Historically, ﬁelds like
distributed artiﬁcial intelligence
manufacturing and environmental control systems
mobile computing
pervasive / ubiquitous computing
Internet computing
peer-to-peer (P2P) systems
got the news early, and tried to face the issues in terms of new models,
technologies, and methodologies

Situatedness: Examples
Control systems for physical domains
manufacturing, traﬃc control, home care, health care systems
explicitly aim at managing / capturing data from the environment
through event-driven models / event-handling policies
Sensor networks, robot networks
are typically meant to sense, explore, monitor and control partially
known / unknown environments

Situatedness I
Situated action [Suc87]
the notion of situated action stresses the relationship between an
action and its context of performance
actions are performed in a context: which aﬀects the actions, and is
aﬀected by them
the notion of environment is what is typically used here to denote the
(computational) context

Situatedness II
Environment as a ﬁrst-class entity [WOO07]
the notion of environment is explicit
components / computations interact with, and are aﬀected by the
environment
interaction with the environment is often explicit, too
Is this new?
every computation always occurred in some context
however, the environment is masked behind some “wrapping”
abstractions
environment is not a primary abstraction

Situatedness III
Does masking / wrapping work?
wrapping abstractions are often too simple to capture complexity of
the environment
when you need to sense / control the environment, masking it is not
always a good choice
environment dynamics is typically independent of system dynamics
the environment is often unpredictable and non-formalisable [Weg97]

Situatedness IV
Trend in CS and SE
drawing a line around the system, explicitly representing
what is inside in terms of component’s behaviour and interaction
what is outside in terms of environment, and system interaction with
the environment
predictability of components vs. unpredictability of the environment
this dichotomy is a key issue in the engineering of complex software
systems

Openness: Examples
Critical control systems
unstoppable systems, run forever
they need to be adapted / updated anyway, in terms of either
computational or physical components
openness to change, and automatic reorganisation are essential
features
Systems based on mobile devices
the dynamics of mobile devices is out of the system / engineer’s
control
system should work without assumptions on presence / activity of
mobile devices
the same holds for Internet-based / P2P systems

Openness
Permeable boundaries
‘drawing lines’ around systems does not make them isolated
boundaries are often just conventional, and allow for mutual
interaction and side-eﬀects
The dynamics of change
systems may change in structure, cardinality, organisation, . . .
technologies, methodologies, and (above all) abstractions should
account for modelling (possibly governing) the dynamics of change

Openness: Further Issues
Where is the system?
where do components belong?
are system boundaries for real?
“Mummy, where am I”?
how should components become aware of their environment. . .
. . . when they enter a system / are brought to existence?
How do we control open systems?
. . . where components come and go?
. . . where they can interact at their will?

Local Control: Examples
Cellular phone network
each cell with its own activity / autonomous control flow
autonomous (inter)acting in a world-wide network
World Wide Web
each server with its own (reactive) independent control flow
each browser client with its own (proactive) independent control flow

Local Control
Flow of Control
key notion in traditional systems and in Computer Science
multiple flows of control in concurrent / parallel computing
however, not an immediate notion to deal with in complex software
systems
a more general / abstract notion is required
Autonomy
is the key notion here
subsuming control flow
motivating multiple, independent flows of control
at a higher level of abstraction

Local Control: Issues of Autonomy
autonomy of execution is an eﬀective model for multiple independent
computational entities, since it prevents the issues of coupling
in an open world, autonomy of execution makes it easy for
components to move across systems & environments
autonomy of components more eﬀectively matches dynamics of
environment
SE principles of locality and encapsulation cope well with delegation
of control to autonomous components

Local Interactions: Examples
Control systems for physical domains
each control component is delegated a portion of the environment to
control
interactions are typically limited to the neighbouring portions of the
environment
strict coordination with neighbouring components is typically enforced
Mobile applications
local interaction of mobile devices is the basis for context-awareness
interaction mostly occurs with the surrounding environment
interoperation with neighbouring devices is typically enabled

Local Interactions
Local interactions in a global world
autonomous components interact with the environment where they
are located
interaction is limited in extension by either physical laws or logical
constraints
autonomous components interact openly with other systems
motion to and local interaction within the new system is the cheapest
and most suitable model
situatedness of autonomous components calls for context-awareness
a notion of locality is required to make context manageable

Summing Up
Complex software systems, then
made of autonomous components
locally interacting with each other
immersed in an environment—both components and the system as a
whole
system / component boundaries are blurred—they are conceptual
tools until they work
Change is ongoing
Computer Science is changing
Software Engineering is changing
a (sort of) paradigm shift is occurring—a revolution, maybe [ZP03]

Complex Software Systems Away from Objects
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Evolution of Programming Languages: The Picture
[Ode02]

Evolution of Programming Languages: Dimensions
Historical evolution
Monolithic programming
Modular programming
Object-oriented programming
Agent programming
Degree of modularity & encapsulation
Unit behaviour
Unit state
Unit invocation

Monolithic Programming
The basic unit of software is the whole program
Programmer has full control
Program’s state is responsibility of the programmer
Program invocation determined by system’s operator
Behaviour could not be invoked as a reusable unit under diﬀerent
circumstances
modularity does not apply to unit behaviour

Modular Programming
The basic unit of software are structured loops / subroutines /
procedures / . . .
this is the era of procedures as the primary unit of decomposition
Small units of code could actually be reused under a variety of
situations
modularity applies to subroutine’s code
Program’s state is determined by externally supplied parameters
Program invocation determined by CALL statements and the likes

Object-Oriented Programming
The basic unit of software are objects & classes
Structured units of code could actually be reused under a variety of
situations
Objects
have local control over variables manipulated by their own methods
variable state is persistent through subsequent invocations
object’s state is encapsulated
are passive—methods are invoked by external entities
modularity does not apply to unit invocation
object’s control is not encapsulated

Agent-Oriented Programming
The basic unit of software are agents
encapsulating everything, in principle
by simply following the pattern of the evolution
whatever an agent is
we do not need to deﬁne them now, just to understand their desired
features
Agents
could in principle be reused under a variety of situations
have control over their own state
are active
they cannot be invoked
agent’s control is encapsulated

Features of Agents
Before we deﬁne agents, we already know that. . .
. . . agents are autonomous entities
encapsulating their thread of control
they can say “Go!”
. . . agents cannot be invoked
they can say “No!”
they do not have an interface, nor do they have methods
. . . agents need to encapsulate a criterion for their activity
to self-govern their own thread of control

Dimensions of Agent Autonomy
Dynamic autonomy
Agents are dynamic since they can exercise some degree of activity
they can say “Go!”
From passive through reactive to active
Unpredictable / non-deterministic autonomy
Agents are unpredictable since they can exercise some degree of
deliberation
they can say “Go!”, they can say “No!”
and also because they are “opaque”—may be unpredictable to external
observation, not necessarily to design
From predictable to unpredictable through partially predictable

Objects vs. Agents: Interaction & Control
Message passing in object-oriented programming
Data flow along with control
data flow cannot be designed as separate from control flow
A too-rigid constraint for complex distributed systems. . .
Message passing in agent-oriented programming
Data flow through agents, control does not
data flow can be designed independently of control
With agents, complex distributed systems can be designed just based
on the design of the information flow

Agents Communication
Agents communicate
Interaction between agents is a matter of exchanging information
toward Agent Communication Languages (ACL) [Sin98]
Agents can be involved in conversations
they can be involved in associations lasting longer than the single
communication act
diﬀerently from objects, where one message just refer to one method

Philosophical Diﬀerences [Ode02] I
Decentralisation
Object-based systems are completely pre-determined in control:
control is essential centralised at design time
Agent-oriented systems are essentially decentralised in control
Emergence
Object-based systems are essentially predictable
Multi-agent systems are intrinsically unpredictable and
non-formalisable and typically give raise to emergent phenomena

Philosophical Diﬀerences [Ode02] II
Analogies from nature and society
Object-oriented systems have not an easy counterpart in nature
Multi-agent systems closely resembles existing natural and social
systems

Towards the Coexistence of Agents and Objects
Final issues from [Ode02]
Should we wrap objects to agentify them?
Could we really extend objects to make them agents?
How are we going to implement the paradigm shift, under the heavy
weight of legacy?
technologies, methodologies, tools, human knowledge, shared practises,
. . .

Towards Agents
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Towards Agents Moving Toward Agent Technologies
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Towards Seamless Agent Middleware
The ﬁrst question
how are we going to implement the paradigm shift, under the heavy
weight of legacy?
Mainstreaming Agent Technologies [OR04]
observing the state of agent technologies
focussing on agent middleware
devising out a possible scenario

The Technology Life-Cycle
A successful technology from conception to abandonment
ﬁrst ideas from research
premiere technology examples
early adopters
widespread adoption
obsolescence
dismissal
Often, however, this does not happen
new technologies fail without even being tried for real
which are the factors determining whether a technology will either
succeed or fail?

Dimensions of a Technology Shift
Technology scenario has at least three dimensions
programming paradigm → new technologies change the way in which
systems are conceived
development process → new technologies change the way in which
systems are developed
economical environment → new technologies change market equilibrium,
and their success is aﬀected by market situations
3-D space for a success / failure story
what is going to determine the success / failure of agent-based
technologies?

The Programming Paradigm Dimension I
Pushing the paradigm shift
evangelists gain space on media
technological geeks follow soon
drawbacks
too much hype may create unsupported expectations
perceived incompatibility with existing approaches
possible dangers for conceptual integrity

The Programming Paradigm Dimension II
Middleware for the paradigm shift
technology support to avoid unsupported claims
seamlessly situated agents vs. wrapper agents
communication actions towards agents
pragmatical actions towards objects
this allows agents to be used in conjunction with sub-systems
adopting diﬀerent component models

The Development Process Dimension
Accounting for real-world software development
availability of development methods & tools is critical
No technology is to be widely adopted without a suitable
methodological support
day-by-day developer’s needs should be accounted, too
Agent-Oriented Software Engineering Methodologies
adopting agent-based metaphors and abstractions to formulate new
practises in software engineering
current state of AOSE methodologies [CHMS14]
early development phases are typically well-studied
later phases are not, neither the tools, nor the ﬁne-print details

The Economical Environment Dimension I
Innovation has to be handled with care
stakeholders of new technologies may enjoy advantages of early
positioning
however, they often focus too much on novelty and product, rather
than on benefits and service
“we are different” alone does not help much
software is a quite peculiar product: nearly zero marginal cost, and
almost infinite production capability

The Economical Environment Dimension II
Agent-Oriented Middleware & Infrastructures
promoting agent-oriented technologies through integration with
existing object-oriented middleware & infrastructures
creating a no-cost space for agent technologies
where (agent) technologies are no longer “sold” as whole packages
whose choice do not require any design commitment
where however agents represent the most eﬀective choice for most
components
allow agent metaphors to add their value to existing systems with no
assumption on the component model
e.g. agent-based simulation frameworks, agent architectures for the
integration of AI techniques, ACL-based messaging services, . . .

Towards Agents The Many Agents Around
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Convergence Towards The Agent
Many areas contribute their own notion of agent
Artiﬁcial Intelligence (AI)
Distributed Artiﬁcial Intelligence (DAI)
Parallel & Distributed Systems (P&D)
Mobile Computing
Programming Languages and Paradigms (PL)
Software Engineering (SE)
Robotics

On the Notion of Intelligence in AI
Reproducing intelligence
AI is first of all concerned with reproducing intelligent processes and
behaviours, where
intelligent processes roughly denote internal intelligence—like
understanding, reasoning, representing knowledge, . . .
intelligent behaviours roughly represent external, observable
intelligence—like sensing, acting, communicating, . . .
Symbolic intelligence
classic AI promoted the so-called symbolic acceptation of (artificial)
intelligence
based on mental representation of the external environment
where the environment is typically oversimplified
and the agent is the only source of disruption

On the Notion of Agent in AI
Encapsulating intelligence
agents in AI have from the very beginning worked as the units
encapsulating intelligence
individual intelligence
within the symbolic interpretation of intelligence
Cognitive agents
AI agents are essentially cognitive agents
they are ﬁrst cognitive entities
then active entities
in spite of their very name, coming from Latin agens [agere]—the one
who acts

AI & Agents: A Note
Reversing perspective [OP06]
results from AI and MAS research are no longer easily distinguishable
agents and MAS work as the introductory metaphors to most of the
AI results
as exemplified by one of the most commonly used AI textbooks [RN02]
classic AI results on planning, practical reasoning, knowledge
representation, machine learning, and the like, have become the most
obvious and fruitful starting points for MAS research and technologies
it is quite rare nowadays that new findings or lines of research in AI
might ignore the agent abstractions at all
altogether, rather than a mere subfield of AI, agents and MAS could
be seen as promoting a novel paradigm, providing a new and original
perspective about computational intelligence and intelligent systems

On the Notion of Agent in DAI [Woo02]
Overcoming the individual dimension
No more agents
as single units encapsulating individual intelligence
acting alone within an oversimpliﬁed environment
Social acceptation of agency
Agents are
individuals within a society of agents
components of a multiagent system (MAS)
distributed within a distributed environment

Agent Features in DAI [OJ96]
A DAI agent. . .
. . . has an explicit representation of the world
. . . is situated within its environment
. . . solves a problem that requires intelligence
. . . deliberates / plans its course of actions
. . . is ﬂexible
. . . is adaptable
. . . learns

A DAI Agent Represents the World: What?
What should be represented?
What is relevant? What is not relevant?
more precisely, which knowledge about the environment is relevant for
an agent to eﬀectively plan and act?
so, which portion of the environment should the agent explicitly
represent somehow in order to have the chance to behave intelligently?
Representation is partial
Necessarily, an agent has a partial representation of the world
Its representation includes in general both the current state of the
environment, and the laws regulating its dynamics

A DAI Agent Represents the World: How?
The issue of Knowledge Representation (KR)
How should an agent represent knowledge about the world?
Representation is not neutral with respect to the agent’s model and
behaviour
and to the engineer’s possibilities as well
Choosing the right KR language / formalism
according to the agent’s (conceptual & computational) model
multisets of tuples, logic theories, description logics, . . .

A DAI Agent Represents the World: Consistency I
Perception vs. representation
Environment changes, either by agent actions, or by its own dynamics
Even supposing that an agent has the potential to observe all the
relevant changes in the environment, it can not spend all of its
activity monitoring the environment and updating its internal
representation of the world
So, in general, how could consistency of internal representation be
maintained? And to what extent?
in other terms, how and to what extent can an agent be ensured that
its knowledge about the environment is at any time consistent with its
actual state

A DAI Agent Represents the World: Consistency II
Reactivity vs. proactivity
An agent should be reactive, sensing environment changes and
behaving accordingly
An agent should be proactive, deliberating upon its own course of
actions based on its mental representation of the world
So, more generally, how should the duality between reactivity and
proactivity be ruled / balanced?

A DAI Agent Solves Problems
An agent has inferential capabilities
New data representing a new solution to a given problem
New knowledge inferred from old data
New methods to solve a given problem
New laws describing a portion of the world
An agent can change the world
An agent is equipped with actuators that provide it with the ability to
aﬀect its environment
The nature of actuators depends on the nature of the environment in
which the agent is immersed / situated
In any case, agent’s ability to change the world is indeed limited

A DAI Agent Deliberates & Plans I
An agent has a goal to pursue
A goal, typically, as a state of the world to be reached—something to
achieve
A task, sometimes, as an activity to be brought to an
end—something to do
An agent understands its own capabilities. . .
. . . in terms of actions, pre-conditions on actions, eﬀects of actions
. . . where “understands” roughly means that its admissible actions
and related notions are somehow represented inside an agent, and
there suitably interpreted and handled by the agent
perception should in some way interleave with action either to check
action pre-conditions, or to verify action eﬀects

A DAI Agent Deliberates & Plans II
An agent is able to build a plan of its actions
An agent builds possible plans of action according to
its goal/task
its knowledge of the environment
An agent
deliberates on the actual course of action to follow
then acts consequently

A DAI Agent is Flexible & Adaptable
Define flexible. Define adaptable.
What do these words exactly mean?
Adaptable / flexible with respect to what?
Can an agent change its goal dynamically?
Or, can it solve different problems in different contexts, or in
dynamics contexts?
Can an agent change its strategy dynamically?
The above properties are both important and potentially misleading,
since they are apparently intuitive, and everybody thinks he/she
understands them exactly

A DAI Agent Learns
What is (not) learning?
Learning is not merely agent’s change of state
Learning is not merely dynamic perception—even though this change
the agent’s state and knowledge
What could an agent learn?
New knowledge
New laws of the world
New inferential rules?
new ways to learn?
A number of areas insisting on this topic
Machine Learning, Abductive / Inductive Reasoning, Data Mining, Neural
Networks, . . .

DAI Agents: Summing Up
In the overall, a DAI agent has a number of important features
It has a (partial) representation of the world (state & laws)
It has a limited but dynamic perception of the world
It has inferential capabilities
It has a limited but well-known ability to change the world
It has a goal to pursue (or, a task to do)
It is able to plan its course of actions, and to deliberate on what to
do actually
Once understood what this means, it might also be ﬂexible and
adaptable
It learns, regardless of how this term is understood

A PL Agent is Autonomous in Control
Complexity is in the control flow
The need is to abstract away from control
An agent encapsulates control flow
An agent is an independent locus of control
An agent is never invoked—it merely follows / drives its own control
flow
An agent is autonomous in control
it is never invoked—it cannot be invoked

A PL Agent is neither a Program, nor an Object
An agent is not merely a program [Gra96]
A program represents the only ﬂow of control
An agent represents a single ﬂow of control within a multiplicity
An agent is not merely a “grown-up” object [Ode02]
An object is invoked, and simply responds to invocations
An agent is never invoked, and can deliberate whether to respond or
not to any stimulus

A P&D Agent is Mobile [FPV98]
An agent is not bound to the Virtual Machine where it is born
Reversing the perspective
it is not that agents are mobile
it is that objects are not
Mobility is then another dimension of computing, just made evident
by agents
A new dimension requires new abstractions
New models, technologies, methodologies
To be used for reliability, limitations in bandwidth, fault-tolerance, . . .

A Robotic Agent is Physical & Situated
A robot is a physical agent
It has both a computational and a physical nature
complexity of physical world enters the agent boundaries, and cannot
be conﬁned within the environment
A robot is intrinsically situated
Its intelligent behaviour cannot be considered as such separately from
the environment where the robot lives and acts
Some intelligent behaviour can be achieved even without any
symbolic representation of the world
non-symbolic approach to intelligence, or situated action approach
[Bro91]
Reactive architectures come from here

A SE Agent is an Abstraction
An agent is an abstraction for engineering systems
It encapsulate complexity in terms of
information / knowledge
control
goal / task
intelligence
mobility
Agent-Oriented Software Engineering (AOSE)
engineering computational systems using agents
agent-based methodologies & tools

A MAS Agent is a Melting Pot
Putting everything together
The area of Multiagent Systems (MAS). . .
. . . is today an independent research ﬁeld & scientiﬁc community
. . . building a coherent conceptual framework, by drawing from the
results of the many contributing areas
Summing up
A MAS agent is an autonomous entity pursuing its goal / task by
interacting with other agents as well as with its surrounding
environment
Its main features are
autonomy / proactivity
interactivity / reactivity / situatedness

A MAS Agent is Autonomous
A MAS agent is goal / task-oriented
It encapsulates control
Control is ﬁnalised to task / goal achievement
A MAS agent pursues its goal / task. . .
. . . proactively
. . . not in response to an external stimulus
So, what is new here?
agents are goal / task oriented. . .
. . . but also a MAS as a whole can be such
individual vs. global goal / task
how to make them coexist fruitfully, without clashes?

A MAS Agent is Interactive
Limited perception, limited capabilities
it depends on other agents and external resources for the achievement
of its goal / task
it should interact with other agents and the environment [Agr95]
communication actions & pragmatical actions
A MAS agent does not live in isolation
it lives within an agent society
it lives immersed within an agent environment
Key-abstractions for MAS
agents
society
environment

Autonomy
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Autonomy Autonomy in Dictionaries
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Oxford Dictionary of English (2nd Edition revised 2005)
Etimology
Early 17th cent.: from Greek autonomia, from autonomos ‘having its own
laws’, from autos ‘self’ + nomos ‘law’.
Dictionary
autonomy
the right or condition of self-government
a self-governing country or region
freedom from external control or inﬂuence; independence.
(in Kantian moral philosophy) the capacity of an agent to act in
accordance with objective morality rather than under the inﬂuence of
desires

Oxford Thesaurus of English (2nd Edition revised 2008)
Thesaurus
autonomy
self-government, independence, self-rule, home rule, sovereignty,
self-determination, freedom, autarchy;
self-suﬃciency, individualism.

Merriam-Webster I
Dictionary
autonomy
1 the quality or state of being self-governing; especially: the
right of self-government
2 self-directing freedom and especially moral independence
3 a self-governing state
Synonyms accord, free will, choice, self-determination, volition, will
Antonyms dependence (also dependance), heteronomy, subjection,
unfreedom

Merriam-Webster II
Thesaurus
autonomy
1 the act or power of making one’s own choices or decisions:
accord, free will, choice, self-determination, volition, will
2 the state of being free from the control or power of another:
freedom, independence, independency, liberty,
self-determination, self-governance, self-government,
sovereignty

Autonomy Autonomy in Philosophy
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Internet Encyclopedia of Philosophy I
Many acceptations of autonomy
general an individual’s capacity for self-determination or
self-governance
folk inchoate desire for freedom in some area of one’s life
personal the capacity to decide for oneself and pursue a course of
action in one’s life
moral the capacity to deliberate and to give oneself the moral law,
rather than merely heeding the injunctions of others
political the property of having one’s decisions respected, honored,
and heeded within a political context

Internet Encyclopedia of Philosophy II
Individual autonomy
after Kant, autonomy is an essential trait of the individual, and
strictly related with its morality, represented by some high-level
ethical principles
then, with the relation between its inner self and its individual actions
that is, mind and behaviour

Internet Encyclopedia of Philosophy III
Independence from oneself
a more demanding notion of autonomy requires not only
self-determination, but also independence from oneself
this conception is connected with notions of freedom and choice, and
(maybe) non-determinism
and requires the ability of reasoning on (and possibly changing) not
just one own course of actions, but one own goals

Autonomy Autonomy in Military
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Unmanned Systems Integrated Roadmap FY 2011-2036 I
Automatic vs. autonomous
Automatic systems are fully pre-programmed and act repeatedly and independently of
external influence or control. An automatic system can be described as
self-steering or self-regulating and is able to follow an externally given
path while compensating for small deviations caused by external
disturbances. However, the automatic system is not able to define the
path according to some given goal or to choose the goal dictating its
path.
Autonomous systems are self-directed toward a goal in that they do not require outside
control, but rather are governed by laws and strategies that direct their
behavior. Initially, these control algorithms are created and tested by
teams of human operators and software developers. However, if machine
learning is utilized, autonomous systems can develop modified strategies
for themselves by which they select their behavior. An autonomous
system is self-directed by choosing the behavior it follows to reach a
human-directed goal.

Unmanned Systems Integrated Roadmap FY 2011-2036 II
Four Levels of Autonomy for Unmanned Systems (Guy Edwards)
Various levels of autonomy in any system guide how much and how often
humans need to interact or intervene with the autonomous system:
human operated
human delegated
human supervised
fully autonomous

Unmanned Systems Integrated Roadmap FY 2011-2036 III
Common terms used for this concept are sliding autonomy or flexible autonomy. The goal is not
about designing a better interface, but rather about designing the entire autonomous system to
support the role of the warfighter and ensure trust in the autonomy algorithms and the system
itself. Table 3 contains the most commonly referenced description of the levels of autonomy that
takes into account the interaction between human control and the machine motions.
Table 3. Four Levels of Autonomy
Level Name Description
1 Human
Operated
A human operator makes all decisions. The system has no autonomous control of its environment
although it may have information-only responses to sensed data.
2 Human
Delegated
The vehicle can perform many functions independently of human control when delegated to do so. This
level encompasses automatic controls, engine controls, and other low-level automation that must be
activated or deactivated by human input and must act in mutual exclusion of human operation.
3 Human
Supervised
The system can perform a wide variety of activities when given top-level permissions or direction by a
human. Both the human and the system can initiate behaviors based on sensed data, but the system can
do so only if within the scope of its currently directed tasks.
4 Fully
Autonomous
The system receives goals from humans and translates them into tasks to be performed without human
interaction. A human could still enter the loop in an emergency or change the goals, although in practice
there may be significant time delays before human intervention occurs.

Unmanned Systems Integrated Roadmap FY 2011-2036 IV
Autonomy & Unpredictability
The special feature of an autonomous system is its ability to be
goal-directed in unpredictable situations.
This ability is a signiﬁcant improvement in capability compared to the
capabilities of automatic systems.
An autonomous system is able to make a decision based on a set of
rules and/or limitations.
It is able to determine what information is important in making a
decision.
It is capable of a higher level of performance compared to the
performance of a system operating in a predetermined manner.

Autonomy Autonomy in Social Sciences & AI
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Autonomy Autonomy in Social Sciences & AI
Autonomy as a Relational Concept [Cas95] I
Autonomy as a social concept
an agent is autonomous mostly in relation to other agents
autonomy has no meaning for an agent in isolation
Autonomy from environment
the Descartes’ problem: (human, agent) behaviour is aﬀected by the
environment, but is not depending on the environment
→ situatedness, reactiveness, adaptiveness do not imply lack of
autonomy

Agents
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Agents
Autonomy as the Core of Agents
Lex Parsimoniae: Autonomy
! autonomy as the essential feature of agents
? which other typical agent features may follow / descend from
this—somehow?
Computational Autonomy
agents are autonomous as they encapsulate (the thread of) control
control does not cross agent boundaries
only data (knowledge, information) do
agents have no interface, cannot be controlled / invoked
looking at agents, MAS can be conceived as an aggregation of
multiple distinct loci of control interacting with each other by
exchanging information

Agents
(Autonomous) Agents (Pro-)Act
Action as the essence of agency
the etimology of the word agent is from the Latin agens
so, agent means “the one who acts”
any coherent notion of agency should naturally come equipped with a
model for agent actions
Autonomous agents are pro-active
agents are literally active
autonomous agents encapsulate control, and the rule to govern it
→ Autonomous agents are pro-active by deﬁnition
where proactiveness means “making something happen”, rather than
waiting for something to happen

Agents
Agents are Situated
The model of action depends on the context
any “ground” model of action is strictly coupled with the context
where the action takes place
an agent comes with its own model of action
any agent is then strictly coupled with the environment where it lives
and (inter)acts
agents are in this sense are intrinsically situated [Suc87]

Agents
Agents are Reactive
Situatedness and reactivity come hand in hand
any model of action is strictly coupled with the context where the
action takes place
any action model requires an adequate representation of the world
any effective representation of the world requires a suitable balance
between environment perception and representation
→ Any effective action model requires a suitable balance between
environment perception and representation
however, any non-trivial action model requires some form of perception
of the environment—so as to check action pre-conditions, or to verify
the effects of actions on the environment
in this sense, agents are supposedly reactive to change

Agents
Are Autonomous Agents Reactive?
Reactivity as a (deliberate) reduction of proactivity
an autonomous agent could be built / choose to merely react to
external events
it may just wait for something to happen, either as a permanent
attitude, or as a temporary opportunistic choice
in this sense, autonomous agents may also be reactive
Reactivity to change
reactivity to (environment) change is a diﬀerent notion
this mainly comes from early AI failures, and from robotics
[Bro86, Bro91]
It stems from agency, rather than from autonomy—as discussed above

Agents
(Autonomous) Agents Change the World
Action, change & environment
any model for action brings about a notion of change
an agent acts in order to change something in the MAS
two admissible targets for change by agent action: an agent could act
in order to change. . .
(agent) . . . the state of another agent
since agents are autonomous, and only data ﬂow among
them, the only way another agent can change their state
is by providing them with some information
change to other agents essentially involves
communication actions
(environment) . . . the state of the environment
change to the environment requires pragmatical actions
which could be either physical or virtual depending on
the nature of the environment

Agents
Autonomous Agents are Social
From autonomy to society
from a philosophical viewpoint, autonomy only makes sense when an
individual is immersed in a society
autonomy does not make sense for an individual in isolation
no individual alone could be properly said to be autonomous
this also straightforwardly explain why any program in any sequential
programming language is not an autonomous agent per se
[Gra96, Ode02]
Autonomous agents live in a MAS
single-agent systems do not exist in principle
autonomous agents live and interact within agent societies & MAS
roughly speaking, MAS are the only “legitimate containers” of
autonomous agents

Agents
Autonomous Agents are Interactive
Interactivity follows, too
since agents are subsystems of a MAS, they interact within the global
system
by essence of systems in general, rather than of MAS
since agents are autonomous, only data (knowledge, information)
crosses agent boundaries
information & knowledge is exchanged between agents
leading to more complex patterns than message passing between
objects

Agents
Autonomous Agents Do not Need a Goal I
Agents govern MAS computation
by encapsulating control, agents are the main forces governing and
pushing computation, and determining behaviour in a MAS
along with control, agent should then encapsulate the criterion for
regulating the thread(s) of control

Agents
Autonomous Agents Do not Need a Goal II
Autonomy as self-regulation
the term “autonomy”, at its very roots, means self-government,
self-regulation, self-determination
“internal unit invocation” [Ode02]
this does not imply in any way that agents needs to have a goal, or a
task, to be such—to be an agent, then
however, this does imply that autonomy captures the cases of
goal-oriented and task-oriented agents
where goals / tasks / . . . play the role of the criteria for governing
control

Agents Deﬁning Agents
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Agents as Autonomous Entities [ORV08]
Deﬁnition (Agent)
Agents are autonomous computational entities
genus agents are computational entities
diﬀerentia agents are autonomous, in that they encapsulate control
along with a criterion to govern it
Agents are autonomous
from autonomy, many other features stem
autonomous agents are interactive, social, proactive, and situated
they might have goals or tasks, or be reactive, intelligent, mobile
they live within MAS, and interact with other agents through
communication actions, and with the environment with pragmatical
actions

“Weak” Notion of Agent
Four key qualities [WJ95]
Weak agents are
autonomous
proactive
reactive (to change)
social

Are Autonomous Agents Intelligent?
Intelligence helps autonomy
autonomous agents have to self-determine, self-govern, . . .
intelligence makes it easy for an agent to govern itself
while intelligence is not mandatory for an agent to be autonomous
however, intelligent autonomous agents clearly make sense

Are Autonomous Agents Mobile?
Mobility could be an expression of autonomy
autonomous agents encapsulate control
at the end of the story, control might be independent of the
environment where an agent lives—say, the virtual machine on which
it runs
mobile autonomous agents clearly make sense
even though mobility is not required for an agent to be autonomous

Do Autonomous Agents Learn?
Learning may improve agent autonomy
by learning, autonomous agents may acquire new skills, improve their
practical reasoning, etc.
in short, an autonomous agent could learn how to make a better use
out of its autonomy
learning autonomous agents clearly make sense
learning, however, is not needed for an agent to be autonomous

“Strong” Notion of Agent
Mentalistic notion [WJ95]
Strong agents have mental components such as
beliefs
desires
intentions
goals
plans
knowledge about the world
. . .
Intelligent agents and mental components
Intelligent autonomous agents are naturally (and quite typically) conceived
as strong agents

Agents Mentalistic Agents
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Modelling Agents with Mental States I
Modelling agents based on mental states. . .
. . . eases the development of agents exhibiting complex behaviour
. . . provides us with a familiar, non-technical way of understanding
and explaining agents
. . . allows the developer to build MAS by adopting the perspective of
a cognitive entity engaged in complex tasks—e.g., what would I do in
the same situation?
. . . simpliﬁes the construction, maintenance, and veriﬁcation of
agent-based applications
. . . is useful when the agent has to comunicate and interact with
users or other system entities

Modelling Agents with Mental States II
[Den07]
The intentional stance is the strategy of interpreting the
behaviour of an entity (person, animal, artifact, whatever) by
treating it as if it were a rational agent who governed its ‘choice’
of ‘action’ by a ‘consideration’ of its ‘beliefs’ and ‘desires’.
The scare-quotes around all these terms draw attention to
the fact that some of their standard connotations may be set
aside in the interests of exploiting their central features: their
role in practical reasoning, and hence in the prediction of the
behaviour of practical reasoners.

Modelling Agents with Mental States III
Agents with mental states
agents governing their behaviour based on internal states that mimic
cognitive (human) mental states
epistemic states representing agents knowledge—their knowledge on
the world
i.e., percepts, beliefs
motivational states representing agents objectives—what they aim to
achieve
i.e., goals, desires
the process of selecting one action to execute among the many
available based on the actual mental states is called practical
reasoning

Practical vs. Epistemic Reasoning
Practical reasoning is reasoning directed towards actions—the process of
ﬁguring out what to do in order to achieve what is desired
[Bra87]
Practical reasoning is a matter of weighing conﬂicting considerations for
and against competing options, where the relevant considerations are
provided by what the agent desires/values/cares about and what the agent
believes.
Epistemic reasoning is reasoning directed towards knowledge—the process
of updating information, replacing old information (no longer
consistent with the world state) with new information

Practical Reasoning
practical reasoning consists of two main cognitive activities
deliberation when the agent makes decision on what state of affairs
the agent desire to achieve
means-ends reasoning when the agent makes decisions on how to
achieve these state of affairs
the outcome of the deliberation phase are the intentions
what agent desires to achieve, or what he desires to do
the outcome of the means-ends reasoning phase is the selection a
given course of actions
the workflow of the actions the agent intends to adopt in order to
achieve its own goals expressed as intentions

Basic Architecture of a Mentalistic Agent
Perception
Action
Plans
Reasoning
Beliefs
Agent

The BDI Framework I
According to [DG11]
one of the most popular and successful framework for agent
technology is deﬁned by Rao and Georgeﬀ [RG92]
there, the notions of belief, desire, and intention are the core ones
hence, agents in this framework are mentalistic ones, and are typically
referred to as BDI agents

The BDI Framework II
beliefs represent at any time the agent’s current knowledge about
the world, including
information about the current state of the environment
inferred from perception devices
messages from other agents
internal information
desires represent a state of the world the agent is trying to achieve
intentions are the chosen means to achieve the agent’s desires, and are
generally implemented as plans and post-conditions
as in general it may have multiple desires, an agent can
have a number of intentions active at any one time
these intentions may be thought of as running
concurrently, with one chosen intention active at any
one time

The BDI Framework III
Besides these components, the BDI model includes
plan library — namely, a set of “recipes” representing the procedural
knowledge of the agent
event queue — where
events — either perceived from the environment or
generated by the agent itself to notify an update of its
belief base
internal subgoals — generated by the agent itself while
trying to achieve a desire
are stored

The BDI Framework IV
Plans & plan library
usually, BDI-style agents do no adopt ﬁrst principles planning at all
all plans must be generated by the agent programmer at design time,
which are then selected for execution at run time
pre-programmed plans are collected in the plan library
the planning done by agents consists entirely of context-sensitive
subgoal expansion, which is deferred until a subgoal is selected for
execution

The BDI Abstract Architecture
Accordingly, the abstract architecture proposed by [RG92] comprise three
dynamic and global structures representing agent’s Beliefs, Desires, and
Intentions (BDI), along with an input queue of events
update (write) and query (read) operations are possible upon the
three structures
update operation are subject to compatibility requirements
formalised constraints hold upon the mental attitudes
the events that the system is able to recognise could be either
external – i.e., coming from the environment – or internal ones—i.e.,
coming from some reﬂexive action
events are assumed to be atomic, and can be recognised after they
have occurred

Summing Up
Agents are autonomous computational entities
agents encapsulate control along with a criterion to govern it
from autonomy, many other features (more or less) stem
autonomous agents are interactive, social, proactive, and situated;
they might have goals or tasks, or be reactive, intelligent, mobile
they live within MAS, and interact with other agents through
communication actions, and with the environment with pragmatical
actions
Intelligent agents are core components for complex systems
strong vs. weak notion of agency
BDI architecture for mentalistic agents
mobility
learning capabilities

Building Agents & MAS
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Building Agents & MAS Paradigm Shifts
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Paradigm Shifts in Software Engineering
New classes of programming languages
new classes of programming languages come from paradigm shifts in
Software Engineering
new meta-models / new ontologies for artificial systems build up new
spaces
new spaces have to be “filled” by some suitably-shaped new (class of)
programming languages, incorporating a suitable and coherent set of
new abstractions
the typical procedure
first, existing languages are “stretched” far beyond their own limits,
and become cluttered with incoherent abstractions and mechanisms
then, academical languages covering only some of the issues are
proposed
finally, new well-founded languages are defined, which cover new spaces
adequately and coherently

The Problem of PL & SE Today
Things are running too fast
new classes of programming languages emerge too fast from the
needs of real-world software engineering
however, technologies (like programming language frameworks)
require a reasonable amount of time (and resources, in general) to be
suitably developed and stabilised, before they are ready for SE practise
→ most of the time, SE practitioners have to work with languages (and
frameworks) they know well, but which do not support (or,
incoherently / insuﬃciently support) required abstractions &
mechanisms
→ this makes methodologies more and more important with respect to
technologies, since they can help covering the “abstraction gap” in
technologies

Building Agents & MAS Programming Agents
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Agents Without Agent Languages
What if we do not have an agent language available?
for either theoretical or practical reasons, it may happen
we may need an essential Prolog feature, or be required to use Java
what we do need to do: (1) define
adopt an agent definition, along with the agent’s required / desired
features
choose agent architecture accordingly, and according to the MAS needs
define a model and the languages for agent actions, both
communicative and pragmatical
what we do need to do: (2) map
map agent features, architecture, and action model / languages upon
the existing abstractions, mechanisms & constructs of the language
chosen
thus building an agent abstraction layer over our non-agent language
foundation

The Agent Abstraction
AOP languages have agent as a basic abstraction
an agent-oriented programming (AOP) language should support one
agent definition (at least)
so, straightforwardly supporting mobility in case of mobile agents,
intelligence somehow in case of intelligent agents, . . . , by means of
well-defined language constructs
required agent features play a fundamental role in defining language
constructs

Agent Architectures
AOP languages support agent architectures [Woo02]
agents have (essential) features, but they are built around an agent
architecture, which deﬁnes both its internal structure, and its
functioning
an agent programming language should support one (or more) agent
architecture(s),
the agent architecture inﬂuences possible agent features

Agent Oriented Programming Languages (AOP) I
Programming languages for autonomous agents
weak agents
architecture to implement autonomous behaviour for distributed
systems
focus on conceptual integrity, interoperability, distribution, mobility,
. . .
agent-oriented middleware

Agent Oriented Programming Languages (AOP) II
Example
Jade a Java-based framework to develop agent-based applications
in compliance with the FIPA speciﬁcations for interoperable,
intelligent, multi-agent systems
where
FIPA stands for Foundation for Intelligent Physical Agents
http://www.fipa.org/
FIPA is the IEEE Computer Society standards organisation that
promotes agent-based technology and the interoperability of its
standards with other technologies

Agent Oriented Programming Languages (AOP) III
Programming languages for cognitive agents
mentalistic agents
either BDI [RG91] or other cognitive architectures
facilities and structures to represent internal knowledge, goals, . . .
architecture for implementing practical reasoning

Agent Oriented Programming Languages (AOP) IV
Examples
Jason (Brasil) http://jason.sourceforge.net/ Agent platform
and language for BDI agents based on AgentSpeak(L)
JADEX (Germany) http://www.activecomponents.org/ Agent
platform for BDI and Goal-Directed Agents
2APL (Netherlands) http://www.cs.uu.nl/2apl/ Agent
platform providing programming constructs to implement
cognitive agents based on the BDI architecture
3APL (Netherlands) http://www.cs.uu.nl/3apl/ A
programming language for implementing cognitive agents
ASTRA (Ireland) http://astralanguage.com/ A distributed /
concurrent programming panguage based on
AgentSpeak(TR).

Building Agents & MAS Programming Agent Interaction
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Intra- vs. Inter-Agent Languages I
Agent computation vs. agent interaction
agents have both
an inner behaviour, and
an external, observable behaviour
this reproduce the “computation vs. interaction / coordination”
dichotomy of standard programming languages [Weg97, GC92]
computation the inner functioning of a computational component
interaction the actions determining the observable behaviour of a
computational component

Intra- vs. Inter-Agent Languages II
What is new here?
agent autonomy is new
the observable behaviour of an agent as a computational component is
driven / governed by the agent itself
e.g., intelligent agents do practical reasoning—reasoning about
actions—so that computation “computes” over the interaction
space—in short, agent coordination
Languages for computation, languages for interaction
agent programming languages should be either / both
intra-agent languages to deﬁne (agent) computational behaviour
inter-agent languages to deﬁne (agent) interactive behaviour

Intra- vs. Inter-Agent Languages III
Example: Agent Communication Languages (ACL)
ACL are the easiest example of inter-agent language [Sin98]
they just deﬁne how agents speak with each other
however, these languages may have some requirements on internal
architecture / functioning of agents

Agent Communication Languages (ACL) I
Speech acts
inspired by the study of human communication
communication is based on direct exchange of messages between
agents
specifying agent communicative actions
speaking agent acts to change the world around
in particular, to change the belief of another agent
every message has three fundamental parts
performative the pragmatics of the communicative action
content the syntax of the communicative action
ontology the semantics of the communicative action

Agent Communication Languages (ACL) II
Examples
examples, working as standard protocols for information exchange
between agents
KQML Knowledge Query Manipulation Language
http://www.cs.umbc.edu/kqml/ [LF97]
FIPA ACL FIPA Agent Communication Language
http://www.fipa.org/repository/aclspecs.html
[FIP02]

Agent Observable Behaviour
AOP languages support agent model of action
agent observable behaviour is not just communication, agent actions
are not just communication actions
agents act through either communication or pragmatical actions
altogether, these two sorts of action define the admissible space for
agent’s observable behaviour
a communication language [Sin98] defines how agents speak to each
others
an action language [GL93] should define how an agent can act over its
environment
AOP languages should account for both sorts of languages

Building Agents & MAS Programming MAS
Outline
Away from Objects
2 Towards Agents
3 Autonomy
4 Agents
Deﬁning Agents
Mentalistic Agents
Paradigm Shifts
Programming Agents
Programming MAS
6 Hot Topics in MAS

Programming the Interaction Space I
The space of MAS interaction
inter-agent language roughly deﬁne the space of (admissible) MAS
interaction
inter-agent language should not be merely seen from the viewpoint of
the individual agent (subjective viewpoint)
the overall view on the space of (admissible) MAS interaction is the
MAS engineer’s viewpoint (objective viewpoint)
subjective vs. objective viewpoint over interaction [Sch01, OO03]

Programming the Interaction Space II
Enabling / governing / constraining the space of MAS interaction
a number of inter-disciplinary ﬁelds of study insist on the space of
(system) interaction
communication
dialogue / argumentation
coordination
organisation
security
workﬂow management
e-institutions / normative systems
. . .

Coordination I
Coordination
many diﬀerent deﬁnitions around [OZKT01]
in short, coordination is managing / governing interaction in any
possible way, from any viewpoint [Weg96]
Coordination in MAS is a middleware issue [OOR04]

Coordination II
Coordination models
coordination models for distributed systems [GC92] perfectly ﬁt MAS
as action languages [COZ00], according to the objective coordination
viewpoint [Sch01, OO03]
e.g., tuple-based language [Gel85]
as software integrators [Cia96], coordination models get together
agents within a MAS
coordination models govern MAS interaction by mediating agent
interaction through coordination primitives [COZ00]
agent societies can be built around coordination media
e.g., tuple spaces [Gel85]

Coordination III
Why coordination models in MAS?
micro-level individual goals
individual agents are built around some individual
goal/task, which represent the micro-level of MAS
macro-level social/global goals
agent societies / MAS are built around some global
goal/task, which represent the macro-level of MAS
agent societies can be built around coordination media
micro-level vs. macro-level
making everything work smoothly is typically the main
issue in governing the interaction space in MAS
thus, the main issue of MAS coordination

Coordination IV
Example
TuCSoN Tuple Centers Spread over the Network
http://tucson.unibo.it/ [LF97], a tuple-based
middleware for the agent coordination [OZ99].
In TuCSoN, the action language is basically represented by Linda-based
primitives—rd, in, out.

Hot Topics in MAS
Outline
2 Towards Agents
3 Autonomy
4 Agents
6 Hot Topics in MAS

Hot Topics in MAS
Technologies, Methodologies & Standards I
Developing usable technologies
a lot of experimental work, a number of nice agent technologies
developed
reliable agent technologies are under development
integration with common technology standards
Providing usable methodologies and tools
industry-level AOSE methodologies are needed [BGZ04, ZO04]
AO development process should be addressed, too [CHMS14]
the issue of tools is a key one

Hot Topics in MAS
Technologies, Methodologies & Standards II
Building a shared AO technology environment worldwide
FIPA started a decade ago [Fou05]
OMG and IEEE (FIPA committee) are taking the lead today
Applications, of course
some papers, already [PM08]

Hot Topics in MAS
Autonomous Systems
MAS for the design and implementation of autonomous systems
agents as autonomous components [HCF03]
social action [Cas98]
agent societies as aggregates featuring autonomy as an overall
properties
architecture for distributed autonomy [SO16]
Autonomous robots
physical autonomy [Ros14]
cognitive architectures for robot intelligence
EASSS 2016 tutorial on “Programming Autonomous Robots using
PROFETA and the BDI model” by Corrado Santoro
agents to model individual robots, MAS to model robot teams
workshop ARMS @ AAMAS 2016

Hot Topics in MAS
Simulation
Using MAS for simulating complex systems
real-world systems [Ode02]
simulation of complex software systems
workshop MABS @ AAMAS 2016
simulation for SE [MCOV13]

Hot Topics in MAS
Knowledge-Intensive Systems (KIS)
Agents for KIS
developing over distributed cognition [Kir99]
within knowledge-intensive environments (KIE) [MO15]
both intelligence and mobility are features, along with self-*
techniques
from ontologies to semantic coordination [NOV13]

Hot Topics in MAS
Multi-Paradigm Integration
Agents as a meta-level paradigm
subsuming OOP in AOP: wrapping objects [Ode02], using objects
[ORV08]
many language paradigms altogether [JBN+15]
middleware integration for complex systems: MAS & EBS
(event-based systems) [OM15]
EASSS 2016 tutorial on “Complex Event Recognition in Multi-Agent
Systems” by Alexander Artikis

Hot Topics in MAS
Complex Systems as MAS
Self-organising MAS
self-* techniques for MAS [DMSGK06]
mapping natural & social systems [Ode02]
self-* techniques for SE
Intelligent Complex Systems
encapsulating intelligence within individual agents
as well as within social abstractions [Cas98]
also adopting self-adaptation techniques

Conclusion
So What?
Motivations, basics & issues for agents and MAS
. . . time for questions. . .
. . . thanks for listening. . .
. . . and enjoy EASSS 2016!

References
References I
Philip E. Agre.
Computational research on interaction and agency.
Artificial Intelligence, 72(1-2):1–52, January 1995.
Special volume on computational research on interaction and agency, part 1.
Federico Bergenti, Marie-Pierre Gleizes, and Franco Zambonelli, editors.
Methodologies and Software Engineering for Agent Systems: The Agent-Oriented Software
Engineering Handbook, volume 11 of Multiagent Systems, Artificial Societies, and
Simulated Organizations.
Kluwer Academic Publishers, June 2004.
Michael E. Bratman.
Intention, Plans, and Practical Reason.
Harvard University Press, November 1987.
Rodney A. Brooks.
Achieving artificial intelligence through building robots.
Technical report, Massachussets Institute of Technology (MIT), May 1986.
Rodney A. Brooks.
Intelligence without representation.
Artificial Intelligence, 47:139–159, 1991.

References
References II
Cristiano Castelfranchi.
Guarantees for autonomy in cognitive agent architecture.
In Michael J. Wooldridge and Nicholas R. Jennings, editors, Intelligent Agents, volume 890
of Lecture Notes in Computer Science, pages 56–70. Springer Berlin Heidelberg, 1995.
ECAI-94 Workshop on Agent Theories, Architectures, and Languages (ATAL) Amsterdam,
The Netherlands 8–9 August 1994. Proceedings.
Cristiano Castelfranchi.
Modelling social action for ai agents.
Artiﬁcial Intelligence, 103(1):157–182, August 1998.
Massimo Cossentino, Vincent Hilaire, Ambra Molesini, and Valeria Seidita, editors.
Handbook on Agent-Oriented Design Processes.
Springer Berlin Heidelberg, 2014.
Paolo Ciancarini.
Coordination models and languages as software integrators.
ACM Computing Surveys, 28(2):300–302, June 1996.

References
References III
Paolo Ciancarini, Andrea Omicini, and Franco Zambonelli.
Multiagent system engineering: The coordination viewpoint.
In Nicholas R. Jennings and Yves Lesp´erance, editors, Intelligent Agents VI. Agent
Theories, Architectures, and Languages, volume 1757 of LNAI, pages 250–259.
Springer-Verlag, 2000.
6th International Workshop (ATAL’99), Orlando, FL, USA, 15–17 July 1999. Proceedings.
Daniel Dennett.
Intentional systems theory.
In Ansgar Beckermann and Sven Walter, editors, Oxford Handbook of the Philosophy of
Mind, chapter 19. Oxford University Press, 2007.
Aniruddha Dasgupta and Aditya K. Ghose.
BDI agents with objectives and preferences.
In Andrea Omicini, Sebastian Sardina, and Wamberto Vasconcelos, editors, Declarative
Agent Languages and Technologies VIII, volume 6619 of Lecture Notes in Computer
Science, pages 22–39. Springer Berlin Heidelberg, 2011.
Giovanna Di Marzo Serugendo, Marie-Pierre Gleizes, and Anthony Karageorgos.
Self-organization in multi-agent systems.
The Knowledge Engineering Review, 20(02):165–189, January 2006.

References
References IV
Foundation for Intelligent Physical Agents (FIPA).
Agent Communication Language Speciﬁcations, 2002.
Foundation for Intelligent Physical Agents.
FIPA home page, 2005.
Alfonso Fuggetta, Gian Pietro Picco, and Giovanni Vigna.
Understanding code mobility.
IEEE Transactions on Software Engineering, 24(5):342–361, May 1998.
David Gelernter and Nicholas Carriero.
Coordination languages and their signiﬁcance.
Communications of the ACM, 35(2):97–107, February 1992.
David Gelernter.
Generative communication in Linda.
ACM Transactions on Programming Languages and Systems, 7(1):80–112, January 1985.
Michael Gelfond and Vladimir Lifschitz.
Representing action and change by logic programs.
The Journal of Logic Programming, 17(2-4):301–321, November 1993.

References
References V
Stan Franklin Art Graesser.
Is it an agent, or just a program?: A taxonomy for autonomous agents.
In Nicholas R. Jennings Jörg P. Müller, Michael J. Wooldridge, editor, Intelligent Agents
III Agent Theories, Architectures, and Languages: ECAI’96 Workshop (ATAL) Budapest,
Hungary, August 12–13, 1996 Proceedings, volume 1193 of Lecture Notes In Computer
Science, pages 21–35. Springer, 1996.
Henry Hexmoor, Cristiano Castelfranchi, and Rino Falcone, editors.
Agent Autonomy.
Multiagent Systems, Artificial Societies, and Simulated Organizations. Springer US, 2003.
Howell Jordan, Goetz Botterweck, John Noll, Andrew Butterfield, and Rem Collier.
A feature model of actor, agent, functional, object, and procedural programming
languages.
Science of Computer Programming, 98:120–139, February 2015.
David Kirsh.
Distributed cognition, coordination and environment design.
In Proceedings of the 3rd European Conference on Cognitive Science (ECCS 1999), pages
1–11, 1999.

References
References VI
Thomas S. Kuhn.
The Structure of Scientiﬁc Revolutions.
University of Chicago Press, Chicago, IL, USA, 1962.
Yannis Labrou and Tim Finin.
Semantics and conversations for an agent communication language.
In Michael N. Huhns and Munindar P. Singh, editors, Readings in Agents, pages 235–242.
Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 1997.
Ambra Molesini, Matteo Casadei, Andrea Omicini, and Mirko Viroli.
Simulation in Agent-Oriented Software Engineering: The SODA case study.
Science of Computer Programming, 78(6):705—714, June 2013.
Special Section on Agent-oriented Design methods and Programming Techniques for
Distributed Computing in Dynamic and Complex Environments.

References
References VII
Stefano Mariani and Andrea Omicini.
Anticipatory coordination in socio-technical knowledge-intensive environments:
Behavioural implicit communication in MoK.
In Marco Gavanelli, Evelina Lamma, and Fabrizio Riguzzi, editors, AI*IA 2015, Advances
in Artiﬁcial Intelligence, volume 9336 of Lecture Notes in Computer Science, chapter 8,
pages 102–115. Springer International Publishing, 23–25 September 2015.
XIVth International Conference of the Italian Association for Artiﬁcial Intelligence, Ferrara,
Italy, September 23–25, 2015, Proceedings.
Elena Nardini, Andrea Omicini, and Mirko Viroli.
Semantic tuple centres.
Science of Computer Programming, 78(5):569–582, May 2013.
Special section: Self-Organizing Coordination.
James Odell.
Objects and agents compared.
Journal of Object Technologies, 1(1):41–53, May–June 2002.

References
References VIII
Gregory M.P. O’Hare and Nicholas R. Jennings, editors.
Foundations of Distributed Artiﬁcial Intelligence.
Sixth-Generation Computer Technology. John Wiley & Sons Ltd., hardcover edition, April
1996.
Andrea Omicini and Stefano Mariani.
Reconciling event- and agent-based paradigms in the engineering of complex systems: The
role of environment abstractions.
In Danny Weyns and Fabien Michel, editors, Agent Environments for Multi-Agent Systems
IV, volume 9068 of Lecture Notes in Computer Science, pages 117–130. Springer
International Publishing, November 2015.
4th International Workshop, E4MAS 2014 – 10 Years Later, Paris, France, May 6, 2014,
Revised Selected and Invited Papers.
Andrea Omicini and Sascha Ossowski.
Objective versus subjective coordination in the engineering of agent systems.
In Matthias Klusch, Sonia Bergamaschi, Peter Edwards, and Paolo Petta, editors,
Intelligent Information Agents: An AgentLink Perspective, volume 2586 of LNAI:
State-of-the-Art Survey, pages 179–202. Springer, 2003.

References
References IX
Andrea Omicini, Sascha Ossowski, and Alessandro Ricci.
Coordination infrastructures in the engineering of multiagent systems.
In Federico Bergenti, Marie-Pierre Gleizes, and Franco Zambonelli, editors, Methodologies
and Software Engineering for Agent Systems: The Agent-Oriented Software Engineering
Handbook, volume 11 of Multiagent Systems, Artificial Societies, and Simulated
Organizations, chapter 14, pages 273–296. Kluwer Academic Publishers, June 2004.
Andrea Omicini and Agostino Poggi.
Multiagent systems.
Intelligenza Artificiale, III(1-2):76–83, March-June 2006.
Special Issue: The First 50 Years of Artificial Intelligence.
Andrea Omicini and Giovanni Rimassa.
Towards seamless agent middleware.
In IEEE 13th International Workshops on Enabling Technologies: Infrastructure for
Collaborative Enterprises (WET ICE 2004), pages 417–422, 2nd International Workshop
“Theory and Practice of Open Computational Systems” (TAPOCS 2004), Modena, Italy,
14–16 June 2004. IEEE CS.
Proceedings.

References
References X
Andrea Omicini, Alessandro Ricci, and Mirko Viroli.
Artifacts in the A&A meta-model for multi-agent systems.
Autonomous Agents and Multi-Agent Systems, 17(3):432–456, December 2008.
Special Issue on Foundations, Advanced Topics and Industrial Perspectives of Multi-Agent
Systems.
Andrea Omicini and Franco Zambonelli.
Coordination for Internet application development.
Autonomous Agents and Multi-Agent Systems, 2(3):251–269, September 1999.
Special Issue: Coordination Mechanisms for Web Agents.
Andrea Omicini, Franco Zambonelli, Matthias Klusch, and Robert Tolksdorf, editors.
Coordination of Internet Agents: Models, Technologies, and Applications.
Springer Berlin Heidelberg, March 2001.
Michal Pˇechouˇcek and Vladim´ır Maˇr´ık.
Industrial deployment of multi-agent technologies: review and selected case studies.
Autonomous Agents and Multi-Agent Systems, 17(3):397–431, December 2008.
Special Issue on Foundations, Advanced Topics and Industrial Perspectives of Multi-Agent
Systems.

References
References XI
Anand S. Rao and Michael P. Georgeff.
Modeling rational agents within a BDI architecture.
In James F. Allen, Richard Fikes, and Erik Sandewall, editors, 2nd International Conference
on Principles of Knowledge Representation and Reasoning (KR’91), pages 473–484, San
Mateo, CA, 1991. Morgan Kaufmann Publishers.
Anand S. Rao and Michael P. Georgeff.
An abstract architecture for rational agents.
In Bernhard Nebel, Charles Rich, and William R. Swartout, editors, 3rd International
Conference on Principles of Knowledge Representation and Reasoning (KR ’92), pages
439–449, Cambridge, MA, USA, 25-29 October 1992. Morgan Kaufmann.
Proceedings.
Stuart J. Russell and Peter Norvig.
Artificial Intelligence: A Modern Approach.
Prentice Hall / Pearson Education International, Englewood Cliffs, NJ, USA, 2nd edition,
20 February 2002.

References
References XII
Bernd Rosslenbroich.
On the Origin of Autonomy. A New Look at the Major Transitions in Evolution.
History, Philosophy and Theory of the Life Sciences. Springer International Publishing,
2014.
Michael Schumacher.
Objective Coordination in Multi-Agent System Engineering. Design and Implementation,
volume 2039 of LNCS.
Springer, April 2001.
Munindar P. Singh.
Agent communication languages: Rethinking the principles.
IEEE Computer, 31(12):40–47, December 1998.
Giovanni Sartor and Andrea Omicini.
The autonomy of technological systems and responsibilities for their use.
In Nehal Bhuta, Susanne Beck, Hin-Yan Liu, Robin Geiß, and Claus Kreß, editors,
Autonomous Weapon Systems. Law, Ethics, Policy, chapter 3, pages 39–74. Cambridge
University Press, July 2016.

References
References XIII
Lucy A. Suchman.
Situated actions.
In Plans and Situated Actions: The Problem of Human-Machine Communication,
chapter 4, pages 49–67. Cambridge University Press, New York, NYU, USA, 1987.
Peter Wegner.
Coordination as constrained interaction.
In Paolo Ciancarini and Chris Hankin, editors, Coordination Languages and Models. First
International Conference, COORDINATION ’96 Cesena, Italy, April 15–17, 1996.
Proceedings, Lecture Notes in Computer Science, pages 28–33. Springer Berlin Heidelberg,
April 1996.
Peter Wegner.
Why interaction is more powerful than algorithms.
Communications of the ACM, 40(5):80–91, May 1997.
Michael Wooldridge and Nicholas R. Jennings.
Intelligent agents: Theory and practice.
Knowledge Engineering Review, 10(2):115–152, June 1995.

References
References XIV
Michael J. Wooldridge.
An Introduction to MultiAgent Systems.
John Wiley & Sons Ltd., Chichester, UK, March 2002.
Danny Weyns, Andrea Omicini, and James Odell.
Environment as a ﬁrst-class abstraction in multi-agent systems.
Autonomous Agents and Multi-Agent Systems, 14(1):5–30, February 2007.
Special Issue on Environments for Multi-agent Systems.
Franco Zambonelli and Andrea Omicini.
Challenges and research directions in agent-oriented software engineering.
Autonomous Agents and Multi-Agent Systems, 9(3):253–283, November 2004.
Special Issue: Challenges for Agent-Based Computing.
Franco Zambonelli and H. Van Dyke Parunak.
Towards a paradigm change in computer science and software engineering: A synthesis.
The Knowledge Engineering Review, 18(4):329–342, December 2003.

Extras
URL
Slides
on APICe
→ http://apice.unibo.it/xwiki/bin/view/Talks/MasfoundationsEasss2016
on SlideShare
→ http://www.slideshare.net/andreaomicini/foundations-of-multiagent-systems

Foundations of Multi-Agent Systems
Andrea Omicini
andrea.omicini@unibo.it
Dipartimento di Informatica – Scienza e Ingegneria (DISI)
Alma Mater Studiorum – Universit`a di Bologna a Cesena
18th European Agent Systems Summer School (EASSS 2016)
Catania, Italy, 25 July 2016

Foundations of Multi-Agent Systems

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to Foundations of Multi-Agent Systems (20)

More from Andrea Omicini (20)

Recently uploaded (20)

Foundations of Multi-Agent Systems