THE A

1
THE AA CONCEPTUAL FRAMEWORK FOR MODELLING
ENGINEERING COMPLEX SYSTEMS- PRELIMINARY NOTES
-

• ALESSANDRO RICCI
• joint work with Andrea Omicini and Mirko Viroli
• DEIS Università di Bologna
• CESENA

2
BACKGROUND

• DEIS group in Cesena / Bologna
• Coordination Models, Languages and
  Infrastructures for MAS
• Agents and MAS paradigm for engineering software
  systems
• Recent interest in MAS Complex Systems, along 2
  main different but related directions
• Complex Systems as MAS
• using MAS as a paradigm to understand, model,
  simulate complex systems
• biologic systems in particular
• cooperation with bio-engineers at DEIS
• gtgt SARAS TALK!
• Software MAS as Complex Systems
• models and tools to specify, design and build
  software MAS with complex system properties
• self-organisation
• specification, modelling

3
THE CONTRIBUTION OF THE TALKIN A SLIDE

• Rethinking the notion of environment in MA4CS
• are grids, fields and pheromones ?
• Introducing the AA conceptual framework
• modelling and engineering high-level
  computational environments for MA4CS
• Supporting tehcnologies TuCSoN and CARTAGO
• Contribution in
• desigining and engineering software MAS
• modelling ( simulating ) complex systems as MAS
  
• FOCUS OF THIS TALK

4
OUTLINE

• THE ROLE OF ENVIRONMENT IN MA4CS
• motivation for introducing AA
• THE AA CONCEPTUAL FRAMEWORK
• the artifact abstraction
• the role wrt complex systems
• CURRENT MODELS TECHNOLOGIES FOR IMPLEMENTING
  AA
• TuCSoN and CARTAGO

5
RECONSIDERING THE ROLE OF THE ENVIRONMENT

• PART ONE

6
BACKGROUNDCOMPLEX SYSTEMS AS MAS

• MAS - and Situated MAS - more and more considered
  as a good paradigm for XXX complex systems
• XXX modelling, simulating, engineering
• M4ACS workshops, MABS workshop, JAAS journal
• Among the distinguishing properties
• high-level of abstraction
• autonomy and pro-activity, situatedness,
  reactivity, social abilities, etc...
• interaction and social behaviour as primary
  dimensions
• organisation notions and models
• ...

7
CATCHING COMPLEX SYSTEMS PROPERTIES

• Systemic dimensions
• interaction and emergent behaviour
• Non-linear dynamic systems behaviour
• feedbacks
• stochastic behaviours
• Openness and open systems
• behaviours far from equilibrium (stable or not)
• Self-organising behaviour
• dissipative structures
• Autopoietic behaviours
• agents and organisations as autopoietic entities
• Hierarchical structures and multiple organisation
  layers
• ...

8
THE ROLE OF THE ENVIRONMENT IN MA4CS

• Sometime missing
• MAS as a dynamic collection / network of agents
  interacting through direct communication
• Grid-like environments
• widely adopted for Agent-Based Modelling
  Simulation
• E.g. Epstein Axtell artificial societies...
• Field-based / pheromone-based and a-like
• widely adopted in situated MAS approaches
• Eg. MMASS

MA4CS Workshop at ECCS06 (Oxford, Sept. 06)
AA for Modelling Simulating Complex Systems
9
REMARKS

• Lack of abstraction
• are grids, fields and pheromones abstract and
  expressive enough to model any useful notion of
  environment?
• what about environments for cognitive /
  intelligent agents?
• Lack of generalisation and specialisation
• can we identify some general abstraction
  functioning as basic building block to model
  high-level environments?
• specific enough to encapsulate environment
  characteristics?

10
ENVIRONMENT ROLE THE CASE OF HUMAN ACTIVITIES

• Key role of the environment in human working
  activities
• well-studied by approaches in social science and
  cognitive science
• Activity theory (AT), distributed cognition
• already applied by computer science related
  fields
• CSCW, HCI, DAI
• Context, mediation artifacts (AT)
• any (non-trivial) human activity is mediated by
  some kind of artifact
• basic building block to understand human
  environment
• designed and constructed by humans to support
  their activities
• any kind of resource, tool, device, ...
• Cognitive artifacts (Norman)
• artifacts designed taking into the account human
  cognitive capabilities
• Key role for emergent and self-organising
  collective behaviours
• artifacts as enablers and constrainers

11
AA AIMS AND OBJECTIVE REUSING AND APPLYING
SUCH A LESSON FROM HUMAN SOCIETY TO AGENT
SOCIETIES.
12
AA CONCEPTUAL FRAMEWORK

• PART TWO

13
ENVIRONMENT AS CONTEXT OF AGENT (SOCIAL)
ACTIVITIES

• Agent computational environment as first-class
  abstraction in modelling and engineering MAS...
• ...and modelling and engineering complex systems
  as MAS
• Modelling in particular what is constructed and
  co-used by agents during their activities
• e.g. passive resources, communication /
  coordination media, etc

E4MAS 2006 _at_ AAMAS 2006 Hakodate, Japan, 8/5/2006
Ricci, Viroli, Omicini - Cartago
14
THE AA (AGENTS AND ARTIFACT) APPROACH

• MAS (agents artifacts) grouped in workspaces
• Agents
• as the autonomous goal/task-oriented entities
  working in one or more workspaces
• Artifacts
• as the passive function-oriented entities
  constructed, shared, used, by agents during their
  activities
• Workspaces
• logic containers for agents and artifacts

E4MAS 2006 _at_ AAMAS 2006 Hakodate, Japan, 8/5/2006
Ricci, Viroli, Omicini - Cartago
15
THE ARTIFACT ABSTRACTION

• Basic building block to model engineer
  high-level computational environments
• analogous to artifacts in human societies
• representing resources, tools, computational
  devices in the agent world
• Function-oriented entities
• explicitly designed to embed some kind of
  functionality
• design-stance (vs. agent intentional stance)
• concept of use usage interface
• agents use artifacts through their usage
  interface
• agents communicate with other agents, agents use
  artifacts
• Strong social dimension
• artifacts as media of agent (social) activities
• coordination artifacts as artifacts providing
  coordination functionalities

16
WORKSPACES

• Analogous to human workspaces
• agents set of artifacts / tools needed and used
  in doing their activities
• Defining a topology for the environment
• structuring the set of agents and artifacts
• defining a scope for agent / artifact
  interactions
• which artifacts are observable and reachable by
  an agent
• Principles of composition and hierarchies
• workspaces containing workspaces

MA4CS Workshop at ECCS06 (Oxford, Sept. 06)
AA for Modelling Simulating Complex Systems
17
ARTIFACT ABSTRACTIONSOME PROPERTIES

• ...indeed different from agent properties....
• Predictability
• not-autonomous behaviour
• designed to provide a specific function
• usage interface as a I/O contract
• Inspectability, controllability malleability
• designed to be inspectable, controllable, tunable
  / malleable
• management interface
• Linkability
• linking / connecting artifacts together to get
  more complex / articulated artifacts
• pure cause / effect relationships

18
A FIRST MODEL

• Usage Interface
• To use it
• operations
• observable state / events
• Function Description
• Why to use it
• Operating Instructions
• How to use it

MA4CS Workshop at ECCS06 (Oxford, Sept. 06)
AA for Modelling Simulating Complex Systems
19
TOWARDS A THEORY OF ARTIFACTSIN AGENCY

• Theoretic and pragmatic investigation of the
  basic dimensions of agent / artifact interaction
• construction
• observation use
• agents act upon artifacts through their usage
  interface
• agents perceive artifacts observable state and
  events
• inspection manipulation
• Impact on agent individual and social reasoning
• investigating how (cognitive) artifacts can be
  suitably exploited by agents to support their
  cognitive activities

20
ARTIFACTS FOR MA4CSSOME KEY POINTS

• Abstraction and modularity
• Revisiting the notion of locality
• Observation and control (governance) of MAS
  interaction
• Balancing design and emergence

21
(1) ABSTRACTION MODULARITY

• Abstraction
• high-level brick for modelling and designing
  articulated environments
• in particular cognitive environments for
  cognitive agents
• heterogeneity
• different kind of artifacts for the same
  environment
• vs. low-level abs. approaches like grid-like /
  field-based approaches
• Modularity
• way to decompose and structure the environment

22
(2) PRINCIPLE OF LOCALITY REVISED

• Impact on the notion of locality
• locality depends on artifacts specific nature,
  function and use
• Examples in human society
• cell phones and TV artifacts (and related
  infrastructure)
• instant communication and synchronisation
  without being in the same spatial locality
• dashboards with post-it
• communication and synchronisation without being
  in the same temporal locality

MA4CS Workshop at ECCS06 (Oxford, Sept. 06)
AA for Modelling Simulating Complex Systems
23
(3) OBSERVATION GOVERNANCE OF MAS INTERACTION

• (Coordination) Artifacts as media of MAS
  interactions
• artifacts that enable, mediate and constraints
  agent interaction
• blackboards, channels, schedulers,
  note-boards,workflow engines, ...
• Natural place where to put mechanisms for
  observation and control
• from tracking / logging interaction...
• ...to online analysis / synthesis of global
  (social) properties
• Enabling intelligent agents observation, reason
  and control of global properties
• by properly using coordination artifacts
  observation / control interface
• enabling autonomic behaviours

24
(4) BALANCING DESIGN EMERGENCE

• Artifacts as the place where to balance design
  emergence dimension in systems
• Encapsulating the knowledge that designers /
  scientists have (or desire) about the environment
  (design)
• rules and constraints / norms
• Encapsulating basic mechanisms promoting the
  emergence of (desired) global behaviours
• analogous to stigmergic mechanisms provided by
  pheromone environments
• the example of cognitive stigmergy

25
COGNITIVE STIGMERGY EUMAS05,E4MAS06 BACKGROUND

• Stigmergy in human social behaviours and working
  activities
• CSCW, Cognitive Sciences
• Focus on working environments and artifacts
• WHAT mediated interaction, implicit
  communication
• HOW environment populated by artifacts
  (cognitive tools)
• triggers, place-holders, entry points,...
• WHY social behaviour emerges from individual
  actions over shared artifacts
• Cognitive/rational behaviours stigmergy
• human intelligent activities when using such
  tools
• observation, expectation
• stigmergic processes provided as artifact
  functionalities

26
EXAMPLES ALREADY THERE

• Some existing applications are already working
  based on these ideas
• platforms for sharing human knowledge, and create
  it incrementally...
• ...exploiting artificial (software) artifacts..
• ..supposing that users are rational!
• Cooperative work
• Wiki Wikipedia
• annotations as sorts of pheromones
• ranking, traces, history
• E-commerce
• Amazon
• customers who bought book A also bought book B
• building a relationship over knowledge/interests

27
COGNITIVE STIGMERGYMAIN INGREDIENTS

• Stigmergic principles
• local interaction of agents with the environment
• emergence of a global behaviour
• Cognitive agents (rational/intelligent)
• bringing about individual and social goals
• keeping a representation of the environment
• acting rationally
• ...not necessarily BDI-based...
• Suitable environments
• suitable set of artifacts...
• ...designed to make rational agents fruitfully
  interact

28
COMPUTATIONAL ENVIRONMENTS FOR COGNITIVE
STIGMERGY

• The environment as a shared field of work
• intelligent agents working in the same one..
• ...being aware of it...
• ...being aware that it is shared (and used by
  others)...
• ...being aware of the functionalities and
  opportunities it provides
• How to effectively model such a field of work in
  MAS?
• should rely on a notion of artifact coupling
  that of agent

29
A FRAMEWORK BASED ON ARTIFACTS

• 3-levels framework
• AGENT LEVEL
• users immersed in the same field of work
• e.g. a network for knowledge sharing
• TOOL LEVEL
• artifacts mediating agent access to the shared
  resources of the domain
• providing functionalities for cognitive stigmergy
• e.g. an annotation system and its management
• DOMAIN LEVEL
• artifacts target of agent working activities
• e.g. the wiki pages to be written/read

30
AN ARTIFACT-BASED ARCHITECTURE
31
FROM PHEROMONES TO ANNOTATIONS

• Symbolic annotations as counterpart of pheromones
• agents accessing resources and annotating them
• representing and embedding knowledge about the
  resources in the domain
• Knowledge / symbolic level
• both quantity and quality information
• Explicit/Implicit information
• annotations created intentionally by agents
• with a specific purpose (individual/social goal)
• annotations created automatically
• to track agent actions

32
ENCAPSULATING MECHANISMS FOR PROMOTING EMERGENCE

• Diffusion
• propagation of annotations according to some kind
  of logical topology (relationships between
  topics)
• Aggregation
• transforming set of annotations in a single
  annotation, according to various criteria
• positive feedbacks
• Selection and Ordering
• annotations ordered according to various
  relevance criteria
• freshness, pertinency, trust
• positive feedbacks
• selection to limit the information overload
• analogy with dissipation and evaporation
  mechanisms

33
FROM MODELLING TO SIMULATIONS

• Simulation as MAS execution
• simulations engineered on top of MAS
  infrastructure
• distributed simulations
• tools for online/offline analysis
• simulations as online experiments
• manual/automated observation control of system
  behaviour
• Benefits challenges
• natural deployment on computer networks /
  Internet
• conceptually scaling with system complexity
• loosing a notion of a global time and
  synchronisation
• any synchronisation must be achieved either by
  using artifacts or by communication

34
CURRENT TECHNOLOGIES FOR AA

• PART THREE

35
CURRENT TECHNOLOGIES FOR AA

• TuCSoN Omicini et al. 1999
• coordination model and infrastructure for MAS
• middleware providing first-class services for
  coordinating agents
• services as coordination artifacts
• CARTAGO Ricci, Viroli, Omicini, 2006
• middleware directly supporting the AA
  abstractions

36
TuCSoN INFRASTRUCTURE
37
TUPLE CENTRES

• Technically Programmable tuple spaces
• logic-based tuple space
• persistent communication spaces
• generative communication
• out, in, rd, inp, rdp primitives
• first-order logic (Prolog) facts as tuples
• programmable
• embedding general-purpose coordination laws,
  programmed as reactions in ReSpecT language
• General-purpose / programmable coordination
  artifacts
• coordination functionality programmed in ReSpecT

MA4CS Workshop at ECCS06 (Oxford, Sept. 06)
AA for Modelling Simulating Complex Systems
38
SIMPLE EXAMPLES
39
MODELLING ARTIFACTS ON-TOP-OF TUPLE CENTRES

• Usage interface modelled upon basic coordination
  primitives
• operations encoded as out, in, rd,...
• Artifact function encoded in ReSpecT language
• in terms of reactions to operation and events
  occurring inside the artifact
• Implementing artifacts for modelling complex
  system
• Examples
• artifacts for cognitive stigmergy
• dashboards, noteboards,...
• compartments in cell modelling / simulation
• Saras presentation

40
TUPLE CENTRES AS BIO-COMPARTMENTS
41
CARTAGO

• Middleware directly supporting the AA conceptual
  framework
• Orthogonal to the agent model
• promoting the integration with heterogeneous
  agent platforms
• Java used as implementation language
• Two parts
• First-class support / API to create new types of
  artifacts
• API available to agents to dynamically create and
  work with environments
• creating, sharing, using, manipulating artifacts
• Available as open-source project
• http//www.alice.unibo.it/projects/cartago

42
RECAPITULATION ANDCONCLUDING REMARKS

• LAST PART

43
RECAPITULATION

• AA contribution for modelling and engineering
  complex systems as MAS
• artifacts as first-class environmental
  abstraction
• role wrt complex systems typical properties
• defining encapsulating mechanisms for promoting
  self-organisation and emergent behaviours
• balancing designed emergent behaviour
• observation and governance of interactions
• Current technologies
• TuCSoN coordination infrastructure
• CARTAGO

44
CONCLUDING REMARKS

• Lots of work still to do, both for...
• AA as a paradigm for engineering software
  systems (not part of the talk)
• AA as a paradigm for modelling and simulating
  complex systems as MAS
• Among the issues (for the second point)
• formal models and theories grounding the approach
  
• engineering distributed simulations
• artifact shape global society properties
• investigating the relationships between artifacts
  shape and the global (emergent) properties of
  the society exploiting the artifacts
• artifact catalog for self-organising systems
• investigating general kinds of artifacts useful
  to support self-organising and emergent
  behaviours despite of the specific application
  domain

45
THANKS!

• END OF THE TALK