the nature of intelligent decision support systems adam maria gadomski [email protected]...
TRANSCRIPT
The Nature of Intelligent The Nature of Intelligent Decision Support SystemsDecision Support SystemsThe Nature of Intelligent The Nature of Intelligent Decision Support SystemsDecision Support Systems
Adam Maria GadomskiAdam Maria [email protected]
1997 ENEA copyright
IDSIDSSSIDSIDSSS
Workshop” Intelligent Decision Support Systems for Emergency Management”, Halden, 20-21 October 1997
p. 2Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSSsIDSSs ENEA, ERG-ING-TISGI, 97
• Contexts of IDSSs
• Theoretical Background
• Technologies and Examples
• Contexts of IDSSs
• Theoretical Background
• Technologies and Examples
Contexts of IDSSs
• Internet Contexts
• Application Context
• RTD Context
33
IDDS - Internet Context (Alta Vista, Infoseek)
Concept number of documents
• DSS (various types)............................… 10 000• DSS for Emergency Management ....……………….....…...... 200 • Operator Support Systems.................... 40
• IDSS ........................................... 100
• Decision-Making Model ....................….. 1900• Multi Agent Systems (MAS) ...............…. 1300• Intelligent Agents ..............................….. 5100• Knowledge Based & Expert Systems ..............................….. 7000
44
IDSSsIDSSs ---- Applications Context
Health
Business
Informatin acquisition
(Internet, Data Mining)
IDSS
Servicies & Public Administration
Industry
Operatorslevel
Manageriallevel
Routine
Routine
Emergency
Emergency
Emergency
Manageriallevel
Emergency
Emergency
55
IDSSsIDSSs ---- RTD Context
Information Systems & Data Processing
Numerical Simulation & Optimization
Knowledge Based & Expert Systems
Logic & Meta programming
Multi-Agent & Intelligent Agent Technologies
Neuro-Fuzzy Technologies
IDSS
INTERESTS and EXPANTIONCognitive sciences & philosophy
66
IDSSsIDSSs ---- Historical Context
1986 - Paradigms for Intelligent Decision Support, David D. Woods in "Intelligent Decision Support in Process Environments" (E. Hollnagel, editor); Springer-Verlag.
...Advances in AI are providing powerful new computational tools that greatly expand the potential to support cognitive activities in complex work environments (e.g., monitoring, planning, fault management, problem solving). The application of these tools, however, creates new challenges about how to "couple" human intelligence and machine power in a single integrated system that maximizes joint performance.
7777
IDSSsIDSSs ---- First ConclusionsHistorical Context
1988, APPROACHES TO INTELLIGENT DECISION SUPPORT,. Editor: R.G. Jeroslow, Georgia Institute of Technology, Atlanta, GA B. Jaumard, P.S. Ow and B. Simeone, A.
1990, Model of Action-Oriented Decision-Making Process: Methodological Approach, A.M.Gadomski, Proceedings of the "9th European Annual Conference on Human Decision Making and Manual Control", CEC JRC Ispra.
88
IDSSsIDSSs ---- First ConclusionsHistorical Context
K. Sycara, Utility theory in conflict resolution P.S. Ow and S.F. Smith, Viewing scheduling as an opportunistic problem-solving process S. De, A knowledge-based approach to scheduling in an F.M.S. T.L. Dean, Reasoning about the effects of actions in automated planning systems D.P. Miller, A task and resource scheduling system for automated planning F. Glover and H.J. Greenberg, Logical testing for rule-base management J.N. Hooker, Generalized resolution and cutting planes D. Klingman, R. Padman and N. Phillips, Intelligent decision support sytems: A unique application in the petroleum industry K. Funk, A knowledge-based system for tactical situation assessment R.R. Yager, A note on the representation of quantified statements in terms of the implication operation L.D. Xu, A fuzzy multiobjective programming algorithm in decision support systems S.D. Burd and S.K. Kassicieh, A Prolog-based decision support system for computing capacity planning From APPROACHES TO INTELLIGENT DECISION SUPPORT, .1988.
99
IDSSsIDSSs ---- First Conclusions
LIST OF QUESTIONS:
Why - emergency management?
Why - cognitive sciences ?
Why - advanced technologies ?
Why - now ?
Let’s go to experience-based and theoretical explanations
1010
IDSS IDSS --- Emergency Management
Characteristics of emergency / crisis domains
Characteristics of emergency managers (IDSS users)
Information available for emergency managers
Characteristics of decisions
1111
IDSSs IDSSs --- Emergency Management
Characteristics of emergency / crisis domains: industrial distributed infrastructure emergencies, it covers high-risk industrial plants accidents, industrial territorial disasters and calamites. In general, it is referred to a high risk, complex domain not formally structured, such as ports, territory with population, airport infrastructure, railways node, oil pipes systems, chemical industry, etc. and to adequate human organizations which contribute as executors and partners in emergency management. Especially - multi-events emergencies where previously prepared plans have to be changed or realized under unexpected conditions.
1212
IDSSs IDSSs --- Emergency Management
Characteristics of emergency managers
They have: qualitative weakly structured knowledge about emergency domain, semi-formal knowledge about competencies of their own organization and other potential partners of emergency managing. They have a strong managerial skill, direct human assistants, an access to different experts and to information about the emergency and resources state.They need to cooperate with other emergency managers.They work under stress. They are not computer specialists.
1313
IDSSs IDSSs --- Emergency Management
Information available: frequently, limited access to information, information is not complete, uncertain, on different levels of details, too much or too dense various information, difficult or time consuming access to specially requested data.
Characteristics of decisions : must be made under time and resources constrains. Every decision depends on risk evaluation and manager competencies. It is focused on what to do in emergency domain (not only how to do), who should intervene and who should serve as an expert. Planned and just activated actions can be not efficient and can require immediate modifications. Erroneous human decision can be cause of serious and essential losses. 1414
IDSSs - Theoretical Background
p. 15Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
Industrial Emergency Management
decision-making model.
passive DSS and active DSS, i.e. Intelligent Decision Support System
architectures & intelligent agents
demo-prototypes
IDSSs IDSSs ENEA, ERG-ING-TISGI, 97
Industrial Emergency Management
Industrial Emergency
A state of risk and/or losses generation:
a) which is over the level accepted by local
administration
b) which is caused by an industrial accidentManagement
A control of autonomous functional units by task communication in
order to achieve an expected goal in the predefined domain.
p. 16Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSS IDSS - ManagementManagement ENEA, Gad, 97
risk
losses
An qualitative indicator of the current state of physical objects proportional to the probability of an event which may generate losses, and to the value of the maximal losses could be caused to this object by such event .* Risk value can be assessed by event specialists or obtained from experts during knowledge acquisition.* Risk value depends on many attributes of the risk objects and attributes of its environment.
An qualitative/quantitative indicator of death, injury,destruction in human, economical, cultural and ecological/environmental sense
p. 17Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSS IDSS - - ManagementManagement ENEA, Gad 97
Emergency domain
control of (human) autonomous functional units,
afu,by comands which activate afu according to
emergency plans. or include specific tasks.autonomous functional units
autonomous functional units:
fire brigades,
police,
...
are characterized by competence (types of interventions), and access to information
sourcesgoal a state of the domain which emergency managers intend to obtain
(consider most important).
domain
p. 18Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSS ------------------IDSS ------------------ ENEA, ERG-ING-TISPI, 97
Emergency
manager
Emergency
manager
Emergency
manager
Experts Executor 1
(afu)
Executors N
(afu)
. . .
different roles
Emergency
Supervisor
EMERGENCY DOMAIN
cooperation
DOMAIN OF ACTIVITYDOMAIN OF ACTIVITY
OF EMERGENCYOF EMERGENCY
MANAGERMANAGER
DOMAIN OF ACTIVITYDOMAIN OF ACTIVITY
OF EMERGENCYOF EMERGENCY
MANAGERMANAGER
. . .
coordination
taskstasks
tasks
cooperation
p. 19Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
actions
IDSS IDSS ENEA, ERG-ING-TISGI, 97
Industrial Emergency Management
decision-making model
Passive DSS and active DSS, i.e. Intelligent Decision Support System
abstract intelligent agents
demo-prototypes
p. 20Some references: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSS IDSS ENEA, ERG-ING-TISGI, 97
Definitions
Decision-making (d-m) is a mental activity implied by the necessity of a choice either • without known criteria or • without known alternatives.
Decision - a result of the choice.
reasoning pathcriticalnode
alternatives
d-mdata decision
??
??
decision
2121
IDSSIDSS ENEA, A.M.Gadomski, 97
decision-making model
Requires definitions of a reasoning mechanizm and the following relative concepts:
--
I - InformationI - Information
P - PreferencesP - Preferences
K - KnowledgeK - Knowledge
Decision (intervention)
domain
DDDD
Decision (intervention)
domain
DDDD
2222
IDSSIDSS ENEA, A.M.Gadomski, 97
Simplified action-oriented decision-making model
Let’s assume:
IIyy = K = Kii ( I ( Ix x ); ); oror K Kii: I: Ix x IIyy IIyy = K = Kii ( I ( Ix x ); ); oror K Kii: I: Ix x IIyy
I represents states/situation/changes of the decision domain, DD
Ki represents an inference association on DD
Ko represents an available operation on DD
AAy y represents an action on DD.
AAyy = K = Koo ( I ( Ix x ); ); oror K Koo: I: Ix x AAyy AAyy = K = Koo ( I ( Ix x ); ); oror K Koo: I: Ix x AAyy
2323
IDSSIDSS ENEA, A.M..Gadomski,
97
Simplified decision-making model
IIinin’ = P ( I’ = P ( Iin in ;I;I ); ); oror P: ( I P: ( Iin in ;I);I) IIinin’’ IIinin’ = P ( I’ = P ( Iin in ;I;I ); ); oror P: ( I P: ( Iin in ;I);I) IIinin’’
P represents a preference relation on DD.
IIin in denotes the currently preferred state of DD, it can
be called intention, max. intention can be called goal.
I I denotes current state of DD.
A Preference depends on the parameter IM :
P( X;I ): If intention_is X and I IM then intention_is Y
what is equivalent to the sentence:
In the state of DD from the class IM, Y is_better then X .2424
IDSSIDSS ENEA, A.M..Gadomski,
97
Simplified decision-making model
In such manner we can construct reasoning pathes on the sets of Preferences and Knowledge.
In such manner we can construct reasoning pathes on the sets of Preferences and Knowledge.
In the reasoning processes modelling,
P, KP, Ki i , K, Koo can be, in natural way, represented by rules and operations (algorithms) on the level of a DD model, i.e. they are referred to classes of information employed in the model.
2525
IDSSIDSS ENEA, A.M..Gadomski,
97
Simplified decision-making model
K1K3
K4K2
A1
An example of the interference path
K9
A2
K5
K6
Decisional node
Here, we may demostrate that for the decison-making we need
or new information or new preferences or new knowledge.
I
2626
Decision-making (d-m) is a mental activity implied by the necessity of choice either without known criteria or without known alternatives.
• The criteria are meta-preferences• The alternatives are possible actions
IDSSIDSS ENEA, A.M..Gadomski,
97
Simplified decision-making model
K6
Decisional node
A1
A2
IxIx
= K6 ( I= K6 ( Ix x ););A1A1
A2A2 ]1st type of Decision-Making rules ( meta-preferences):
mP: ( if mP: ( if A1A1 AX andAX and A2 A2 AY; Ix IM then A2 A2 ) 2727
IDSSIDSS ENEA, A.M..Gadomski,
97
mP: ( if mP: ( if A1A1 AX andAX and A2 A2 AY; Ix IM thenthen A2 A2 )
where AX , , AY are classes of actions of the
decision-maker, and IM is a class of the states of DD.
In such conceptualization, IDSS has to have a fixed base of mP rules, such as (in informal way):
if is a fire then activation of fire-men is better than activation of police station.
2828
IDSSIDSS ENEA, A.M..Gadomski,
97
Passive DSS and active DSS, i.e. Intelligent Decision Support System
Passive classical DSSs provides information
Active Intelligent DSSs suggest possible actions (knowledge) and inform about used criteria (preferences).
2929
• Unfortunately, their application requires from
their users continous learning and training to
which typical emergency managers are not
enough motivated
• Large part of the user decisions relies on the
choice of the concrete button from menubars or
menutools being parts of a visualized
hierarchical menu structures (menu-driven
paradigm)
Passive DSS gives data and tool choice
for decision making.
Passive DSS
3030
EMERGENCY MANAGER
Computer network
EMERGENCY DOMAIN
( Human)
Human Organization
cooperation
Interventiondecisions Continuous
monitoring
Images,MeasuredData
Data request
dialoguemenu-driven
Computer specialists
Assistance of
DATA BASES MANAGEMENT SYSTEMSFunctional algorithmsGeographical DBDangerous Materials DBEmergency organiz. DBPlannes, Instructions DB
PassiveDECISIONSUPPORTSYSTEM
DSS
(Information System)
ENEA, A.M.Gadomski,97
3131
dataacquisition
. PASSIVE DSS.
Why Intelligent DSS ?
IDSSs are expecially important when:
the amount of information necessary for the management is so large, or
its time density is so high, that the probability of human errors during
emergency decision-making is not negligible
the coping with unexpected by managers (and DSS designer)
situations requires from the managers the remembering, mental
elaboration and immediate application of complex professional knowledge,
which if not properly used, causes fault decisions.
3232
INTELLIGENT DECISION-SUPPORT SYSTEM - OVERVIEW
link to computer networks
Interventions, decisions
Continuousmonitoring
Emergency Domain
dialogue,suggestions,explanations
IDSS (Artificial Intelligent Agent )
data flowon requestt
Decision support is based on:Informationcurrent data on Em.Domainand Em. OrganizationKnowledge:rules, instructions, procedures, plansPreferencesrisk criteria, role criteria,resource criteria
Human Organizations
Emergency Management Staff
Information system
Continuousmonitoring
amg,94
( Human Agents )
MIND
Adam M. Gadomski, 1995
ENEA
3333
IDSS - Domains of interventionsIDSS - Domains of interventions ENEA, A.M.Gadomski,97
Intelligent Decision Support System
Suggested intervention
Suggested executors
Suggested
experts
Suggested request of information
Suggested cooperation
p. 34Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
How to do?
The IDSS should based on:• application of a generic ideal model of decision
maker (his role)
• its decomposability into human and computer
decision-makers
Ideal Managermodeling
IDSS Human Manager
decomposition
Interface
3535
Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
p.36
Why Intelligent Agent Technology (IAT)?IAT offers various reasoning tools to support classical passive menu driven DSSs to be “intelligent”. Specific advantage is the autonomy of intelligent agents in task execution. Intelligent agent has capabilities to: information filtering and interpretation according to the manager role and situation model. It may suggest new goals, alternative decisions or elaborate plans of the intervention. Intelligent agent can use various Artificial Intelligent methods which enable to copy with uncertain and incomplete data, qualitative reasoning, constrains satisfactions an so on. Its flexibility, modularity and reusing depend strongly on the type of architecture accepted. An “organization” of task-dependent intelligent agents can be considered as the kernel of IDSS.Now, a multiagent architecture based on a repetitive structure, the possibility of (user friendly) modifications of the specific emergency domain and user roles, are considered as a key research fields in the IDSS development.
ENEA, A.M.Gadomski,97
IDSS - Frame System IDSS - Frame System
Emergency Domain
Real-timeData Bases
Real-timeImage Bases
Passive DS (decision support)
Simulators ofmain events
Plumedispersion
Firepropagation
Explosionconsequences
MMI
Passive DS
ComputerNetworkInterface
USER
Intelligent Kernel
EvaluatorAgent
DiagnosticAgent
CommunicationAgent
Action ChoiceAgent
CommonKnowledgetools
amg
ExternalManualSymulatorSupport
EmergencyorganizationData Bases
GIS
DATA BASES SYS
A.MGadomski
ENEA
3737
IDSS developmentIDSS developmentA.MGadomski
ENEA
Symptoms Actions
Toxic Substancesand Risk Industries
DataBases
ConsequencesAnalysis
Algorithms
InterventionProcedures
Diagnosticmodule
Decision-makingmodule (agent)
Whathappens
What will happenor could bappen
What to do
GEOGRAPHICAL DATABASES
EVENT
Predictivemodule
CONSEQUENCES
3838
p. 39Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSSIDSS
First type architecture
Second type architecture
a structural intelligence of multi-agent system
or behavioral intelligence of multi-functionalsystem
Here we can have
?
4040
p41. Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
Abstract Simple Agent : ActionData acquisition
NewInformation
Decision
Goal
DS
PS KS
DS
PS
KS
Domain System:a representation of Physical Domain ofActivity
Agent PreferenceSystem
Agent KnowledgeSystem
Physical Domainof Activity
State- Information
IDSS - STRUCTURAL INTELLIGENCE
p.42 Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
Real Domain
Domain System
PreferencesSystem
KnowledgeSystem
DS
PS KS
DS
PS KS
DS
PS KS
DS
PS KS
DS
PS KS
Second meta level
First meta level
DS
PS KS
inf
inf
goal
inf
Abstract Simple Agent
A Multi-level Abstract Intelligent Agent Architecture
act.
IDSSIDSS ENEA, 97
p.43Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
Domain-Representation Module
Suggested Interventionsamg
Cause of
emergency-eventModification of DomainModel
Preferences System
Possible
consequences
Assessment max.negative consequences
Generation ofIntervention-goal
Knowledge System
Action planning
Decision-Making
Availableprocedures
information
goal
Decision-Making Module based on Abstract-Intelligent-Agent Architecture
Final decision
Domain-Representation Module
Suggested Interventionsamg
Explosion in
chemical plant Plant object in the state of losses generation
Preferences System
Possibleconsequences:
Assessment of large scale human losses
Choice of Intervention-goal: EVACUATION
Knowledge System
Evacuation planspreparation
plans selectionaccording to strategies criteria
Availableprocedures
information
goal
Decision-Making Module: An Example
-toxic plumegeneration- local damage- impact area
of Evacuation
Final decision
p44 Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSS - Domains of interventionsIDSS - Domains of interventions ENEA, A.M.Gadomski,97
Intelligent Decision Support System
Suggested intervention
Suggested executors
Suggested
experts
Suggested request of information
Suggested cooperation
p. 45Some References: [“TOGA Theory...”, A.M.Gadomski, ,AIA Proc.,!993,;[Gad. at al., TIEMEC95], [DiCostanzoat al.,TIEMEC97] ENEA,A.M.Gadomski
IDSSIDSS ENEA, 97
NEW ONTOLOGY
( DSS PROBLEMS ARE RECONCEPTUALIZED)
STRONG INTERDYSCIPLINARY APPROACH
NEW TECHNOLOGIES
(REASONING TOOLS and INTELLIGENT AGENTS ARCHITECTURE)
NEW POSSIBILITES OF UNCERTEN, COMPLEX AND HIGH RISK DOMAIN MANAGEMENT.
CONCLUSIONSCONCLUSIONS
4646
IDSS REFERENCES
Some references and other meta-information you can find on my Home-Pages:
wwwerg.casaccia.enea.it/ing/tispi/gadomski/gadomski.html
4747