a unified framework for collective systems
DESCRIPTION
Emma Hart: Edinburgh Napier University Jeremy Pitt: Imperial College London Ulle Endriss: University of Amsterdam Presentation from ECAL 2013TRANSCRIPT
A UNIFIED FRAMEWORK FOR COLLECTIVE SYSTEMS
Emma Hart, Edinburgh Napier UniversityJeremy Pitt, Imperial College LondonUlle Endriss, University of Amsterdam
G d Vi iGrand Vision
Applications
p
A Software Toolkit of Design Patterns and
Components
y
A Unified Theory of Operations for CAST
Systems
Wh d d h ?Why do we need a new theory ?• Existing engineering
approaches provide some theoretical basis• E.g. control theory –
ensure/prove stability• But most methods don’t
t f d fi iaccount for defining properties of CAST systems• Lead to systems that are
oscillatory or at worst unfit foroscillatory or at worst unfit for purpose
• Existing methods often domain-driven (e gdomain-driven (e.g. telecoms, robotics)• Not generalisable or
transferabletransferable
CAS i M l i Di i liCAS is Multi-Disciplinary
• Many theories from individual disciplines
• Hard to compare theoriestheories
• Theories address different aspects ofdifferent aspects of CASDon’t account for• Don t account for engineering constraintsconstraints
T d ifi d hTowards a unified theory
• Unifies concepts from multiple disciplines into a single framework
• Qualitative theoryBiological Systems
Qualitative theory represented in axiomatic form Computational Organisationala o at c o
• Can be formalised and analysed
Social Choice Theory
analysed• Operationalised via design patternsdesign patterns
Bi l i l SBiological Systems• Immune-neuro-endocrine mechanisms lead to homeostasis
• Long-term stability• Adapt over multiple ti llead to homeostasis
• Cohen’s cognitive immune system :
timescales• Coordinate multiple heterogeneousimmune system :
• Decision making via co-respondence
heterogeneous components
• Deal with limited and• Swarm insects
• Coordination, partial info
• Deal with limited and partial information
• Decision making• Symbiosis between multiple species:
C ti
Decision making• Conflict resolution
• Cooperation
S i l Ch i ThSocial Choice Theory• Originates in economics
and political science• Concerns design &Concerns design &
analysis of methods for aggregating preferences of multiple agents into
• Heterogeneous agents• Multiple objectivesmultiple agents into
collective decisions• Social choice considers
formal aspects of
• Collective decisions• Open-ness
Fair division of resourcesformal aspects of democratic decision making (e.g electoral
t )
• Fair division of resources• Stability
systems)• Computational Social
Choice add an algorithmic gperspective
O i i l ThOrganisational Theory
• Elucidates principles for stable resource • Collective Actionmanagement
• Study of engineered
• Collective Action• TrustStudy of engineered
systems• Insights into
• Cooperation• Stable and enduring • Insights into
engineering socio-technical
gsystems
technical ‘organisations’ in a top down mannerdown manner
A U ifi d Th f O iA Unified Theory of Operation
A Unified Theory of Operations for CAST Systems
Computational n esns es…
Social Choice
Biologicaln-
ness
form
atio
n
Obj
ectiv
e
nflic
ts
tera
ctio
n
prop
ertie
Systems
Organisational
Ope
n
Noi
sy In
f
Div
erse
O
Con
Soc
ial I
n
w C
AS
T
Theory DS
New
Engineering Requirements of CAST Systems
Wh d h i i ?What does synthesis give ?
Biological Organisational Biological Computational gSystems
OrganisationalTheory
Biological Systems
pSocial Choice
>>Computational Social Choice
EngineeringConstraints
OrganisationalTheoryEngineering
>>Constraints TheoryEngineering
Constraints
• Addresses weaknesses in individual theories• Addresses conflicts• Addresses conflicts• Respects engineering constraints
I di id l W kIndividual Weaknesses• Biological Systems:
• Tend to rely on homogeneous collectivesGlobal rather than individual objectives• Global rather than individual objectives
• Considerable physical differences
• Computational Social ChoiceComputational Social Choice• Based on standard models from economics• Abstracted from human decision making (different goals but same
model)
• Institutional Theories• Easy to get locked into sub-optimal states due to path
dependencies• Not clear how to evaluate ‘fitness’ of an institutionNot clear how to evaluate fitness of an institution
C l iConclusions• Unification addresses current fragmented approach to
inter-disciplinary researchDiff t l i t l tl hi d l id ti• Different analysis tools currently hinder elucidating connections between fields
• Many existing theories don’t account for engineering• Many existing theories don t account for engineering constraints of CAS
• A unified theory will:• A unified theory will:• Enable formal comparison between concepts from different
disciplines• Drive innovation in field