Engineering systems with emergent behaviour

1
Engineering systems with emergent behaviour

• Graeme Smith
• The University of Queensland Australia

2
Draft Research Agenda

• Specification and Design
• Abstractions and models for massively parallel,
  open-ended systems. We currently lack the
  necessary abstractions and foundations to
  describe, model, and design massively parallel
  systems

3
Draft Research Agenda

• Specification and Design
• Deducing a global specification from local
  rules, and finding local rules that produce a
  desired global behaviour. we will have to
  integrate services as parts into a larger
  environment. We will thus no longer be able to
  use a top-down approach
• addresses deployment
• BUT not system development

4
Is top-down development possible?

• According to the literature, top-down
• development of emergent behaviour is either
• 1. Impossible
• 2. Impracticable
• 3. Difficult
• But which?

5
Impossible

• By definition
• By analogy with emergence in natural systems
• By Gödel's incompleteness theorem
• Because it is an undecidable problem (?)
• Because it requires high and low level languages

6
Impracticable

• As an accepted fact
• By analogy with emergence in natural systems
• However
• Smale and Cucker (2005)
• behaviour of individual birds ? flocking
• Zhu (2005)
• autonomous sorting

7
Topics of interest

• Categorisation and formalisation of emergent
  properties (macro level)
• statistical distribution, convergence,
  stability,
• Modelling of intelligent, adaptive components
  (micro level)
• goals, model of environment,
• Coordination of components (meso level)
• interaction patterns, mediating infrastructure
  (inspiration from biology, physics, society)

8
Strategies for the meso level

• Scenarios (cf. Zhu, engineering emergence)
• Abstract mediators (cf. Hayes, deadline command)
• Time scales (cf. Burns, time bands)
• Islands of interaction (cf. Hogg, aliasing
  control)
• Key is incremental development.

9
Example shape-forming atoms

• L1 all atoms are in a position of desired shape
• L2 a subset of the atoms that are not in
  position, move into a position of desired shape
  (all atoms already in a position do not move)
• L3 an atom not in the desired shape moves next
  to one in the desired shape that needs a
  neighbour (initially at least one atom is in the
  desired shape)

10
Example continued

• L4 an atom in the desired shape broadcasts its
  need for a neighbour (atoms in position store a
  model of the desired shape)
• an atom not in the desired shape responds to
  broadcast by moving to broadcaster
• L5 a message is broadcasted by moving between
  neighbouring atoms which increment a number it
  carries. This number is stored by the receiving
  atom and hence creates a gradient which can be
  followed to the broadcaster

11
Revised Research Agenda

• Specification and Design
• Deducing a global specification from local
  rules. Influencing the global behaviour by making
  local changes. we will have to integrate
  services as parts into a larger environment. We
  will thus no longer be able to use a top-down
  approach

12
Revised Research Agenda

• Specification and Design
• Finding local rules that produce a desired
  global behaviour. To guarantee the presence of
  desired global behaviour, and the absence of
  undesired global behaviour, new top-down
  development approaches are needed. Such
  approaches must deal with the massive scale and
  complex interactions of ensembles.