Transcript Autonomous Agents - Department of Computer Science

Workshop
on
Causal/Influence Networks
July 2009
C.A. Hooker
PhD (physics), PhD (phil.) FAHA
1
Autonomy = I/MR synchrony
2
Autonomy is important
• demarcation of living systems
• Organisation, global constraint (not
order) is fundamental
• grounding for agency
• frames the evolution of intelligence and
intentionality,
3
Comparative system order
•
•
•
•
•
•
Property
Internal bonds
Directive ordering*
Constraints
Organisation
System Kind
GAS
None
Vweak/s
None
None
CRYSTAL
Rigid, passive
Vstrong/s
Local
None
CELL
Adaptive, active
Mod/Vcomplex
Global
Very high
• * Directive ordering is spatio-temporally selective energy flow
4
Autonomous Agents [AAs]
AA interrelations are grounded in autonomy, →
SDAL:
• Self-directed (= feedback-evaluated
behavioural adaptation)
• Anticipative (= feedforward on evaluation)
• Learning (= feedback-evaluated adaptation of
self-directedness)
AAs are finite, → uncertainty, heuristics,
satisficing
5
SDAL
Example: detective
• Synergy between profile development and investigation
method → simultaneously moves itself towards its goal
and improves its capacity to move towards its goal.
• Solves open problems: ill defined = problem, method,
solution criteria [all deep design problems]
• Captures science research cycles
– E.g. ape language research
– Adaptive method, e.g. error treatment
• Captures integrated modelling & management method
6
SDAL and scientific niche creation
• Key to scientific progress is its capacity for synergistic
•
•
new multiplexed niche creation.
Cf. Lasers as distance/time measuring, imaging, energytransferring devices, and impact on sci. instruments,
methods & models + economic technologies with $
feedback to sci.
Sci. SDAL: sci. uses its new niches, created from specific
problem solutions, to improve its learning capacity.
– e.g. observation
→ context dependence, many weak bonds, idiosyncrasy
(curbs current network enthusiasm)
Contrast military constraints?
7
Evolution of endogenous regulation
• Darwinian model:
‘Transparent phenotype’
Open VSR → regulated VSR
• Autonomous Systems
Model: Organised
Phenotypes
8
The major organisational evolutionary transitions
LIFE’S CONSTRAINTS [SUFFER!]
•
FINITUDE + FALLIBILITY
•
DISSPATIVENESS + DELICACY
LIFE’S SOLUTION [ORGANISE!]
•
AUTONOMY
•
ANTICIPATIVENESS
•
APTNESS
•
ADAPTIVENESS
LIFE’S BELL’S & WHISTLES [ENJOY!]
•
SOCIALITY
•
SELF-DIRECTEDNESS
•
AGENCY
•
INTELLIGENCE
•
CULTURE
9
Enabling constraints for adaptiveness
•
•
•
•
Communal
Social
adaptiveness Insects
dominates
•
•
Multicellular
Body Cultures
•
Chimpanzee
•
Bonobo
•
Human
•
•
•
•
•
•
•
•
•
Social
Birds
Individual
adaptiveness
Dominate
Slime Moulds
A-social
Organisms
0
1
2
Ratio of usable individual parametric plasticity between isolate and communal states.
10
Culture: technology
Technologies are amplifiers
Technology as culture = technology as a-cultural:
• Objects, methods/tools, possibilities, language
common across diverse agents
• Each group and agent exploits idiosyncratic
possibilities context-dependently
Example – computers in markets
• Import (tech) ≈ possibilities, agent range, access
Tech as kth order culture = tech as < kth order a-cultural
• music, fashion as cultural technologies
• language as head-altering tool = technology 11
Culture as dynamical
• Technologies as dynamical entrainments in a rugged
entrainment landscape
• Institutions as self-organised emergents: Hayek to
Lansing to Shi
Modelling
• genetic Darwinism: bioevolution :: memes: cultural
dynamics [Dawkins: Jablonka/Lamb :: Blackmore: ?]
• Rubber sheets & oscillators: shaped/shaping
• Agency, idiosyncrasy & coherence limits (e.g. control
functions, Woese on gene sharing)
Military culture: Centralised →? Technologies?
12
Some proto-cultural dynamical distinctions I
SHMO: simple harmonic oscillator.
DCC: dynamically collective constraint.
Model 1: a set of independent SHMOs.
• System state = aggregate of individual states.
• No DCCs. All collective phenomena are patterns
determined only by initial (or boundary)
conditions.
• Social example: the distribution of objects in
13
refuse.
Some proto-cultural dynamical distinctions II
• Model 2: model 1 + small, local pair-wise interactions
•
•
•
•
between SHMOs.
System state = perturbation of model 1 state by
addition of local pair-wise corrections.
Weak local DCCs responsible for collective wave-like
perturbation propagation.
For increased interaction strength &/or less local
interaction, stronger &/or more global DCCs emerge
generating further collective phenomena, e.g.
entrainment, chaotic behaviour.
Social example: pair-wise reflex interaction behaviour.
14
Some proto-cultural dynamical distinctions III
• Model 3: model 2 + interactions modified by
SHMO integrative memory.
• System state = joint product of SHMO states
and interaction states. Memory is some
function of past interactions and constrains
current interaction form and strength.
• Emergence of global DCCs constraining SHMO
behaviour in relation to collective properties.
• Social example: pre-recording socially
referenced behaviours.
15
Some proto-cultural dynamical distinctions IV
• Model 4: model 3 + integrative memory referenced to a
•
•
•
shared global field.
System state = joint product of SHMO states,
interaction states, and field state. Field interacts
locally with all SHMOs (realised, e.g., by a rubber
sheet to which they are attached or an electromagnetic
field which their movements collectively generate).
Emergence of strong global DCCs constraining SHMO
behaviour in relation to collective properties based on
inherent field dynamics.
Social example: socially recorded referenced
16
behaviours.