Guided Self!Organization:" from chaos to planning?!
Francis Heylighen!
Evolution, Complexity and Cognition Group!
Vrije Universiteit Brussel! Newtonian Science!
Making simple models, by means of:#
•!Analysis! •!Reductionism! •!Determinism! •!Reversibility! Complexity!
complexus = entwined# •!Many parts! •!that are connected so that they are di"cult to separate! In between Order and Disorder# •!the #edge of chaos$! From Newton to Complexity!
Analysis → Connection#
Reductionism → Emergence#
•! The whole is more than the sum of its parts!
Deterministic change → Uncertainty#
•! We cannot predict exactly what will happen!
Matter → Organization#
•! Systems more important than elements or particles!
Reversible movement → Evolution#
•! Creativity and progress!
Emergence!
whole > sum of the parts#
Whole has emergent properties # •! properties not reducible to properties of the parts! examples# •!car: maximum speed = emergent, weight = reducible! •!music: melody = emergent, volume= reducible! Open Systems!
Inseparable from environment:# •!input = incoming %ows of matter, energy, information! •!output = outgoing %ows!
SYSTEEM
input throughput output
grens
OMGEVING Coupling!
di$erent systems can be coupled via their input and output# three basic types#
sequentiëel parallel circulair
Networks!
Many coupled systems form a network# Which can be seen as a system itself# Subsystems and Supersystems!
subsystemen
supersysteem
Hierarchy!
Linearity!
Linearity = e$ect proportional to cause# •!using k times as much raw material → & produces k times as much output! •! f(k.a) = k.f(a) Therefore: Small causes ! small effects Large causes ! large effects Non'linearity!
E$ect can be much larger or much smaller than cause# Usually caused by e$ect feeding back into cause#
# positive feedback negative feedback!
amplifying deviations suppressing deviations !
→ unstable → stable # Chaos!
sensitive dependence on initial conditions# •!Unobservably small perturbations can produce huge e(ects! the butter%y e$ect: %apping of wings may create a tornado!
→ intrinsic unpredictability and uncontrollability# Self'organization!
Spontaneous emergence of order or organization# •!not imposed by outside or inside agent! Organization distributed over all the components# •!Collective, decentralized! •!Robust against disturbances! Mechanisms of self'organization!
Global order emerges from local interactions#
•!Components initially interact only with their neighbors!
•!Until they #align$!
•!i.e. )nd a stable, #&t$ arrangement!
•!Alignments tend to propagate to further neighbors!
•!until they spread over the system as a whole! Magnetization!
Magnetization!
B*nard convection!
surface surface
bottom (hot) bottom (hot) B*nard convection!
Dynamics of Self'organization!
Local variation !
Selective elimination of the #un)t$ arrangements!
Growth of the )t arrangements!
•!Positive feedback!
Suppression of deviations from that arrangement!
•!Negative feedback!
Result: Initially rapid growth → stabilization → robustness and continuing adaptation# Attractor Dynamics!
Non!linear systems typically have several attractors#
Attractor=#
•!!stable region in state space!
•!that the system can enter& but not leave!
•!surrounded by basin!
•!System descends & spontaneously from & basin into attractor!
•!and comes back to the & attractor if forced out of it!
Order out of Chaos!
Self!organization is stimulated by random variation 'noise, chaos(#
•! more con)gurations are explored in the same time interval!
•! higher probability of )nding a #deep$ attractor !
•! therefore, a stable outcome is reached more quickly and reliably! Example: shaking!
Throw cubes randomly in a box#
•!preference for position close to bottom!
•!But cubes mutually obstruct → #friction$!
Shaking reduces volume#
•!Random variation explores many possible con)gurations!
•!But preferentially retains the most stable ones at the bottom! Complex Adaptive Systems!
Collection of interacting agents#
Characterized by self"organization#
•! global coordination and adaptation!
•! emerging from local interactions!
Examples#
•! Societies, groups, communities!
•! Cities, regions!
•! Ecosystems, markets!
•! World'Wide Web ! Agent!
Goal!directed individual#
•! tries to maximize #)tness$ !
Follows condition!action rules #
•! if condition A sensed, then perform action B!
•! A → B!
•! where B in general increases )tness! Agents are diverse!
Di$erent agents have in general di$erent abilities:#
•! Knowledge of rules!
•! Information about conditions!
•! Capacity to perform actions! Bounded rationality!
Agents have only # •! local information! •! Limited information processing ability!
The environment is complex and unpredictable#
•! non'linearity may give rise to chaos!
Therefore, agents cannot predict the long!term e$ects of their actions# •! they must rely to some degree on trial'and'error!
•! They must constantly adapt to the circumstances! Evolution of Synergy!
Agents co!evolve#
•!Variation and selection lead to mutual adaptation!
•!Each acts so as )t in with its neighbors, and vice versa!
Selected interaction characterized by#
•!Minimal friction +con%ict, negative sum, !
•!Maximal synergy +#win'win$, positive sum,!
Leads to stable )connection* or )alignment* #
•!No more con%ict, but cooperation! Propagation of Alignment!
More agents join alignment#
•!Often ampli)ed by positive feedback!
•!Larger aligned assembly exerts larger in%uence on non' aligned agents!
•!Pressure to #join$ increases!
Eventually, the whole system becomes aligned#
All agents act synergetically# Emergence of & collective intelligenc"e
Distributed cognition#
•! di(erent agents contribute di(erent results at di(erent times and places!
•! results are integrated!
•! together, they solve the global problem:!
•!co!ective inte!igence! Coordination mechanisms!
Alignment of targets# minimizing con%ict or friction!
Division of labor#
Di(erentiation or specialization!
Work%ow#
Actions performed in right sequence!
Aggregation of results#
•!Into collective solution! Coordination!
workflow
division of labor aggregation Self'organization of coordination!
Stigmergy# •!Result of action stimulates performance of subsequent action! •!Right agent incited to intervene at right moment! •!Aggregation into shared workspace! Reinforcement learning# •!Successful sequences of actions are reinforced!
•!Unsuccessful ones are eliminated! Planning?!
Planning is intrinsically di+cult for complex systems# •!Uncertainty, chaos, autonomy, openness….!
Better: guided self!organization#
•!Formulate desired outcomes!
•!Let system itself evolve towards these targets!
•!While adapting to any unforeseen complications!
Don,t try to rigidly control the process!#
Anticipating outcomes!
Self!organizing systems evolve towards attractors#
Attractors are complex and emergent#
But can be foreseen to some degree#
•!By using multi'agent simulations!
•!Or by mapping out positive and negative feedback loops!
•!Systems dynamics models!
This allows us to target desired attractors#
•!and avoid undesired ones! Controlling Emergent Outcomes!
How can we make sure to get the desired result?# Initial situation# # problem to be dealt with by the process! Desired outcome# #situation that solves the problem! We need to stimulate moves in the right direction! But not impose a trajectory! Controlling direction!
De&ne &tness criterion: # •!what is better or worse?! Reinforcement# • reward moves toward )tter situation! •!e.g. Subsidizing education or sustainable development! • inhibit moves that reduce )tness! •!e.g. Taxing pollution or waste ! Controlling boundary conditions!
Set up architectural )sca$old*# Constraining structure# Make #right$ moves easy! Make #wrong$ moves di"cult! But leave enough space for exploration!! Examples# #Organs$ grown from cells! seeded on arti)cial frame! Stigmergy!
Actions leaves a trace in a shared workspace#
The trace stimulates agents to perform additional actions!
Very robust way to collaborate#
•!No need for planning or explicit coordination#
•!Division of labor and work%ow emerge spontaneously#
•!agents do not need to communicate!
•!workspace stores and aggregates contributions!
•!Successful contributions grow through positive feedback! Stigmergy: Wikipedia!
Person A writes text on topic X#
•!Action leaves trace on website!
Person B independently reads text#
Person B is stimulated to add to the text#
Further contributions are aggregated on the same webpage!
Positive feedback: #
•!more edits → better text → more readers → more edits → ...! Stigmergy: termite hill construction! individual termite drops mud randomly! new termite attracted to add mud to already present mud! positive feedback: mud → more mud! the mud heap grows into a column! columns grow towards each other!
→ cathedral'like structure with arches# Stigmergy: Ant trails!
Ants coming back from food source leave pheromone trace#
Ants searching for food preferentially follow pheromone trace#
•!deviate from trace occasionally!
•!Allows them to )nd new food source!
Strong trails get reinforced as more ants use them#
Trails to exhausted food sources evaporate#
Outcome: adaptive network of trails connecting food sources and nest in most e+cient way# Adaptation to disturbances! Sca(olding Stigmergy!
A stigmergic medium can be pre!structured, e.g. :#
•!Encyclopedia with templates +#stubs$, for missing articles!
•!Plastic sca(old to grow an ear!
•!City with some roads and landmarks, but no buildings yet!
This can incite and facilitate desired self!organization#
•!Attractive empty slots invite developments!
•!Positive feedback: development → more development!
•!Remains adaptive to unforeseen situations! Conclusion!
Complexity precludes accurate prediction and control#
•! → planning intrinsically unreliable!
Self"organization produces robust, adaptive systems#
•!Even in very complex, unpredictable environments!
•!But outcomes may not be as desired!
We need to investigate guided self"organization#
•!Using reinforcement signals and sca(olds !
•!to steer stigmergic coordination in the desired direction!