06-04-09 The Interesting In Between: Why Complexity Exists Scott E Page University of Michigan Santa...

Preview:

Citation preview

06-04-09

The Interesting In Between: Why Complexity Exists

Scott E Page

University of Michigan

Santa Fe Institute

scottepage@gmail.com

06-04-09

Background Reading

06-04-09

Outline

Attributes

Properties

The Interesting In Between

Why Complexity

Conclusions

06-04-09

Complex Adaptive Systems: Attributes

06-04-09

Complex Adaptive Systems

Networks

Source: MIT

06-04-09

Complex Adaptive Systems

NetworksAdaptation

Source: Exploring Nature

06-04-09

Complex Adaptive Systems

NetworksAdaptationInteractions

Source: Uptodate.com

06-04-09

Complex Adaptive Systems

NetworksAdaptationInteractionsDiversity

Source: Scientific American

06-04-09

Complex Adaptive Systems: Properties

06-04-09

Complex ≠ Complicated

06-04-09

Complex ≠ Equilibrium (w/ Shocks)

06-04-09

Complex ≠ Chaos

Source: Andrew Russell

06-04-09

Complex ≠ Difficult

Source: Biology-direct

06-04-09

Complex = Dancing Landscapes

Source: Chris Lucas

06-04-09

Epi-Phenomena

Emergence Structures and Levels

Source: Boortz.com

06-04-09

Diffusion Limited Aggregation

Start with a seed on a plane.

Create drunken walkers who start from a random location and walk in random directions until touching the seed, at which point the walkers become immobilized.

Witten and Sander (1981)

06-04-09

Diffusion Limited Aggregation

seed

walker

06-04-09

Diffusion Limited Aggregation

06-04-09

Conway’s Game of Life

X 5

76

41 2 3

8

Cell has eight neighbors

Cell can be alive

Cell can be dead

Dead cell with 3 neighbors comes to life

Live cell with 2,3 stays alive

06-04-09

Examples

X

06-04-09

Bigger Space

06-04-09

A New Kind of Science - Wolfram

Binary state objects arranged in a line using simple rules can create- “perfect’’ randomness- chaos- patterns- computation

06-04-09

Epi-Phenomena

Emergence Structures and LevelsEmergent Functionalities

Source: Biology-direct

06-04-09

Wolfram’s 256 Automata

N X N X

06-04-09

Rule 90

N X N X

2

8

16

64

Sum = 90

06-04-09

Rule 90

N X N X

2

8

16

64

Sum = 90

06-04-09

Four Classes of Behavior

06-04-09

Emergent Computation

Source: U of Indiana

06-04-09

Epi-Phenomena

Emergence Structures and LevelsEmergent FunctionalitiesInnovation

http://media-2.web.britannica.com/eb-media/54/4054-004-F5EB3891.jpg

06-04-09

Epi-Phenomena

Emergence Structures and LevelsEmergent FunctionalitiesInnovation Large Events

06-04-09

A Long Tailed Distribution

06-04-09

A Long Tailed Distribution

cities size words citations web hits book sales phone calls earthquakes moon craters wars net worth family names

06-04-09

Large Events

Source: www2002

06-04-09

Per Bak’s Sandpile

sand

table

floor

06-04-09

Per Bak’s Sandpile

sand

table

floor

06-04-09

Self Organized Criticality

Systems may self organize into critical states. If so “events” may not be normally distributed. They may instead have long tails. Small events could have enormous consequences.

06-04-09

Epi-Phenomena

Emergence Structures and LevelsEmergent FunctionalitiesInnovationLarge Events Robustness

Source: NBC

06-04-09

“Imagine how difficult physics would be in electrons could think.”

-Murray Gell-Mann

06-04-09

Robustness: The World of Thinking (or adapting)

Electrons

06-04-09

The Langton Graph

06-04-09

The Langton Graph

06-04-09

A Thought Play

A Simple Model of Forest Fires & Bank Failures

06-04-09

The Bank Model

Banks choose to make a risky loan each period with probability p

06-04-09

The Bank Model

Banks choose to make a risky loan each period with probability p

Risky loans fail with probability q but have a higher yield

06-04-09

The Bank Model

Banks choose to make a risky loan each period with probability p

Risky loans fail with probability q but have a higher yield

Failures spread to neighboring banks only if those banks have a risky loan outstanding

06-04-09

The Forest Fire Model

Trees choose to make a grow each period with probability p

Trees get hit by lightening with probability q

Fire spreads to neighboring locations only if those locations have a tree

06-04-09

Example

Period 1: 00R00R000RR0RPeriod 2: R0R00R00RRRRR

06-04-09

Example

Period 1: 00R00R000RR0RPeriod 2: R0R00R00RRRRR Period 3: R0R00R00FRRRRPeriod 4: R0R00R00FFFFFFPeriod 5: R0R00R00000000

06-04-09

Key Insight Revisited

Bank managers should be smarter than trees!

06-04-09

Forest Fire Model Results

Yield increases in p up to a point and then falls off rather dramatically

Physicists call this a ``phase transition’’

06-04-09

Poised at the ‘edge of chaos’

rate of risky loans p*

yield

06-04-09

Smarter Banks

Let each bank learn (using a standard learning rule from psychology called Hebbian learning) whether or not to make risk loans.

06-04-09

Emergent Robustness

rate of risky loans p*

yield

06-04-09

Emergence of Firewalls

11101101110111100111

06-04-09

The Interesting In Between

06-04-09

The Barn

Mutation/Adaptation Network

Big Area Real Novelty

InteractionDiversity

06-04-09

Tuning Complexity

06-04-09

06-04-09

Rates of Adaptation/Learning

0: - rule aggregation

11: - rational expectations

06-04-09

Interdependencies

0: - decision theory

11: - mangle

06-04-09

Network/Connectedness

Mathematically tractable models: N = 2

-game theory N = Infinity

-averaging- random mixing

06-04-09

Rock, Paper, Scissors

Rock: All D Toxic E ColiPaper: TFT Resistant E ColiScissors: All C Sensitive E Coli

06-04-09

Rock, Paper, Scissors

Rock: All D Toxic E ColiPaper: TFT Resistant E ColiScissors: All C Sensitive E Coli

Simulations: we get “ stone soup” but diversity on a lattice or a line

06-04-09

Rock, Paper, Scissors

Rock: All D Toxic E ColiPaper: TFT Resistant E ColiScissors: All C Sensitive E Coli

Real Experiments: we get one type E Coli ina flask but diversity on a slide

Kerr, Riley, Feldman, Bohannan (Nature 2002)

06-04-09

alone lattice network soup

complexity

06-04-09

Diversity

0: - representative agent model

11: - statistical averaging

06-04-09

A complex adaptive system requiresthe right amount of “interplay” between our agents.

06-04-09

The Small Barn

Mutation Network

Equilibrium

Interaction Diversity

06-04-09

The Big Barn

Mutation Network

Lack of Structure: Limiting Distributions

Interaction Diversity

06-04-09

The Interesting in Between

Mutation Network

Complex

InteractionDiversity

06-04-09

Why Complexity?

06-04-09

Emergent Complexity

A lurking theory of homeocomplexus: learning rates, interaction effects, networks, and diversity adjust to maintain complexity.

06-04-09

Enlarging the Barn

If the barn is small, the system is often both stable predictable. These two properties create an opportunity for faster adaptation – this can mean a new action (diversity), a new connection, or a new interdependency.

06-04-09

Shrinking the Barn

If the barn is big, the system tends to be either a mangle or random. Either state creates an incentive for simple strategies, fewer connections, less diversity, and reducing interdependencies.

06-04-09

Dali’s Barn

Mutation Network

Big Area Real Novelty

Interaction

Diversity

06-04-09

Conclusions

06-04-09

Big Successes

Crowds and PanicsSpatial SegregationResidential LocationTransportation NetworksInternet StructureScaling Laws

06-04-09

We Have No Choice

Global WarmingGlobal Financial MarketsDisease TransmissionTransportationInternet Terrorism NetworksCrime

06-04-09

Normal Science

Step 1: Construct a ModelStep 2: Produce HypothesesStep 3: Test the Hypotheses

06-04-09

All New

Non Analytic

Ensembles of Models

Interactions of disciplines in same model

Recommended