Adaptation on Rugged Landscapes



Adaptation on Rugged Landscapes

0 0


Levinthal-1997-summary


On Github usuallycwdillon / Levinthal-1997-summary

Adaptation on Rugged Landscapes

A working summary of D. Levinthal's 1997 paper

CW Dillon

for Computational Organizational Theory, CSS739

George Mason University

I am a PhD student in the Department of Computational Social Sciences, at the Krasnow Institute for Advanced Studies. We focus primarily on neurosciences and applications of complexity theory from a behavioral perspective. This seminar, by Dr. Maksim Tsvetovat is presented as a special topic in social network analysis.

Overview

  • A model of organizational adaptation
    • I assumed that this is an ABM because of the description
  • Pretty rigid (lots of assumptions)
  • L's examples
  • Interesting because...

What the Model is About

  • Organizations assumed to adapt to fit their environment (or die)
  • Survival of the fittest, most adaptable, and luckiest
  • Includes a complex adaptive version

Organizations

  • Orgs have N attributes
    • Sample model has 10 attributes
    • Initial values are either 0 or 1
  • Each attribute assumed to be influenced by K other attributes
    • K=0 means each attribute is independent of all others (loosely-coupled)
    • K=N-1 means each attribute is dependent on all others (tightly-coupled)
    • Oops! I had that wrong before. Updated 1015a, Tuesday
  • Over time, orgs increase their internal connectivity (level of epistatic intreaction) which reduces their adaptibilty

Environment

  • Environment is a space with 1 or more peaks
  • Number of peaks can reach $2K$, where K=N-1
  • Changes in response to number of ``fit'' firms, in CAS version

Adaptations

  • In each turn, orgs can alter one attribute to better fit the landscape
  • No cost of adaptations
  • Assumed that all adaptations are for the better
  • Some firms not able to adapt:
    • where K approaches N-1
    • when level of epistatic interactions is high
  • Death of an org allows a new org to be born (population is constant)
  • Can also change every attribute to random value: ``long jump''

Experiments

  • When K is 0:
    • only 1 optimal fit to environment
    • environement is smooth
  • As K increases:
    • more combinations and permutations of attributes fit the env
    • environment is rugged

Results

  • Diversity of orgs can be explained by diversity of the environment
  • Tightly-coupled orgs are less likely to survive by adaptation:
    • change in one attribute impacts many others
    • better to make ``long jumps'' in dynamic landscapes
  • Loosely-coupled orgs survive by:
    • adapting quickly (change is easy)
    • making few ``long jumps''

It's Interesting Because...

  • Consider two org types: government offices, restaurants
    • How many attributes determine success? N
    • How tightly-coupled are they? K
    • How radically does their landscape change?
  • Survival or death (and replacement)
    • Survival rate of encumbants decreases as K increases
    • Diversity of fit forms decreases if environment is adaptive