More Than Fit For The Future

1
More Than Fit For The Future

• An EvoDebate Position Statement
• Robert E. Smith
• Director
• The Intelligent Computer Systems Centre
• The University of The West of England, Bristol
• www.csm.uwe.ac.uk/icsc

2
Disheartening News

• Empirical Demonstrations that show the region
  where EC is best amongst search algorithms is
  very narrow
• Criticisms of Hollands Theories
• The No-Free-Lunch Theorem
• Leaving Us Asking
• What is the niche for EC? Is it really that
  useful for anything? Have we all been deluded?

3
My Argument

• Craft and Technology almost always lead
  scientific understanding
• Because of the hard edges of computing, computer
  scientists arent used to this fact
• Complex systems based paradigms (like EC) best
  lend themselves to craft

4
An Example

• Consider the longitude problem
• How does one determine ones longitude on Earth?
• This was once the most important technical (and
  commercial) problem in the world
• Clocks (to determine ones local time relative to
  home time) are provably inadequate for this task,
  since all clocks are always inaccurate

5
Clearly

• The only solution for this task is an
  astronomical one, since time of day (and thus
  longitude) is an astronomical problem
• It was once accepted scientific fact that the
  only adequate solution for the longitude problem
  was to look up, at the perfect clockwork of the
  stars

6
The Real Solution
7
Harrison identified the problem

• As technological, not scientific or mathematical.
  He asked
• Could one make the inherently flawed clock good
  enough to solve the problem
• The fact that one could do so was a shock to the
  scientific world, but not to the world of
  craftsmen

8
Craft works beyond scientific understanding
We in computing dont usually face this fact
9
Another example

• Simon Newcomb (another astronomer) offered strong
  scientific arguments that heavier-than-air flight
  was impossible
• His chief arguments concerned scale up
• Weight scales as the cube, lift scales as the
  square, therefore, there is a limit to the size
  of heavier-than-air vehicles.
• Newcomb argued that large birds were clearly near
  this limit

10
And, of course, he was right
3/2 slope
Heavier-than-air flight has a very narrow space
of usefulness
11
No Free Lunch? No St.

• Wolpert and Macready, 1995
• When performance is averaged over all possible
  search spaces, all search algorithms perform
  equally well (including enumeration)
• This fact has nearly no practical value, and
  isnt a surprise to technologists and engineers
  at all

12
So, I say

• Delivering pretty good solutions is an
  important goal
• Concepts of crisp problems and crisp optimality
  are illusions that seldom exist in the real world
• Filling the pretty-good-solution niche is the
  appropriate goal for soft computing (particularly
  EC)

13
Suggested EC Concentrations

• co-evolution
• adaptive agents
• online adaptation to changing environments,
• machine innovation and creativity
• opening our minds to industrial problems where
  computer algorithms are yet to be applied, but
  where pretty-good, pretty-fast solutions have
  real cash value.