Microsoft Word Kurzweil, Ray The Singularity Is Near doc

THE EVOLUTIONARY ALGORITHM
Create N solution "creatures." Each one has: 
•
A genetic code: a sequence of numbers that characterize a possible solution to the 
problem. The numbers can represent critical parameters, steps to a solution, rules, etc. 
For each generation of evolution, do the following: 
•
Do the following for each of the N solution creatures: 
(i)
Apply this solution creature's solution (as represented by its genetic code) to the 
problem, or simulated environment. 
(ii)
Rate the solution. 
•
Pick the L solution creatures with the highest ratings to survive into the next generation. 
•
Eliminate the (N – L) nonsurviving solution creatures. 
•
Create (N – L) new solution creatures from the L surviving solution creatures by: 
(i)
Making copies of the L surviving creatures. Introduce small random variations into 
each copy; or 
(ii)
Creating additional solution creatures by combining parts of the genetic code (using 
"sexual" reproduction, or otherwise combining portions of the chromosomes) from 
the L surviving creatures; or 
(iii)
Doing a combination of (i) and (ii), 
•
Determine whether or not to continue evolving: 
Improvement = (highest rating in this generation) – (highest rating in the previous 
generation). 
If Improvement < Improvement Threshold, then we're done. 
•
The solution creature with the highest rating from the last generation of evolution has the 
best solution. Apply the solution defined by its genetic code to the problem. 
Key Design Decisions 
In the simple schema above, the designer needs to determine at the outset: 
•
Key parameters: 
N 
L 
•
Improvement threshold 
•
What the numbers in the genetic code represent and how the solution is computed from 
the genetic code. 
•
A method for determining the N solution creatures in the first generation. In general, 
these need only be "reasonable" attempts at a solution. If these first-generation solutions 
are too far afield, the evolutionary algorithm may have difficulty converging on a good 
solution. It is often worthwhile to create the initial solution creatures in such a way that 

---

they are reasonably diverse. This will help prevent the evolutionary process from just 
finding a "locally" optimal solution. 
•
How the solutions are rated. 
•
How the surviving solution creatures reproduce.