Changes

== An Example ==

== An Example ==

An engineer starts with 10,000 different sets of plans for a hydroelectric dam.  His dam must complete two basic tasks: it must hold back a lake with minimal flooding of the surrounding area, and it must maximize power generation.  The intrinsic worth of a dam is determined by how well it completes these tasks; a dam which produces 100 kW of electricity while only raising water levels 2 feet is superior to one which produces 10 kW of electricity while raising water levels 5 feet.  In our simplified example, let us assume that the design of a dam is based on just a few numerical parameters: height, width, thickness, concrete mixture, hydroelectric turbine size, curvature, etc.  After evaluating all the dams, the engineer picks the 100 best and discards the rest.  He then copies each of these dams 100 times.  When copying, the engineer randomly introduces minor differences: say he rolls a die each time he copies a design parameter; if he rolls a six, he then changes that parameter by 1%, flipping a coin to determine if he will increase or decrease the parameter.  Once he has finished copying, he will have 10,000 more dams, most of them very similar to the original 100 but with some small differences.  He then evaluates these 10,000 dams again, picks the best 100, and copies them again, introducing random "errors" in the same way.  After completing this process several thousand times, the engineer has 10,000 dams, all of which perform excellently.  He picks the best of them and thus has an excellent design, without ever doing any real "design" himself.

An engineer starts with 10,000 different sets of plans for a hydroelectric dam.  His dam must complete two basic tasks: it must hold back a lake with minimal flooding of the surrounding area, and it must maximize power generation.  The intrinsic worth of a dam is determined by how well it completes these tasks; a dam which produces 100 kW of electricity while only raising water levels 2 feet is superior to one which produces 10 kW of electricity while raising water levels 5 feet.  In our simplified example, let us assume that the design of a dam is based on just a few numerical parameters: height, width, thickness, concrete mixture, hydroelectric turbine size, curvature, etc.  After evaluating all the dams, the engineer picks the 100 best and discards the rest.  He then copies each of these dams 100 times.  When copying, the engineer randomly introduces minor differences: say he rolls a die each time he copies a design parameter; if he rolls a six, he then changes that parameter by 1%, flipping a coin to determine if he will increase or decrease the parameter.  Once he has finished copying, he will have 10,000 more dams, most of them very similar to the original 100 but with some small differences.  He then evaluates these 10,000 dams again, picks the best 100, and copies them again, introducing random "errors" in the same way.  After completing this process several thousand times, the engineer has 10,000 dams, all of which perform excellently.  He picks the best of them and thus has an excellent design, without ever doing any real "design" himself.  Note that this engineer need not be conscious of the overall goal--he is just applying a simple algorithm over and over.  Indeed, the engineer need not be conscious of anything at all--his task could be performed automatically and without sentience or intelligence.

== Local and Non-Local Improvement ==  

== Local and Non-Local Improvement ==