- optimal decision for an imaginary simplified world (operations research methods)
- decisions that are "good enough" ===>
"satisfycing" solutions (heuristic search)
- can cope with non-quantifiable variables
- is applicable to non-numerical as well as numerical
information
traveling salesman problem
warehouse location problem
location of central power stations
the set of available alternatives is "given" in a certain abstract sense ---> i.e. we can define a generator guaranteed to generate all of them eventually
they are not "given" in the sense that it is practically relevant
within practical computational limits we cannot generate all the admissible alternatives
we cannot recognize the best alternative , even if we are fortunate enough to generate it early
applied to middle levels of management
example: number of keystrokes needed to do a task
problems with extending operations research methods
to ill-defined problems: uncertainty, computational complexity,
lack of operationality
applied to top management decision, involving judgment
applicable to non-quantifiable problems
example: relationship between cognitive effort and physical effort
- generate ---> produce variety (e.g. genetic mutation)
- test ---> to evaluate the newly generated forms (e.g. natural selection)
- example: see genetic programming in chapter 7
- is myopic
- reaches local maxima (instead of global ones)
- moving away from a local maxima implies: going across a valley
- guided
- there is a goal (question: can we design without a final goal in mind?) one can look back over a design and "clean it up"
- one can examine failures and see what went wrong
- faster than evolution (guidance, remembering previous successes and failures)
- "installed base" problem (Qwerty typewriter, English measurement system, FORTRAN/COBOL, .....)
- standards
- knowledge is cumulative
- nature of memory
- learning of categories
- nature of schemas as prototypes
- experiential human knowledge