12
« on: January 08, 2008, 11:08:15 am »
i think rewarding extraordinary good results in a skill with advancement in capabilities is only one side of the coin. more often you learn from failure as much as from success.
(can you imagine to learn a lot from a lucky shot? i can't.)
usually the (unaided) learning speed in some completely new skill is as follows:
* one usually has some "inherent" probability of success, applying the skill untrained
* when new to a skill you experiment rather than train the skill since you don't even know how to perform the skill perfectly. this is rather slow and you learn mostly from success (which is rather rare).
* you find out how to perform the skill. you learn now more rapidly and you learn as much from success as from failure.
* you perfect the skill. you learn mostly from failure (which is rather rare)
so seemingly a realistic function for learning must depend on both success and failure. the easiest way to implement this is the use of two flags for each skill: one flag denoting successfull usage of the skill while the other denotes the faulty usage of a skill. after each usage of the skill check if this was faulty/neutral/successfull and set the according flag to one. if both falgs are one, increase the skill infinitesimally and reset the flags.