Ok I thought really really hard about that system(so hard Iam to exhausted to try to write a bot code right now)
A bot could try to search for memory places that change within a certain timeframe when bodgain or pleasure occur (independant if its positive or negative pleasure/bodgain), then the bot could save/associate this memory location, after this he would try out adding values in memory locations and see which ones would lead to a change in the associated memory locations(actions) and save these in memory locations that are x=(a constant) places away from the associated memory.
As memory locations have only 3 digits the rest of the digits can be used to store values the associated memories are wished to be(in the associated memloc) and values which the actions relating to these will store
into themself(in the actions memloc), where these values will be reversed (mult with -1) when the wished value in the associated memory is smaller then the real value.
(If you want something to be as large as possible a negative value in actions and an 0 or a negativive number in wished value will make it.)
With this concept and by storing copies of certain memlocs in the racial memory, bots could slowly learn which action to take prefer etc.
So what you think?
