an idea for the basis of an ai that should not need updated.

[psychopompos]

based on my perception of the human mind doing no more then discern, weigh and act on probabilities, i wonder if an ai could work on this basis.

humans are as it where, programmed with natures known working strategies, strength in numbers, communication and the drive to get the info we need to properly compete in the world.
we then take that pre-set, and develop environmental skills, social skills, job and world related skills in order to out compete our peers.
this gives the probability matrix more accurate weights toward what actions should be beneficial.

with that, would it be possible to make an ai on a similar framework?
one that did not start with end strategies, but was a learning framework?

[Zarel]

Yes, um, such an idea is great in theory, but they tend not to work very well in practice.

[psychopompos]

they learn stupid things?
or dont learn?
or ... ?

[nux]

Sometimes, i dream of such an AI, that would be able to learnt from its own mistakes, or even from the opponent's strategy. But i am afraid it will be only a dream for a long time...
In my opinion, the most difficult to implement in "discern, weigh and act on probabilities" is "discern". It require pattern recognition which is what the human brain is very good at, and is awfully difficult to program in a computer.