secret of AI from a noob

Putz

Fly shit on a window is what you are! The last thing on your mind, was your ass!

Maybe that wasn't fly shit, ah, that was SLow-Man.

Jamie

Jamie as a multiple personality, with every last component inadequate.

Jamie learned how to do personal insults in primary school and clearly hasn't learned anything since, as we already knew.

Jump off a cliff or not.

OK. For the sake of humouring you.

How would your "insight" allow an improvement on the basic maximum entropy solution for mastermind aka bulls and cows.

Target is 0-9 in four positions. Player has to guess the digits. C= right digit wrong position, B= right digit right position

Maxent always chooses the next guess to partition the remaining subset of possible answers into the largest number N of approximately equal sized chunks that it can find. This converges onto the right answer reliably in the shortest number of steps. Various shortcuts exist depending on how close of far the first guess is to the target.

It aims to maximise entropy defined as

S = -sum{0..N} (Xi.ln(Xi))

This is *NOT* the same as variance.

Hi,

An action that creates an environment of higher interconnected complexity is what I call variance, as I have no other good definition for it taking into account the environment/AI and memory. If jumping off a cliff that has higher interconnected complexity, then ya jumping off the cliff is good, but otherwise no don't jump. Whether the jumper dies is not the main question, it only depends on what action creates the higher interconnected complexity.

A human cell is a good example of very high interconnected complexity, high variance in multiple co-dependent systems, working with lots of feedback and rules. If a cell gets cancer, many of the interconnected rules break down and the interconnected complexity diminishes.

A system with very high interconnected complexity is metastable, and nature is a good example of this. In a complex world, many things are taken into account at the same time.

For an AI the goal should be to maximize the interconnected complexity in space and time.

For a trivial situation a lot of different algorithms could give ideal results.

I'm not sure if Maxent can take into account hierarchical systems, which have interconnected complexity, ie nested state machines. To determine the way to achieve high interconnected complexity from the viewpoint of an AI, requires a lot of computation to determine the existing complexity, just from observation, and attempt to interact in a way to maintain and ideally increase the complexity.

cheers, Jamie