This refer to my previous post : https://steemit.com/science/@modaf/first-ai-will-be-created-before-2070
Here we will try a "one-layer" AI, which will try to solve the "quickest path" problem : finding the quickest path between a point light blue and a red one, here is an example :
To do so we first solve a lot of random problems, saving some of them (we do not save too close problems, by a chosen norm). Then when we want to solve a new problem we compare it to what we saved : this is what human do (even if he doesn't save it but simulate it). Here is the success rate of this AI with time allowed for calculation :
This is the same as human : we can really quickly find a solution but we need a tremendous amount of time to be sure it's the correct one. Here is another one :
But, when the problem to solve is becoming much harder (here it will be just increasing the number of nodes), the success rate of this AI is proportional to the time allowed for calculation ! As we can see it here :
This can sound strange, but humans do so too. Indeed, when a problem is really hard, the first solution that we guess is really often wrong and therefore we need to correct each part of it (if we split it in different parts), doing so our success rate is proportional to the number of part verified and therefore proportional to time spent for calculation.
Now lets compare to Dijkstra's algorithm : Dijkstra's algorithm takes 0.4ms to execute, whereas this AI can in 0.1ms have a +77% success rate, but it needs 100ms to get +99% success rate. It confirms our intuition : human can be really quick but unsure, or sure but so slow.
A cool thing about this AI is that we can stop calculation at any time to get an answer : just close it and get variables, which is quite cool :)
In next part we will try to deal with the traveling salesman problem, which is so hard to compute to get a correct answer (and where humans can use their intelligence to get really quickly a solution : we will therefore see what this type of AI can do is those hard situation)
