i used a list to simulate nodes, then made it so it keeps track of its moves and when it succeeds it adds its moves to the move list and when it fails it deletes its move history it slowly gets faster at escaping Snap! Build Your Own Blocks
(oh yeah the edge is its goal)
i... dont get it
AI is for Artificial Insect?
It feels like a truly realistic simulation of a wasp, stubbornly trying to get outside through a fixed glass surface, while ignoring the adjacent open window. 
insects do have brains
no
Agreed.
So, if you wish, you may consider my observation a compliment for the AI-simulation's realism.
yes a mosquito slowly realizing how to leave
I watched it for a few minutes, and the AI wasn't succesful yet.
I guess that's what @joecooldoo hinted at.
The demonstration might work better if you program it to "learn" somewhat faster.
bro what most ai takes years to train
Very clever 
This way, the program is much faster but still gives some feedback.
This "ai" is stateless. I'm curious if it can handle more complex scenarios, like avoiding extra walls.
Semantics, but AI can be me throwing a ball at a wall to determine the weather in two weeks. Nothing about that - not the ball, not the wall, especially not the idea - is actually intelligent (what really is intelligence anyways), but anyone could claim that it is, thus the artificial bit of artificial intelligence. It isn't real.
That's voodoo
with the state of silicon valley i am inclined to say that the two are one and the same
I've added some metrics to the training process: Username=dardoro&ProjectName=boxed%20in%20ai&editMode&noRun
For the simplest arena (costume "Basic"), after 20.000 iterations, I've got a success ratio of about 50%.
For the "Narrow" costume I've got only ~25%.
For the "Obstacle" there was 0% after 20.000 tries.
I'm not sure if it can go any better ... and the "model" tends to be huge (10^6 items).
You can improve the ai by making the ai also learn based off of what it has previously learned
