When I was 13, I don't remember if it was mandatory to participate in the Science Fair or not. Either way, I did... and I decided I wanted to write a program that had an "artificially intelligent" maze solving mouse. I found the pages from the display and typed them in and scanned the pictures into a Google Doc:
It had been sort of an excuse to try doing some machine language programming, which I used a small bit of in order to make the maze drawing faster. I also learned about how to do direct accesses to screen memory, so that I could read and write the bytes without needing additional storage. I was getting really into how to encode large amounts of data efficiently in binary memory blocks, and so it was a good time of discovering all that.
I didn't really do any research on maze-solving algorithms. Seems that I recall trying, but having not taken any computer science... there would have been a lot of learning required to get started on algorithms. And back in my day we didn't have this fancy Internet. So if the local library only had a book talking about maze solving that was over your head then you were just kind of screwed.
So in the terminology I used, I equated a mouse "learning" a maze with being able to solve it...and then have its "brain" saved and reused to do the same thing. I gave it a seemingly-arbitrary 2048 byte limit on its memory, which I used for it to store its decision for how it would react when it saw a certain type of local configuration. Then each time it saw that configuration again, it would go the same direction.
Unsurprisingly, if solving a maze required visiting two identical-looking intersections and a different choice was needed to be made in the two locations...such a mouse was hopeless. But I didn't want the mouse to be just remembering a log of what it had done. Its brain was supposed to be the DNA for reasoning about what it should do.
Note
I probably should have called it Insanity Mouse instead of Learning Mouse. "Insanity: doing the same thing over and over again and expecting different results." I'll also point out that it's completely unsurprising that a student in the American public school system would equate "memorization" with "learning". :-/
The better idea I had was to let the mouse leave behind a kind of trail-of-bread-crumbs in screen memory, which could be used to help decide where it had already looked or not looked. If I find the time I might look into whether the technique I called "Triple Trail Priority" actually can solve any category of maze I was using...and if it can, if it's minimal or if there's some other way of doing it. But in the meantime, here it is to be amused by. :-)
