I’ve been developing an AI that can learn from the user’s input (basically it’s a bot with neural system), in Rust. It’s in the beginning stage, but as soon as possible I would have some working stuff.
I took the MegaHal’s idea of selecting the best answer based on its information (the more information, the better the answer is), and I’m studying hard about neural system and brain stuff.
Oh, about the learning system: the whole program’s brain is based on signals being sent through other neurons. A short circuit of neurons would receive these signals and match some patterns on them.
From a bird eye view, the patterns represent a data (a word, a signal etc), which can be matched with other data, like so:
// Let's say that each number represents a word:
0001 = "food"
0011 = "melon"
0111 = "fruit"
0001 = "fries"
1000 = "table"
// To check whether a word would match with
// another, we just do an AND operador on them:
food & melon -> 0001 & 0011 = 0001 // food
food & table -> 0001 & 1000 = 0000 // no match
So, the thing is that you can compare the pattern of words and see whether they match (to form a linkage between them) or not, and by that, construct memories and answers… And to learn!