Some notes about applying human psychology to creating a chatbot.read more
We know reinforcement based learning is a possibility for making an AI that can play video games.
Trying to emulate how babies learn:
Randomly output characters of varying lengths, based on certain input (textual? or maybe this should be something else, like teaching the machine how to count, so we input 1…2..3… and they should continue with 4…5…6…), with some spaces. Allow for feedback to reinforce/discourage certain output. The reason for suggesting counting is because that way we don’t have to deal with decoding and attempting to understand/represent language.
More sophisticated: we would input a sentence, and the machine would begin by randomly responding with characters, again rewarding for good responses. The problem here is that we would need the machine to get some representation of the textual input, I believe.