How to make agents learn from human input?

I do understand it is possible to use human_input for the task but , how to make the agent learn from that human_input in training mode?