Learning from Little Human Feedback [R] [P]

I have an environment where the input is an image, which may or may not include a bounding box. An agent that determines if a human likes the bounding box based on the provided image and the human's actions. Now, I want to modify this setup so that the agent can adapt to any other human's preferences with minimal feedback. How can I achieve this?

adapt agent box environment feedback human human feedback image machinelearning setup