Artificial learners adopting normative conventions from human teachers

Abstract

Abstract This survey provides an overview of implemented systems, theoretical work, as well as studies of biological systems relevant to the design of artificial learners trying to figure out what a human teacher would like them to do. Implementations of artificial learners are covered, with a focus on experiments trying to find better interpretations of human behavior, as well as algorithms that autonomously improve a model of the teacher. A distinction is made between learners trying to interpret teacher behavior in order to learn what the teacher would like the learner to do on the one hand, and learners whose explicit or implicit goal is to get something from the teacher on the other hand (for example rewards, or knowledge about how the world works). The survey covers the former type of systems. Human teachers are covered, focusing on studies that say something concrete about how one should interpret the behavior of a human teacher that is interacting with an artificial learner. Certain types of biological learners are interesting as inspiration for the types of artificial systems we are concerned with. The survey focus on studies of biological learners adopting normative conventions, as well as joint intentionality team efforts.