Self-generated learning occurs when a learning system behaves to generate useful data for a particular context. A domain-general learning system should be able to shape their future experiences and enhance learning across a wide variety of contexts.
One example of self-generated learning is in human development. Children physically interact with and explore their environment to learn about the world. These actions include picking up and examining objects, and tracking parents’ gaze.
According to the Interventionist theory of causation, the data created in self-generated learning is most effective for determining causal relationships (i.e., forming scientific explanations) in the world.