Universal Learning

While machine learning has been quite successful in many different domains, we are still quite far from machines that can learn at the level of humans in disparate domains. Can be there a succinct specification that can place us much closer to human-level learning when achieved? We suggest one such specification: universal learning