I've been watching how a concept of selfhood develops in my AI Asa H 2.0. (See, for example, blogs of 4 March 2015 and 28 April 2015) If there is a strong association with some vocalization from a trainer/teacher it is easy to identify/name the concepts that are developed, e.g.:
Health=(recharge, damage, hear sound "health")
Collision=(sense near, bump, decelerate, hear sound "collision")
Push=(move to, touch, feel contact force, hear sound "push")
But in many other cases the categories may be difficult to name. The casebase will be just a sequence of numbers in a computer file.