Seven Desirable Properties for Artificial Learning Systems

We propose seven desirable properties for artificial learning systems, namely: incrementality, nonmonotonicity, inconsistency and conflicting defaults handling, abstraction, self-organization, generalization, and computational tractability. This proposed set of properties is not claimed to be complete, nor does it imply that all properties must necessarily be present in all learning models. Rather, it focuses on issues that are not often explicitly addressed and provides a basic foundation for the design of more efficient algorithms. As most other computer-based systems, learning systems are difficult to reverse engineer, and desirable properties should be part of the original design rather than retrofitted into the system after the fact. Much effort has been devoted to understanding learning and reasoning in artificial intelligence, giving rise to a wide collection of models. For the most part, these models focus on some observed characteristic of human learning, such as induction or analogy, in an effort to emulate (and possibly exceed) human abilities. We propose seven desirable properties for artificial learning systems: incrementality, non-monotonicity, inconsistency and conflicting defaults handling, abstraction, selforganization, generalization, and computational tractability. We examine each of these properties in turn and show how their (combined) use can improve learning and reasoning, as well as potentially widen the range of applications of artificial learning systems. An overview of the algorithm PDL2, that begins to integrate the above properties, is given as a proof of concept. In the following sections, we examine each property in turn and show how their use can improve learning and reasoning, and potentially widen the range of applications of artificial learning systems. As a proof of concept, the algorithm PDL2 [5] is overviewed. PDL2 begins to integrate the above properties into a unified framework in which inductive learning supplements the use of prior knowledge to yield a model capable of dealing with a greater variety of interesting problems.