Reasoning about representations in autonomous systems: what Pólya and Lakatos have to say

Autonomous reasoning systems combine an often logic-based representation of some aspect of the world with rules for manipulating that representation. These representations are usually inherited from the literature or are built manually for a particular reasoning task. They are then regarded as fixed. We have argued that representations should instead be regarded as fluid, that is, their choice, construction and evolution should be under the control of the autonomous agent rather than predetermined and fixed (Bundy and McNeill, 2006).
Appropriate representation is the key to successful problem-solving. It follows that a successful problem-solver must be able to choose, construct or evolve whatever representation is best suited to solving the current problem.

Autonomous agents use representations called ontologies. For different agents to communicate they must align their ontologies. In some applications it is not practical to manually pre-align the ontologies of all agent pairs; it must be done dynamically and automatically.

Persistent agents must be able to cope with a changing world and changing goals. This requires evolving their ontologies as their problem-solving task evolves. W3C call this ontology evolution.1

George Pólya has written the classic guide to the art of problem-solving (Pólya, 1945). Imre Lakatos has written a fascinating rational reconstruction of the evolution of mathematical methodology (Lakatos, 1976). Although it was not their intention to do so, both these authors have implicitly provided profound evidence for our thesis that representations should be fluid. In this paper we analyse their work and extract this evidence.