Where is Knowledge in Robotics? Some Methodological Issues on Symbolic and Connectionist Perspectives of AI
In this chapter we consider a number of methodological issues to which little importance is normally attributed in robotics but which we consider essential to the development of integrated methods of soft and hard Computing and to the understanding of the artificial intelligence (AI) purpose and fundamentals.
The basic conjecture in this chapter is that knowledge always remains at the knowledge level and in the external observer’s domain. To the robot only pass the formai model underlying these models of knowledge. Consequently, there are neither essential differences between symbolic and connectionist techniques nor between soft and hard Computing. They are different inferences and problem-solving-methods (PSMs) that belong to a library and that are selected to be used in a sequential or concurrent manner according to the suitability for decomposing the task under consideration, until we arrive to the level of inferential primitives solved in terms of data and relations specifie of the application domain.
The distinctive characteristics of hard and soft Computing methods are related to the balance between knowledge and data available in advance, the granularity of the model or the necessity and capacity of learning in real time. Nevertheless, in ail these cases the knowledge (the meaning of the entities and relations of the model) always is outside the robot, at the knowledge level, the “house” of models.
In many publications, the robotic programs are described without including âny distinction between levels and domains of description of a calculus. As a resuit, it is generally difficult to determine what the robot actually performs, which knowledge has been represented, and which is artificially injected during the human interpretation of the robots behavior.
In order to make clear this methodological issues we consider the taxonomy of levels introduced by Marr  and Newell  (Knowledge, Symbols, and Hardware) put on the top of the two domains of description of a calculus (the domain proper of each level and that of the observer external to the computation of the level). Then, we describe the usual approach to modeling and reduction of models from the knowledge to the symbol level and finally we illustrate the analogies and differences between different models and reduction processes including the opera-tional stage, either symbolic, connectionist, probabilistic or fuzzy. In ail the cases our conviction is that most of the work must be made by modeling tasks and PSMs at the knowledge level, where it is crystal clear that soft and hard Computing are complementary and ready to be integrated.
KeywordsNatural Language Knowledge Level Physical Level External Observer Automaton Theory
Unable to display preview. Download preview PDF.
- 1.Marr D (1982) Vision. Preeman, New YorkGoogle Scholar
- 2.Newell A (1981) The knowledge level. AI Magazine, summer 1–20Google Scholar
- 3.Mira J, Delgado AE, Boticario JG, Diez FJ (1995) Aspectos básicos de la in-teligencia artificial. Sanz y Torres, SL MadridGoogle Scholar
- 4.Chomsky N.(1968) Language and Mind. Harcourt Brace & World Inc., N. YorkGoogle Scholar
- 5.Pylyshyn ZW (1986) Computation and cognition. Towards a foundation for cog-nitive science. The MIT Press. Cambridge, MassGoogle Scholar
- 6.Clancey WJ (1985) Heuristic classification. Int J Artif Intell Neural Netw, Com-plex Problem-Solving Technol 27: 289–350Google Scholar
- 8.Breuker J, Van de Velde W (eds) (1994) Common KADS library for expertise modelling. IOS Press, AmsterdamGoogle Scholar
- 10.Schreiber et al. (1999), Engineering and managing knowledge: The Com-monKADS methodology. The MIT PressGoogle Scholar
- 11.Fensel D et al., (1999) UPML: A framework for knowledge System reuse. Proc 17th Int’l Joint Conf Artificial Intelligence (IJCAI99), Morgan Kaufmann, San FranciscoGoogle Scholar
- 13.Gennari JH, Grosso W, Musen M (2000) A method-description language: An initial ontology with examples. In: Proc Europ Knowledge Acquisition Conference (EKAW-2000). LN in Artificial Intelligence. Springer, BerlinGoogle Scholar
- 15.Varela FJ (1979) Principles of biological autonomy. The North Holland Series in General Systems Research. North HollandGoogle Scholar
- 16.Mira J (1995) Reverse neurophysiology: The ‘Embodiments of Mind Revisited’. In: Moreno R, Mira J (eds) Brain processes, theories and models. The MIT Press.Massachusetts, pp 37–49Google Scholar
- 17.Mira J, Delgado AE (1987) Some comments on the antropocentric viewpoint in the neurocybernetic methodology. Proc of the Seventh International Congress of Cybernetics and Systems; 2: 891–95Google Scholar
- 19.Herrero JC, Mira J (2000) Causality levels in SCHEMA: A knowledge edition interface. IEE Proceeding-Software Engineering, vol 147 No 5 pp 191–200Google Scholar
- 20.Fensel D, Anciele J, Studer R (1994) The specification language KARL and its declarative semantics. IS-CORE94 Workshop, pp 27–30Google Scholar
- 21.Mira J (2000) Computación y antropomorfismo en robótica emocional. In: Mora F (ed) El cerebro sintiente. Ariel, Barcelona, pp 153–185Google Scholar
- 24.Chandrasekaran B (1986) Generic tasks in knowledge based reasoning: High-level building blocks for expert System design. IEEE Expert fall: 23–29Google Scholar
- 26.Benjamins R (1995) Problem-solving methods for diagnosis and their rôle in KNOWLEDGE acquisition. INET F of Expert Systems: Research and Applications, vol 2 no 8: 93–120Google Scholar
- 32.Herrero JC, Mira J (1998) In search of a common structure underlying a representative set of generic tasks and methods: The hierarchical classification and therapy planning cases study. In: Mira J, del Pobil AP, Moonis A. (eds) Method-ology and Tools in Knowledge-Based Systems. LNAI 1415. Springer-Verlag, Berlin, pp 21–36CrossRefGoogle Scholar
- 35.Mira J, Herrero JC, Delgado AE (1998) Where is knowledge in computational intelligence?: On the reduction F THF KNOWLEDGE level to the level below. Proc 24th EUROMICRO Conference, IEEE vol II pp 723–732Google Scholar
- 36.Mira J, Delgado AE (1995) Computación neuronal avanzada: fundamentos bi-olôgicos y aspectos metodológicos. In Barro S, Mira J (eds) Computación neuronal. Universidad de Santiago de Compostela, Cap. VI, 125–178Google Scholar
- 37.Mira J, Martínez R, Álvarez JR, Delgado AE (2001) DIAGEN-WebDB: A connectionist approach to medical knowledge representation and inference. In: Mira J, Prieto A (eds.) Connectionist models of neurons, learning processes, and artificial intelligence. LNCS 2084, Springer Verlag, Berlin, pp 772–782CrossRefGoogle Scholar
- 38.Moreno R, Mira J (1995) Logic and neural nets: variations on themes by WS McCulloch. In: Moreno R, Mira J (eds) Brain processes, theories and models. The MIT Press. Massachusetts, pp 24–36Google Scholar