Abstract

This dissertation examines the functional roles symbols play in biological organisms, scientific models and adaptive learning devices, analyzing the process of how symbols acquire new functions. The semiotic categories of syntactics, semantics, and pragmatics are used to examine the functioning of symbols in organisms, models, and devices. The dissertation explores how we would go about designing self-organizing devices which adaptively construct their own relationships to the physical world . It bears on the frame/feature-generation problem in artificial intelligence, the problem of machine creativity in philosophy, the measurement problem in physics, the problem of generating new observables in science, and the problem of emergent functions in evolutionary biology. Much of the perspective comes out of the work of theoretical biologists Howard Pattee and Robert Rosen. ;An analytical framework is developed for distinguishing between computation, measurement, control, and nonsymbolic functionalities. A taxonomy of adaptive cybernetic devices is proposed in which nonadaptive devices are distinguished from computationally adaptive ones and semantically adaptive ones . Evolutionary devices adaptively construct their own sensors and effectors contingent upon their performance, and are thus qualitatively different from devices now in use. Evolutionary devices are analogous to the biological evolution of new sense and effector organs. Relative capacities and limitations of the device types are outlined. ;Three contemporary conceptions of emergence are distinguished: computational emergence, thermodynamic emergence, and emergence-relative-to-a-model. Levels of adaptivity are connected to types of emergence. ;Throughout, purely computational, logic-driven strategies for generating new functions are criticized as being incapable of generating new primitives. It is argued that computers have completely nonemergent behavior because of their completely specified, digital, symbolic nature, while biological organisms are capable of emergent behavior because they also include analog, nonsymbolic processes. An evolutionary robotics research program is suggested as an alternative to purely computational approaches