------------------------- This directory contains the text of: @PHDTHESIS{ McMullin:ArtificialKnowledge, AUTHOR = "Finbarr (Barry) Vincent McMullin", TITLE = "Artificial Knowledge: An Evolutionary Approach", SCHOOL = "Ollscoil na h{\'E}ireann, The National University of Ireland, University College Dublin, Department of Computer Science", YEAR = "1992", URL = "ftp://ftp.eeng.dcu.ie/pub/autonomy/bmcm_phd/" } ------------------------- ABSTRACT: --------- I present a new analysis of the _problem situation_ in Artificial Intelligence (AI), grounded in a Popperian epistemology. I first review arguments purporting to establish that no purely ``computational'' system can realise _genuine_ mentality. I conclude that the question is still open; but that the more pressing question is whether such a system can even exhibit intelligent _behaviour_. Attention is thus directed at the computational embodiment of _knowledge_, and its growth_. I suggest that much of the work in this area incorporates a flawed, naive empiricist, epistemology. I adopt Popper's view that the growth of knowledge is possible _only_ through a process of _unjustified variation and selective retention_. In particular, the inate knowledge of biological organisms has arisen by such a process, in the form of Darwinian evolution. I review previous work on the realisation of Darwinian processes in computational systems. In particular, I present a critical reinterpretation of von Neumann's pioneering work in this area. I conclude that no system to date has exhibited substantive growth of artificial knowledge via a process of Darwinian evolution. More importantly, I argue that this problem is deeper than is generally recognised, requiring the effective integration of _autopoiesis_ with _evolvability_. To achieve this it may ultimately be necessary to realise something analogous to the _genesis of life_. I review one proposal for such a phenomenon: Holland's so-called Alpha-Universes. I present an _implementation_ of a specific Alpha-Universe and review the (largely negative) results of empirical tests carried out on it. I conclude with the claim that the problem of realising the spontaneous genesis of ``artificial life'' is of great difficulty, but that its solution may yet prove to be an essential prerequisite for the realisation of anything deserving to be called ``artificial intelligence''. ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ | Barry McMullin, Autonomous Systems Group, | McMullin@EENG.DCU.IE | | School of Electronic Engineering, | Voice: +353-1-704-5432 | | Dublin City University, Dublin 9, IRELAND. | FAX: +353-1-704-5508 | ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++