The Critical Rationalist Vol. 01 No. 04 ISSN: 1393-3809 31-Dec-1996
References
5.2 ... or Metaphysics?
... I do not really believe that we shall succeed in creating life artificially; but after having reached the moon and landed a spaceship or two on Mars, I realize that this disbelief of mine means very little. But computers are totally different from brains, whose function is not primarily to compute but to guide and balance an organism and help it to stay alive. It is for this reason that the first step of nature toward an intelligent mind was the creation of life, and I think that should we artificially create an intelligent mind, we would have to follow the same path.
(109) My interest in Darwinian theory stems from attempts to realise something like a spontaneous growth of knowledge, by Darwinian means, in artificial systems--a problem which is now generally captured under the rubric of Artificial Life (Langton 1989). I may note in passing that, as the quotation above indicates, Popper deserves to be recognised as at least one of the founders of this field; indeed, as early as 1961, in his Herbert Spencer lecture Evolution and the Tree of Life, Popper presented a moderately elaborate, if schematic, discussion of the possible evolutionary growth of knowledge of a strictly artificial (robotic) system (Popper 1961).
(110) So, in this epilogue I would like to briefly wave my hands in the direction of some recent work in Alife, and the problems considered in the article.
(111) First, let me note that any attempt to realise
the growth of adaptive complexity, via (genuinely)
Darwinian processes, in artificial systems, could, in
itself, represent a further kind of test of
. While such tests could hardly provide a
strong refutation of (due to the
necessarily limited scale of artificial systems
compared to biological evolution) the
successful demonstration of significant,
spontaneous, growth in adaptive complexity in
artificial systems, were it to be achieved, might still
represent a significant corroboration for
.
(112) I hasten to add that I do not consider that the achievements to date in this direction should be overrated--but there are some intriguing straws in the wind, at least.
(113) Among the best known efforts in this direction are the BioMorphs of Richard Dawkins (1986) and the Tierra system of Tom Ray (1992). However, the BioMorphs system is limited to static (albeit very "lifelike") images of artificial "organisms"; and while Tierra does involve dynamic entities (fragments of computer code) their environment is so alien to our normal experience that it is very difficult to assess whether "knowledge", in anything like its Popperian sense, is being created or discovered.
(114) In my view then, the most impressive work of this sort to date is that on evolving "virtual creatures" by Karl Sims (Sims 1994b, Sims 1994a). Sims first constructed a "virtual" (computational) environment which reasonably faithfully models normal, three dimensional, Newtonian space--optionally including a viscous medium and/or a uniform gravitational field. He was able to embed within this environment virtual or simulated creatures. These creatures are "grown" from "genetic" descriptions. They consist of roughly cuboid components, of various sizes, jointed together in various ways, and controlled by a distributed artificial "nervous system". Starting from randomised descriptions, and using artificial selection (i.e. controlled by a predetermined "fitness function") Sims has been able to demonstrate the spontaneous evolution of creatures exhibiting swimming, walking, and jumping behaviours; creatures which can track an environmental stimulus; and creatures which can compete (in very stylized combat rituals) for possession of "resources".
(115) Furthermore, in achieving these behaviours, many of Sims' creatures exhibit morphologies and organisation that would undoubtedly be called "adaptations" if found in biological organisms. I suggest that there is at least a prima facie case for claiming that these creatures have spontaneously acquired significant knowledge of their world; that this growth of knowledge was by fundamentally Darwinian means; and that this lends some degree of independent corroboration to conventional, biological, Darwinism. At the very least, I suggest that this work demonstrates that there is more to Darwinism than a tautology!
(116) Having said that, there are also limitations in this work. For example, reproduction and development do not take place within the virtual environment, thus limiting the opportunities for anything like somatic time learning.
(117) But, more importantly I think, selection is driven by a predefined, and externally imposed, evaluation function. I suggest that, for this reason, evolution in this system may be effectively limited to the refinement of what Popper (1961, p. 275) called the skill-structure of the creature. The higher level aim-structure is effectively determined by the imposed evaluation function (as "walk", or "swim", or "fight" etc.). Popper conjectured that mutations in the aim-structure might, in the biological world, explain certain very significant long term phylogenetic trends. In my terms, this seems to offer something like a mechanism for a long term correlation between knowledge and S-value--which is to say, a potential explanation for major episodes of Darwinian growth of knowledge. But it seems to me that, for the moment, this possibility is effectively closed off in Sims' system.
(118) The central difficulty in all of this is, of
course, that the empirical phenomena with which
deals are of such a scale in both time and
space that they are very difficult to subject to any
severe test, even via computer simulations. I remain
convinced that is, in principle, testable;
but I will leave the last word with Popper again:
... we have to add that the phrase in principle is a very important restriction. Neither Darwin nor any Darwinian has so far given an actual causal explanation of the adaptive evolution of any single organism or any single organ. All that has been shown--and this is very much--is that such explanations might exist (that is to say, that they are logically possible).
References
5.2 ... or Metaphysics?
The Critical Rationalist Vol. 01 No. 04 ISSN: 1393-3809 31-Dec-1996
Copyright © 1996 All Rights Reserved.
TCR Issue Timestamp: Tue Dec 31 17:37:08 GMT 1996