A major practical barrier to creating artificial life is simply the vastness of life in the biological realm, both in terms of the numbers of molecules comprising the biosphere, and of the billions of years over which evolution has been proceeding. Even if artificial life were a theoretical possibility, why should we expect it to proceed any faster on a computer only capable of supporting a trivially small population?
If we are eventually hoping to produce intelligent artificial life, the chances seem even slimmer. Gould remarks that ``Homo sapiens is an entity, not a tendency'' [Gould 89] (p.320). Expanding upon this, he says:
``Life arose at least 3.5 billion years ago, about as soon as the earth became cool enough for stability of the chief chemical components ... [However], a good deal more than half the history of life is a story of prokaryotic cells alone, and only the last one-sixth of life's time on earth has included multicellular animals ... But cosmologists tell us that the sun is just about at the halfway point of existence in its current state; and that some five billion years from now, it will explode ... Since human intelligence arose just a geological second ago, we face the stunning fact that the evolution of self-consciousness required about half of the earth's potential time. Run the tape again, and even if the same general pathways emerge, it might take twenty billion years to reach self-consciousness this time--except that the earth would be incinerated billions of years before.'' [Gould 89] (pp.309-311).
Another argument against the possibility of artificial life concerns the nature of the major evolutionary innovations in biology, which seem to have provided organic evolution with an unlimited supply of phenotypic novelty. It has been suggested by Howard Pattee that most of these innovations involve the invention of techniques for measuring new aspects of the environment, and of arbitrarily mapping these patterns to symbols or to specific actions [Pattee 88]. Pattee emphasises that one of the major hurdles for (strong) artificial life will be to develop a satisfactory theory of measurement which suggests how we might equip our programs with the potential to evolve new measurement devices (ibid.). The evolution of new forms of action (e.g. new effectors) is equally problematic.
Furthermore, such innovations usually develop from existing phenotypic structures employed for unrelated functions, but which happen to perform (however crudely) the novel task as a side-effect. To the extent that this is true, this presents a problem for the designers of artificial life systems. A computer model, like any other scientific model, is an idealisation of the real world, designed to model (what the designer thinks are) the important components of a system, while ignoring irrelevant aspects. However, the present argument suggests that a primary source of evolutionary novelty might lie in precisely those aspects of the physical world which might ordinarily be deemed irrelevant for the purposes of constructing a model. We are again confronted by our lack of an adequate theory of measurement, as pointed out by Pattee. Even if we were to model a wide variety of properties for each component in our artificial life system, we might only expect complex forms to emerge if they existed in a very complex environment which afforded many possibilities for perception and action. Some steps which we might take to ensure a rich environment are discussed in Chapter 7, but the issue of evolving novel sensors and effectors in a computer model remains problematic.
These are formidable arguments, and there are a number of others which can also be made against the likelihood of success (see [Bonabeau & Theraulaz 94]). The number of points that computer models have in their favour seem to be few in comparison (speed of execution might be one example).3.7 I take the pragmatic attitude that we simply do not know how successful the approach might be until we try it. Eric Bonabeau and Guy Theraulaz argue that ``rather than true limitations, [issues such as these] constitute questions asked to AL [Artificial Life]. And AL is precisely a constructive way of checking whether these limitations are real obstacles'' [Bonabeau & Theraulaz 94] (p.314). A detailed discussion of the philosophy of artificial life can be found in [Boden 96]. Even if the quest for strong artificial life ultimately ends in failure, the attempt can highlight questions and gaps in our knowledge which were not apparent, or did not seem particularly relevant, beforehand, and indeed it already has done.