To digress a little, with regard to the issue of how symbolic information arises in evolution (discussed, for example, in [Pattee 95b]), this requirement ensures that the matter-symbol relationship is inherent in the system from the beginning. The material is selected for its phenotypic properties, but it is its genetic information which is passed on to its offspring. In this situation, it is necessary to assume that by inheriting this genotype, the offspring will also share the phenotypic properties. For example, in a simple RNA-world scenario (see Section 2.2), we could imagine that molecules which inherit a particular sequence of bases would adopt a particular three-dimensional structure, which might, say, confer specific catalytic properties (as demonstrated in [Zaug & Cech 86]). We could therefore regard the genetic information (the sequence of bases on the RNA molecule) as a symbolic representation of its phenotypic properties (its catalytic action in this example). However, the question of how explicit interpretation machinery evolves is more complicated (as mentioned in the previous section).
Returning to the main topic of discussion, Barricelli was well aware of the need for reproducers to perform other tasks when he designed his artificial life platform in the early 1950s. He says ``It may appear that the properties one would have to assign to a population of self-reproducing elements in order to obtain Darwinian evolution are of a spectacular simplicity. The elements would only have to: (1) Be self-reproducing and (2) Undergo hereditary changes (mutations) in order to permit evolution by a process based on the survival of the fittest'' [Barricelli 62] (pp.70-71). He goes on to describe a simple discrete one-dimensional model where each cell is either empty or contains an integer number. The numbers reproduce according to the implicit rules of the system (`trivial reproduction' in the common use of the phrase), and mutations arise under certain circumstances. This simple model therefore fulfils the fundamental requirements for an evolutionary process. However, as Barricelli notes, this model of evolution ``clearly shows that something more is needed to understand the formation of organs and properties with a complexity comparable to those of living organisms. No matter how many mutations occur, the numbers ... will never become anything more complex than plain numbers'' (ibid. p.73). Barricelli therefore concentrated on looking for the `missing ingredient'.7.16 It should be noted that von Neumann, also, was not so much interested in machines which could only self-reproduce, but rather in machines which could perform other tasks as well ([von Neumann 66] p.92; see also [McMullin 92a] pp.174-175).
The preceding arguments are leading us in the direction of requiring a form of proto-DNA which reproduces due to the implicit laws of the environment in which it exists, but which also explicitly specifies some properties which can be selected for or against in an evolutionary process. At this point we might note that artificial evolutionary systems which have just these properties already exist, and indeed their use is widespread--I am of course referring to genetic algorithms (e.g. [Holland 75], [Goldberg 89]), genetic programming (e.g. [Koza 92]) and similar techniques. The difference is that we require a system with the potential for a large degree of intrinsic adaptation for modelling biological evolution, rather than a system where the selection of individuals is determined by an externally-defined fitness function (see Chapter 3, Sections 3.1.1, 3.2.1 and 3.2.2). Intrinsic adaptation is introduced when the domain of interaction of the individual replicators is within the evolving system itself. This is in contrast to systems with an explicitly defined fitness function, where the replicators do not directly interact with other replicators. Ray recognised this point himself when discussing the design of artificial life platforms:
``What all of this discussion points to is the importance of imbedding evolving synthetic organisms into a context in which they may interact with other evolving organisms. A counter example is the standard implementations of genetic algorithms in which the evolving entities interact only with the fitness function, and never `see' the other entities in the population. Many interesting behavioral, ecological and evolutionary phenomena can only emerge from interactions among the evolving entities.'' [Ray 94b] (Section 11.1).
A small but nevertheless very important point should be emphasised here. In considering particular types of interactions and sources of selection, we are no longer considering the nature of generic evolutionary processes per se. Rather, we are now starting to think specifically about the particular kinds of evolutionary processes that might be capable of supporting phenomena associated with biological organisms (e.g. specific ecological interactions). This involves consideration not only of the nature of the individual replicators, but also of how they interact with each other, and of the general properties of the environment in which they exist. Such issues are fundamental to the design of artificial life platforms, but have so far received little attention from the Artificial Life community. We will return to this topic in Sections 7.3.2 and 7.3.3.
Similar arguments for proto-DNA with the properties of implicit reproduction and the potential for explicitly-encoded attributes with selective significance have been put forward by McMullin [McMullin 92a] (p.267), who points out the connection with Cairns-Smith's general model for the original of terrestrial life based upon inorganic information carriers (i.e. clay minerals, as mentioned in Section 2.2).