The notion of a (spatially) heterogeneous environment, and indeed the very notion of individuality, requires that the environment has some spatial structure. Spatial structure not only introduces the notion of individuality (in the sense that a particular component in the system can be distinguished at any point in time by virtue of its particular relationship with other components), but also makes possible concepts such as compartmentation and the control of the local environment. Such concepts have important consequences for the evolution of cooperative organisations (e.g. [Maynard Smith & Szathmáry 95]), as indeed has already been demonstrated in a number of artificial evolutionary systems (for example, [Boerlijst & Hogeweg 91], [van Baalen & Rand 98]). These studies are a good start, but a great deal of further research is required to improve our knowledge of how the spatial structure of the environment affects the evolutionary behaviour of the system.
With spatial structure comes the requirement for components to move (either actively or passively), so that they may experience different environments. Active motility opens up a wide range of phenotypic and ecological possibilities, but it may be that passive diffusion is sufficient for open-ended evolution in general.
Another fundamental aspect of the environment is related to the choice between a purely logical model (where reproducing entities are configurations of states) or a material model (where the entities are related to material structures, composed of atomic units of matter, and with energetic considerations of one form or another). These choices are probably best viewed as opposite ends of a spectrum, with a host of other possibilities in between. Some of the issues concerning the consequences of this choice on the evolution of the system were discussed in Section 7.1.4. Whether a purely logical model is sufficient to capture an open-ended evolutionary process remains to be seen. However this issue is resolved, it is likely that the development of a theory of constructive dynamic systems (i.e. dynamic systems where new operators may appear intrinsically over time) will be a central requirement. The work described in Section 3.2.3 represents a useful step in this direction.
Regardless of the degree to which states are grounded in a material environment, the discussion of the advantages of genetic reproduction over self-inspection (Section 7.2.3) suggests that a fairly fundamental distinction may exist between reactive states and (quasi-)quiescent states. Whether this distinction is absolutely necessary for open-ended evolution is unknown, but, as Waddington remarked ``in practice--and perhaps because of a profound law of action-reaction--it is difficult (impossible?) to find a [molecule] which is stable enough to be an efficient store and at the same time reactive enough to be an efficient operator'' [Waddington 69] (p.115).
With respect to the types of bonding between atoms in the more materialistic models, it has generally been found that at least two types of bond are required: one strong and fairly permanent, and another weak and temporary. Examples include Penrose's analysis of self-replication ([Penrose 62]: see Section 3.2.1), Myhill's model ([Myhill 64]: see Section 3.2.1), and an unpublished model of my own, named Nidus. Holland's α-Universes had only one type of bond [Holland 76],7.25 but McMullin's implementation of this model revealed that its evolutionary behaviour was much more restricted than Holland had anticipated ([McMullin 92a], [McMullin 92b]). It is possible that this is related to the apparent requirement for two kinds of matter; in particular, a mechanism for temporary association might be necessary to allow the two kinds to interact. At this stage, however, this is just a speculation.
Finally, even though it has been argued that the genetic material should be quasi-quiescent, the fact remains that the DNA of biological organisms is actually rather active; mobile segments of DNA--transposons--play an important role in gene regulation. Furthermore, it has been suggested that features such as splicing and mobile genetic elements were present even at the prebiotic stage of evolution (e.g. [Reanney 79], [Buss 87] p.194). Taking a broader view of mobile genetic elements, the process of symbiogenesis, described in Section 2.1.1, can also be included. Barricelli's remarkable results can therefore be seen as an example of the potential of such genetic mobility. It is also relevant that Ray's latest work with Tierra has included additional system-defined operations such as insertion, deletion and crossover ([Ray & Hart 98]: see Section 3.2.1). We can therefore say that in nature, individual DNA segments (genes) seem to have retained at least some kind of individuality even when collected together on a chromosome, and that in artificial systems, mobile genetic elements sometimes seem to enhance the system's evolvability. It is easy to think of reasons why this may be the case. Whether any of this has any bearing on the necessary features of a genetic system able to support open-ended evolution is an open question.