next up previous contents
Next: Predefined Organism Structure Up: Problems with Tierra-like Models Previous: Problems with Tierra-like Models

Lack of an Explicit Theoretical Grounding

Although Cosmos was designed to study evolution, and in particular (originally, at least) the evolution of multicellular organisms from unicellular ones, it was not built around any particular theory of what the important features of this transition might have been. Features such as the two-dimensional environment, energy tokens and so on were included for the reasons discussed in Chapter 4, but there were no coherent theoretical reasons for deciding which features should be modelled, and which should be left out. This weakness is not specific to Cosmos, but is shared by all of the other Tierra-like systems I have come across.

In describing the philosophy behind the Tierra system, Tom Ray explains that ``rather than attempting to create prebiotic conditions from which life may emerge, this approach involves engineering over the early history of life to design complex evolvable organisms, and then attempting to create conditions that will set off a spontaneous evolutionary process of increasing diversity and complexity of organisms'' [Ray 91] (p.373). However, in order to `engineer over' several billion years of evolution, we would need to have a very good idea of the design and behaviour of the resulting organisms, and an understanding of why they had evolved in such a way (in order to know which aspects of their design and behaviour were the most important for us to model).7.1 Unfortunately we do not possess such details of the organisms which existed immediately prior to the Cambrian explosion.

I am certainly not the first person to criticise artificial life models on these grounds (see the references in Section 3.1.3). For example, Howard Pattee warns that ``simulations that are dependent on ad hoc and special-purpose rules and constraints for their mimicry cannot be used to support theories of life'' [Pattee 88] (p.68).

To be fair, Ray does offer a definition of life in his work with Tierra. He says ``I would consider a system to be living if it is self-replicating, and capable of open-ended evolution'' [Ray 91] (p.372). From the discussion of the concept of life in Chapter 2 it will be clear that this definition is controversial. Now, this lack of agreement is not in itself a particular problem when formulating a scientific model, as long as a precise definition has at least been proposed (and is falsifiable). However, in Section 7.2 I will argue that the concept of self-replication, when simply stated, is not a precise concept, and is potentially the source of much confusion. Additionally, determining necessary and sufficient conditions for a system to be capable of open-ended evolution is half of the problem, and Ray's definition tells us nothing about how we should go about building such a system. This being the case, the definition does not provide an adequate theoretical grounding for Tierra and similar models.

A feature of Ray's definition of life is that it does not define what sorts of environments might support life, or the sorts of ecological interactions which should be available. In Section 2.3.1 it was suggested that ecological processes may play a primary role in promoting evolutionary progress and the evolutionary increase of complexity of some organisms. Furthermore, in Section 3.2.2 we saw that some of the most spectacular examples of artificial evolution rely upon coevolutionary interactions between organisms. This suggests that we should think more carefully about such issues, rather than treating them in the rather ad hoc way that has often been used in the past. This point has been made by Pattee, who says:

``... life must have arisen and evolved in a nonliving milieu. In real life we call this the real physical world. If artificial life exists in a computer, the computer milieu must define an artificial physics ... What is an artificial physics or physics-as-it-could-be? Without principled restrictions this question will not inform philosophy or physics, and will only lead to disputes over nothing more than matters of taste in computational architectures and science fiction.'' [Pattee 95a] (p.29).

The ad hoc feel of Tierra-like systems is a direct consequence of this lack of theoretical grounding. The unmanageable parameter space of many of them can also be attributed to this lack of direction. As a result of these weaknesses, even if interesting behaviours are observed in these systems, we are unlikely to be able to explain why (as was demonstrated in Sections 3.3 and 6.8). It may be that a model of self-replication and open-ended evolution is necessarily somewhat complex, but, even if this is so, the theoretical framework upon which it is built should prescribe the implementational details as much as is practically possible.

next up previous contents
Next: Predefined Organism Structure Up: Problems with Tierra-like Models Previous: Problems with Tierra-like Models
Tim Taylor