... `Zerox'^2.1

From the album Dirk Wears White Sox by Adam and the Ants ©1979 Do-It Records.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... position.^2.2

The major extension that produced the neo-Darwinist position was the synthesis of Darwin's original theory with Mendelian genetics.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... forms.^2.3

Darwin did not claim his theory accounted for the origin of life. However, more recently, a number of theories have been proposed which put the original transition from chemistry to biology in an evolutionary context (see, for example, [Eigen 71], [Maynard Smith & Szathmáry 95]).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... emphasis).^2.4

The enclosed quotation is from [Oyama 85].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Species^2.5

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Species^2.6

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Species^2.7

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

The ideas in this section are partially based upon a talk given by Richard Dawkins at the `Digital Biota 2' conference in Cambridge, England on 10th September 1998.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... impossible.^2.8

In fact, it is still not entirely clear whether the existence of species is a consequence of sexual reproduction ([Maynard Smith & Szathmáry 95], Chapter 9), but this does not affect the argument being made in this section.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... literature,^2.9

For discussion of this, see, for example, [Margulis 91], [Daida et al. 96].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... prokaryotic^2.10

Present-day bacteria are examples of prokaryotic cells.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... eukaryotic^2.11

All present-day animal, plant and fungi cells are eukaryotic. They are far more complex than prokaryotic cells, housing a wide array of specialised structures.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... flux.^2.12

Although the extent to which this is true varies greatly between different organisms, and between different parts within a single organism [Maynard Smith 86] (p.2).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... systems'^2.13

They invented the word `autopoiesis' from the Greek $\alpha\upsilon\tau\acute{o}\varsigma =$ self, and $\pi o\iota\epsilon\iota\nu =$ to produce.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... is:^2.14

Unfortunately the definition, at least in its English translation, is somewhat opaque.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... changed.''^2.15

Quoted from [Kelly 94] (p.447).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Earth.^2.16

Here I am only talking about models that assume a terrestrial origin of life.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... conditions)^2.17

One necessary condition is that the self-maintaining organisations were enclosed in some form of compartment, to enable natural selection to act between the compartments [Szathmáry & Demeter 87].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... speciation,^2.18

That is, speciation occuring as the result of the restriction (or total absence) of gene flow between two populations caused by some sort of physical barrier (e.g. a sea or mountain range). See, for example, [Maynard Smith 89].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... theory.^2.19

In the terminology of evolutionary algorithms, Wright's theory is a mechanism whereby an evolving population can avoid getting stuck in local (sub-optimal) fitness peaks. It involves the population being divided into a large number of small, partially isolated demes. Individuals in one deme may move to a higher fitness peak by chance (i.e. genetic drift), and subsequently spread throughout the whole population. See, for example, [Wright 31] and [Wright 82]; and also [Coyne et al. 97].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Thompson 17].^2.20

Quoted from [Nitecki 88] (p.14).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Darwin 72].^2.21

Quoted from [Gould 89] (p.257).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... complexity.^2.22

McShea also mentions a third possible dichotomy: differentiation versus configuration. The four types of complexity mentioned above are differentiational. Configurational complexity, on the other hand, is ``irregularity of arrangement of parts and interactions, independent of their differentiation'' [McShea 96] (p.480). McShea claims that configurational complexity has received little attention in the biological literature, so he does not consider it further in his work.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Bronowski 73].^2.23

Although Bronowski used the term in a somewhat broader context.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... applies.^2.24

Note that I am not talking about the distinction between gradualism and punctuated equilibria here; this view of evolution does allow for evolution in the periods between major transitions. However, the view holds that evolution during these periods will not necessarily be in the direction of increased complexity.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Gould 89].^2.25

Although Gould's view is also connected to the concept of punctuated equilibria, which is not specifically related to transitions between units of selection.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... next).^2.26

Salthe, in considering the hierarchical nature of life, argues that there is a clear distinction between the ecological hierarchy and the genealogical (evolutionary) hierarchy [Salthe 85]. The confusion in terminology under discussion here can also be seen as a confusion between entities in these two distinct (but related) hierarchies.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... process.^2.27

That is to say, S-lineages are the fundamental types of things which act as units of selection. McMullin is not arguing here about whether it is more appropriate to look at evolution from the point of view of the gene or the organism, for example, but claims that ``whichever of these viewpoints may be adopted, there will be a crucial distinction between actors and S-lineages, with only the latter being properly regarded as units of selection, or entities for whose benefit (Darwinian) adaptations may be said to exist'' [McMullin 95] (p.166, original emphasis).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Karakotsios^3.1

Quoted from [Kelly 94] (p.451).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Langton 86]),^3.2

Although others had used the term in earlier publications, such as [Overton 82] (in which it appears in the title without definition in the text, and in fact, by comparing the contents of the main text with that of the editor's note at the beginning, it appears that the title (including the term `artificial life') was chosen by the editor, Edmund C. Berkeley, rather than by Overton himself).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... subject^3.3

Although some would argue over the extent to which it is a coherent subject. See Sections 3.1.1-3.1.3.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... purposes,^3.4

For example, see Steven Rooke's online portfolio at http://www.concentric.net/~Srooke/ and the list of genetic art-related sites at http://gracco.irmkant.rm.cnr.it/luigi/alg_art.htm.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... p.372).^3.5

The concern with self-replication has also been a preoccupation for other artificial life researchers from von Neumann [von Neumann 66] to the present day, and can be related to the work of biologists such as Muller and Dawkins, discussed in Section 2.2.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... terms,^3.6

This, of course, is a big assumption. Those who maintain that life is of necessity a biochemical phenomenon would argue that concepts such as autopoiesis do not capture all of the relevant features of life. However, this eventually boils down to a linguistic debate, and scientific progress can only be made when we do adopt precise definitions, even if these are not universally agreed upon.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... example).^3.7

But note that artificial life does not restrict itself to using digital computers as the medium of implementation.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... axioms)^3.8

David Marr's [Marr 82] analysis of complex systems on three levels (the computational theory level, the representation and algorithm level, and the hardware implementation level), and his insistence that the theoretical distinction between these levels should be recognised when devising a model--especially the distinction between what is being computed (level 1) and how (level 2)--is also relevant for the formulation of scientific artificial life models.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... machines.^3.9

Von Neumann had difficulties in defining precisely what the term `complicated' meant. He said ``I am not thinking about how involved the object is, but how involved its purposive operations are. In this sense, an object is of the highest degree of complexity if it can do very difficult and involved things.'' [von Neumann 66].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... above.^3.10

The general constructive machine A of this design is often referred to as a `universal constructor'. However, this term should be used with caution; from the above description of the architecture it is clear that A can build any machine X that can be described upon a tape $\mathbf{\phi(X)}$. For cellular automaton models it can be proved that there are some configurations that the universal constructor cannot build (e.g. [Moore 62], [Myhill 63]). These are referred to as `Garden of Eden' configurations, as the only way they may exist is if they are programmed in as the initial state of the space at time zero.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[von Neumann 66].^3.11

Quoted from [Pattee 88] (p.69).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... self-reproduction^3.12

`Trivial' self-reproduction, at least in the sense being used by Langton, occurs when reproduction of a particular sort of configuration happens purely due to the rules of the system rather than to anything explicitly encoded in the configuration itself. For example, a CA with a transition rule such as ``if this cell is empty (in the quiescent state) and one of its neighbouring cells is in state $\cal A$, then change the state of this cell to $\cal A$'' is an environment in which the state $\cal A$ trivially self-replicates. I suspect that Langton's work was the seed for the preoccupation of many artificial life researchers over the past decade with the `problem of trivial self-reproduction'. However, I believe that this is not the most relevant distinction with respect to evolvability, as I will argue in Chapter 7.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... 1990s.^3.13

Although Eigen and Schuster's work in the 1970s should also be mentioned. In the context of prebiotic evolution, they discussed ways in which the amount of selectively maintainable information in a system could be increased beyond the capacity of an individual replicator (see, for example, [Eigen & Schuster 77]). They introduced the `hypercycle' (a cyclically catalytic group of replicators) as an functional organisation which could achieve this.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... `Coreworld'^3.14

Specific configurations of Coreworld were named Venus I, Venus II, and Luna.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Dewdney 84].^3.15

The ancestry of this approach can be traced back even earlier (e.g. [Bratley & Millo 72], [Burger et al. 80]), although these examples were presented in a fairly informal manner.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... universality.^3.16

A proof that the language has this property is presented in [Maley 94].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... template-driven^3.17

With the template-driven branching scheme, a branching instruction is followed by a template (a sequence of bits). The operating system will search for the nearest matching template in the rest of the code and move the instruction pointer to that position.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... problem,^3.18

That is, some programs might never terminate.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... set^3.19

Examples include issuing a jmp instruction with a template pattern for which no match can be found, and attempting to write to a memory address for which the program does not have write access.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... sin.''^3.20

From an online report on Tierra, available at http://www.hip.atr.co.jp/~ray/tierra/netreport/
netreport.html#Philosophy.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... symbiogenesis^3.21

According to Barricelli, the symbiogenesis theory claims that if genes are to evolve into ``relatively higher forms of life'', they must only be able to reproduce through a symbiotic relationship with other genes [Barricelli 57] (p.145). A group of genes that collectively reproduces in this way (a `symbioorganism') can be considered a special case of a hypercycle [Eigen & Schuster 77], in which each component in the group is absolutely required for the reproduction of the next component, rather than just acting as a catalyst.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... community,^3.22

His papers contained many ideas on subjects which have since become popular research areas, such as the use of real DNA for performing computations ([Barricelli 63] pp.121-122).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Bedau 98a]).^3.23

But see [Mayer & Rasmussen 98] for a recent study in which emergent hierarchical structure has been successfully simulated.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[McMullin & Varela 97].^3.24

Interestingly, Barricelli also predicted that his symbioorganisms would need to develop means of controlling their local environment (such as a membrane) if they were to evolve past a certain level of complexity [Barricelli 63] (pp.122-124).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... coexisted.^3.25

Another is Lindgren's model of a population of individuals playing a variation of the iterated Prisoner's Dilemma [Lindgren 91].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Buss.^3.26

Rasmussen et al. also discuss the relationship between these and other universal formalisms [Rasmussen et al. 91] (pp.243-244).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... structures'';^3.27

This is similar to von Neumann's conclusion that ``complication is degenerative below a certain minimum level'' [von Neumann 49] (p.482).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... principles.^3.28

More precisely, there may be some general principles concerning parasitism that Tierra shares with Nature, but the point is that no-one has explicitly stated what these might be, and Tierra was therefore not designed to be a particularly good test of any such principles.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... sort.^3.29

Maynard Smith and Szathmáry discuss ways in which the evolutionary potential of hypercycles can be enhanced through compartmentation, and suggest that this might eventually lead to the evolution of a genetic system [Maynard Smith & Szathmáry 95].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Cosmos,^4.1

`Cosmos' is an acronym for COmpetitive Self-replicating Multicellular Organisms in Software.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... complexity^4.2

Primarily, hierarchical object complexity (see Section 2.3.2). In other words, many of the design features of Cosmos were intended to promote the evolution of multicellular organisms from unicellular ones.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... squares.^4.3

The system has been designed to deal with arbitrary n-dimensional environments, but the current implementation requires some minor revisions to allow this.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... execute.^4.4

As I am usually referring to the contents of the Genome, rather than to the structure itself, when I use the term `genome', I will use the standard typeface from now on.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... instruction),^4.5

See Section A.2 for an explanation of the instruction set.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... genome^4.6

Or on eligible InfoStrings in the Received Message Store. See Section 4.3.7 for details.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... string.^4.7

The search begins at the current Read position on the genome, and proceeds outwards in both directions simultaneously.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... genome,^4.8

Or on eligible InfoStrings in the Received Message Store. See Section 4.3.7 for details.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... tokens,^4.9

In this situation, the choice of which cells to kill is actually stochastic, with the level of a cell's Energy Token Store determining the probability of its being killed.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... area,^4.10

The only restriction is that there is a maximum length to which these strings are allowed to grow, defined by the global parameter info_string_size_limit. This is to prevent the situation in which a program evolves which gets stuck in an infinite loop writing to the Nucleus Working Memory, eventually using up all of the memory in the system.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...)^4.11

The function of these latter two structures is explained in Section 4.3.7.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... it.^4.12

The type of the cell's genome cannot be directly altered, and is passed on to children when the cell splits or divides. However, it is subject to mutation like any other part of the cell (see Section 4.5.7). Therefore, it is possible for organisms with different genome types to emerge in the system.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Organism^4.13

A capital `O' is used here to emphasise that we are talking about the specific implementation details. However, as the Organism class encapsulates the functionality of an organism, the two terms can be used interchangeably. Therefore, in the rest of the document I shall just use the term organism (with a small `o'). The same applies for cells and the Cell class.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... REPLiCa,^4.14

`REPLiCa' is an acronym for Robust Evolvable Programming Language for Cosmos.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... promoter.^4.15

Of course, when programs are evolving, especially when we are considering parallel programs, there may be more than one promoter in the Promoter Store at one time. However, here we are describing how a human might design a program that performs a jump--evolution would probably go about designing a program in a very different way.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... amount.^4.16

To be precise, the magnitude of the increment is kI^p, where I is the current intensity, and k and p are constants defined by the global parameters envinfostring_decay_constant and envinfostring_decay_power respectively.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... limits.^4.17

To be precise, the flaw rate can change by plus or minus n parts per thousand, where n is determined by the parameter flaw_period_max_change_per_thou.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... genome.^4.18

Tierra features both mutations and flaws (although the mechanisms for flaws is somewhat different) but in subsequent work by Chris Adami and Titus Brown with their Avida system the authors suggested that flaws played only a minor role in evolution compared to mutations [Adami & Brown 94]. Informal observations from preliminary runs of Cosmos suggested that flaws in the execution of instructions significantly increase the rate at which useful mutants are produced.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... rate^5.1

Precisely, what matters is the rate at which selectively significant errors occur, i.e. errors which exclude the offspring from that particular S-lineage.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... rate,^5.2

Actually, we looked at the inverse of flaw rate, the flaw period. This is the expected number of successful instruction executions in a program between successive flaws.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[^5.3

Note we are looking at genotypes here, i.e. groups of identical programs, rather than individual programs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Bedau & Brown 97]^5.4

The rewording at the indicated places in this extract is to make it consistent with later changes in the terminology that Bedau et al. use to describe these techniques.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

Again, this formula has been changed from the original text to make the notation more consistent. In the original paper, tex2html_wrap_inline$A_new(t)$ is defined as displaymathA_new(t) = 1D(t)_i,a_0a_i(t)a_1a_i(t) (i.e. as above, but divided by tex2html_wrap_inline$D(t)$). This is what will be referred to here as Mean New Activity, tex2html_wrap_inline$A_new(t)$.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Bedau et al. 98]^5.5

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... unicellular.)^5.6

Therefore, at any given time, the number of cells in the neutral shadow is always the same as in the standard run, although the number of organisms may be lower if the standard run contains some multicellular (parallel) programs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... bits.^5.7

There were in fact a handful (7) of programs of different lengths, but these are not recorded in Figure 5.2, because of the data pruning methods described in Section 5.1.1.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... length.^5.8

Again, the data pruning techniques discussed in Section 5.1.1 mean that information is not displayed in these graphs for genotypes that are represented by only a very few individual programs. Such individuals typically account for about 1% of the total population.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...$\bar{A}_{new}$.^5.9

As explained earlier, Bedau calls this measure simply the new activity, A_new, rather than mean new activity. The name has been changed here to make it consistent with the naming of other measures.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... bounded^5.10

The definition of (un)boundedness given in [Bedau et al. 98] is: The function f(t) is unbounded iff

where sup(.) is the supremum function.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... positive^5.11

Similarly, the definition of a positive function is given in [Bedau et al. 98] as follows: The function f(t) is positive iff

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Chance^6.1

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Chance^6.2

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Chance^6.3

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

This section is based upon a previously published paper, [Taylor & Hallam 98].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

....^6.4

These experiments were run on slower machines than most of the others, which is why the size of the runs (i.e. the maximum number of cells allowed in the population and the run duration) is smaller than most of the others. The other differences in parameter values came about because these runs were actually conducted before most of the others, and the default values of a few of the parameters were changed in the intervening time. The differences are only minor, and it is not expected that they affect the applicability of the results to the other experiments reported in this chapter.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... compared.^6.5

In the following, the pairs of run results displayed in Figure 6.1 and in Figure 6.8 were generally chosen because they illustrate noticeably different results.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... allocated.^6.6

Recall from Chapter 4 that under this scheme, the parameter number_of_energy_tokens_per_ grid_pos_per_sweep determines the mean number of energy tokens that should be distributed to each grid position.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... instruction^6.7

Or sometimes a migrate instruction, which is equivalent to move for a single-celled program.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... readable.^6.8

In contrast, the mean number of faithful reproductions of programs of the ancestor genotype 348AAAA were similar in the two sets of experiments. In the present experiment, the figure was 1.077 (standard deviation 0.045), and in the standard model re-runs, it was 0.993 (standard deviation 0.037).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... energy.^6.9

This section was included for the practical reason that when a run is first inoculated, several hundred time slices pass before the females emit their first messages. During this time, collecting energy is the only useful task that the males can do.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... speciation^6.10

I am using the term `speciation' very loosely here. For a start, as the programs are reproducing asexually, it is debatable whether the concept of species is applicable at all. Even if we do allow the term to be used, it might be argued that the programs in the different length groups are fundamentally very similar, and are more appropriately viewed as different varieties of the same species. However, remember that they did not only differ in their lengths, but also sometimes in their ability to move, and by the amount of energy they collected.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... issue,^6.11

They were not designed with this in mind. A proper test would need to be based upon an explicit set of assumptions, and would almost certainly require sexually reproducing programs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... environment.^6.12

This theory, which he named `cannibalistic altruism', was discussed during a talk by Wolpert at the Royal Museum of Scotland, Edinburgh, on 20 February 1997.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... reserves),^6.13

But note that in this case the prey organisms would only be a resource of energy, not of matter. This distinction is discussed in more detail in the next chapter (e.g. Section 7.1.4), along with possible evolutionary consequences arising from it.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... model).^7.1

Ray himself recognises these difficulties, but is more optimistic that they can be overcome [Ray 91] (p.399).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... (p.36).^7.2

Although he also stresses that ``some epistemic principles must restrict physics-as-it-could-be if it is to be any more than computer games'' [Pattee 95a].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... organisms).^7.3

By `environment' I mean the shared space in which (at least some aspects of) all of the organisms exist, which I will refer to as the `arena of competition' (see Section 7.2.3), together with the universal `laws of physics' of this space which determine how entities within it act and interact.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... information.^7.4

For example, a virus requires information contained in its host's genome in order to reproduce. This information is more than the matter from which the host's DNA is constructed; it involves a particular ordering of matter.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... matter.^7.5

This is also the case in von Neumann's cellular automata models. In contrast, structures in his kinematic model, as in models proposed by various others (such as Myhill, and Holland's $\alpha$-Universes), are constructed from atomic units of matter (see Chapter 3).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Rosen 59]).^7.6

Although Löfgren has shown that complete self-reproductive functions can exist through axiomatisation [Löfgren 68].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... division).^7.7

The important distinction between these two types of system in the present context is that logical systems manipulate states, whereas material systems manipulate matter. Reproduction in material systems, in contrast to logical systems, therefore requires an object to collect the `raw materials' to build a copy of itself.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... distinction.^7.8

Any attempt to classify a reproducer as trivial or non-trivial according to the explicit-implicit distinction is bound to be somewhat arbitrary, because we are generally considering examples of second-level reproduction rather than complete self-reproduction (Section 7.2.1). Therefore, the surroundings will always play some role in bringing about the reproduction of the object in question. There is considerable irony in much of the recent work relating to this issue. For example, Langton not only ignores von Neumann's actual solution to the `problem' of trivial reproduction, but furthermore says that ``von Neumann's work suggests an appropriate criterion ... : the configuration must treat its stored information in ... two different manners ... : interpreted, as instructions to be executed ..., and uninterpreted, as data to be copied'' [Langton 84] (p.137). As far as von Neumann's actual analysis of the subject is concerned, this distinction between interpreted and uninterpreted is only important insofar as it gives the reproducing automaton the capacity to support inheritable mutations, potentially leading to the evolution of more complicated and more efficient machines.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...C,^7.9

This is not an inherent property of the architecture per se, but von Neumann's analysis of evolvability did assume a `compositional' structure in the language of the tape descriptions (see Section 3.2.1).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... (p.574).^7.10

The passage to which I am referring is ambiguous. Ibáñez and colleagues talk about ``the possibility of using our self-inspection based reproductive scheme as a basis for artificial evolution. This attempt is not new; it has quite successfully been applied in other environments, and the most paradigmatic of them is probably Tierra'' [Ibáñez et al. 95] (p.574). I take this to imply that they also regard Tierra as an example of a self-inspection based scheme.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... design.^7.11

Indeed, for organisms in any kind of evolving system, the notion of a phenotype fundamentally involves behaviour, in the form of interaction with the (biotic and abiotic) environment.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Replication.^7.12

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

Illustration © Genentech, Inc. (http://www.gene.com/ae/AB/GG/). Reprinted with permission.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... cell.^7.13

For eukaryotic cells we would have to include the DNA contained in mitochondria and, in the case of plants, chloroplasts, as well as that contained in the nucleus. There is some evidence that particular features of cells can occasionally be inherited without apparently requiring a change in DNA sequence (e.g. abnormal patterns of cilia on the surface of particular protozoans), but such exceptions are very infrequent [Maynard Smith 86] (pp.24-25).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... genome.^7.14

To be precise, the initial interpretation machinery is derived from information contained in the parent cell's genome, in the same way as in von Neumann's self-reproducing automata.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...A.^7.15

Note however that the laws of physics and chemistry still play a vital role in the construction of this machinery. For example, the genome only encodes information about the linear sequence of amino acids in any protein that it can construct. As the ribosomes build the protein based upon this information, the growing protein folds into a three-dimensional shape due to attractive and repulsive forces between its subsections. Thus the primary (linear) structure of the protein is encoded in the genome, but its secondary and higher structure is determined by the laws of physics (and by the manner in which it is constructed).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... ingredient'.^7.16

His solution was to require that elements could only reproduce in symbiotic association with other elements (see Section 3.2.2). While this may indeed be an important aspect of the `missing ingredient', it is extremely doubtful that it is the only important aspect.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... GA.^7.17

The same applies to similar artificial life platforms with two-dimensional environments, such as Avida [Adami & Brown 94].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... addresses.^7.18

Recall that the allocation of memory for reproduction is performed by the external operating system.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Goals^7.19

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Goals^7.20

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Goals^7.21

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

The phrase `Beyond Digital Naturalism' is borrowed from [Fontana et al. 94].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... addressed,^7.22

Even in the situation where some high-level `emergent' phenomenon is under investigation, and the model is expressed in terms of low-level entities and interactions, it may still be viewed as a potential explanation in the sense that it could demonstrate that no additional entities or interactions are required to produce the phenomenon. David Chalmers refers to such accounts as `mystery removing' explanations [Chalmers 96].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... p.115).^7.23

Note that this description closely resembles Löfgren's definition of self-reproduction given in Section 7.2.1, if we substitute $\mathcal{A}$ in those definitions for $\mathcal{P}$ above, d for $\mathcal{Q}$, and $\mathcal{S}$ for $\mathcal{E}_{j}$. Sticking to Waddington's labels, the difference is that Löfgren's definition results in $\mathcal{P}$ being reproduced, whereas Waddington's results in $\mathcal{Q}$ being reproduced. If we assume that $\mathcal{Q}$ is the only aspect of $\mathcal{P}$ that affects the reproduction process, then the end result is the same.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... (p.120).^7.24

In the original paper, the final word of this paragraph appears as $\mathcal{Q}^{\prime}$s rather than $\mathcal{Q}^{*}$s. This is fairly clearly a typographical error.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...[Holland 76],^7.25

Holland refers to strong and weak bonds in his model, but the weak bond effectively denotes the absence of a bond in the normal sense of the word.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... program.^9.1

Actiview is a program developed by Emile Snyder and Mark Bedau at Reed College in the USA, to produce various summary statistics and graphs depicting the evolutionary activity of the run. These are described in Section 5.1.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.