Mediamatic Magazine Vol. 9#1 Hein Masseling 1 Jan 1998


Artificial Evolution in an Artificial World

Until recently, one of the most frustrating problems for evolutional biologists was the fact that the process of evolution could not be directly perceived as a process. With the recent development of the discipline of Artificial Life (in brief, Alife), this seems to be changing.


Tierra - Program developed by Tom Ray

In its search for key characteristics of life, and constants in the emergence and the development of life forms, the profession has reached a new high point with the computer program Tierra, developed by the American ecologist Tom Ray.

Ray designed Tierra due to his frustration over the fact that, as an ecologist, he could only study the products of evolution, and not the process of evolution itself. Despite the apparent oversimplicity of the concept, Tierra has turned out to be a very attractive model for the study of evolutional processes as such. Programs such as Tierra have made it possible, for the first time, to rewind the tape, as Stephen Gould formulates it. With the help of such programs, we can, as it were, cause evolution to happen again. In spite of, or perhaps rather thanks to, this apparent simplicity, Tierra has attracted much attention, both from within the profession as well as from outside.


Artificial Life -

With Tierra, Ray answers in all respects to the adage of research into artificial life. Or, as Chris Langton, who gave the discipline a name with his article Artificial Life (1987), formulates it: To have a theory of the actual, it is necessary to understand the possible.

1 C.G. Langton, 1995, 'Editor's introduction', in: C.G. Langton (ed.), Artificial Life - An overview, Cambridge (Mass.) 1995, p.x.

Researchers believe that the study of forms of artificial life will enable them to gain a better insight into the way in which natural processes of life take place. Most of the research in the field of Artificial Life consists of designing computer models. The notable progress in the calculative capacity of computers has made this possible. The diversity is great. Apart from programs such as Tierra, which is centred around evolution, there are also for example: models of the development of colour patterns on shells, the morphological development of plants, cell fission and differentiation, and of how birds move in flocks. Besides models which use software as a medium, part of the research into artificial life is also carried out with hardware or wetware. An example of a project with hardware as a medium is the mobile robot, which has to find its way through a complex environment. The study by, among others, Gerald Joyce, is an example of the use of wetware as a medium. In a process called 'directed molecular evolution', RNA chains with the right catalytic properties are selected artificially. Apart from yielding a useful product, such research also provides an insight into the possible role of RNA in the origination of the first life forms on earth.
For the majority of researchers, this artificial life goes beyond being just a help in the study of natural life processes. In particular, Tom Ray holds the view that at least a number of artificial life forms, among which the organisms inhabiting Tierra, are genuinely alive. Although until recently this kind of thought perhaps seemed to belong in the realms of science fiction, the preserve of such writers as Isaac Asimov (one of his books is called I, Robot...), with the development of models such as Tierra, the origination of life forms with silicon as their central chemical element has become a real possibility.

Key concepts in the study of artificial life, that also play a conspicuous role in Tierra, are emergence, autonomy, and self-organization. While studies in the related discipline of artificial intelligence are characterized by models whose organization and development are imposed by the designer (contrary to a wide-spread belief, this also applies to neural networks), models designed for research into artificial life have no pre-conceived development or organization. The interactions between the components of the model generate a certain organization, a certain structure. Many properties of the models designed are emergent properties, properties which cannot be explicitly retraced to the starting conditions of the model. Visible consequences of the organization of the total system, according to Douglas Hofstadter

2 D.R. Hofstadter, Gödel, Escher, Bach - An Eternal Golden Braid, Basic Books 1979, p.357

For example, the phenomenon of parasitism which developed within Tierra was not demonstrably part of Ray's original set-up. In a model by Craig Reynolds, that describes the swarm behaviour of bird-like organisms, no explicit rule can be found to indicate that these organisms should form a swarm. Nevertheless, the birds will form a flock, which remains intact and regroups after it has split up to pass an obstacle (Langton, 1987). Chris Langton describes it as follows: Local behaviour supports global dynamics, which shapes local context, which affects local behaviour, which supports global dynamics, and so forth.

3 C.G. Langton, 'Artificial Life', in: M.A. Boden (ed.), The Philosophy of Artificial Life, Oxford 1996, p.68

To understand a model, we should not study the separate components, but rather, the interplay of these components. The organization is not top-down, but bottom-up.


Tierra - Program developed by Tom Ray

Tom Ray, ecologist and evolutional biologist at the University of Delaware, spends most of his time working in the tropical rainforests of Costa Rica. Although he was (and is still) inspired by the great diversity of life forms, he also became aware of the limitations of the study of existing diversity. We are shown the conditional end products of evolution, but for the greater part the evolutional process itself remains out of sight. With the development of Tierra, Ray made an effort to free himself of these limitations.

What he had in mind was a model, in which self-replicating computer programs would evolve according to Darwinian principles

4 For the history of origin of Tierra, the relevant chapter from Roger Lewin's Complexity is the main source of reference.

It would have to include a certain mutative pressure, which would result in a variation in programs, and then a process of selection would have to come about. The stakes were high: the replicating programs would have to qualify as living organisms (Ray, 1992, 1994). When Ray launched his idea in 1989, it was received with the necessary scepticism. Not only by fellow ecologists, but also by those such as Chris Langton, one of the pioneers of the recently emerged discipline of artificial life. Their objections were obvious. Firstly, it seemed unlikely that a computer program, after a random mutation, could still be an active program. Computer programs would simply seem too vulnerable. Secondly, should any active programs survive the mutations, so much the worse. In view of the danger of computer viruses, a model that could generate randomly changing computer programs simply seemed too threatening.
Ray took the objections to heart and, late in 1989, he began to program (in programming language C) what was to become Tierra. Early in 1990, the first version was ready. And what Ray had hoped actually happened. The programs in Tierra mutated and replicated themselves. Tierra, Spanish for earth, is in fact a virtual computer designed within the RAM memory of a real computer. In this way, Tierra is protected, so that if a potentially detrimental programme should develop, it cannot escape. Tierra is inhabited by populations of computer programs which have only one task: to reproduce themselves. As with natural organisms competing with each other for food and space, the programs in Tierra compete with each other for the CPU time they need to reproduce themselves, and for the memory space they have to reserve for their descendants.

Initially there was only one kind of program, a program of 80 machine instructions, the 'ancestor' program. This is the only program, of the many which would eventually populate Tierra, which was written by Ray himself. All the other programs have developed by mutation or recombination. Mutation in Tierra consists of random bit flips: random bits change from a 1 into a 0, or the other way around. There are two mutation regimes: one with a slight chance of mutation, comparable to the chance of mutation by cosmic background radiation, and the other with a better chance of mutation, depending on the number of self-replications. Recombination consists of an exchange of parts of programs between the various programs. Most mutations and recombinations (approximately 80%) will result in mutants and recombinants not being active. Computer programs remain vulnerable to random changes. To prevent Tierra from becoming full up, that is to say, all the memory space being taken, there is a so-called 'reaper function', the Grim Reaper, which makes space by clearing away the old and defective programs.

Very soon after a Tierra session has been started, it can be seen that, alongside the original program, new active programs are emerging. Thus, parasites will come into existence, which, being unable to reproduce themselves, will make use of the reproductive modules of other programs to multiply themselves. The host programs will suffer greatly, but will not become extinct. Indeed, the parasites cannot live without them. Parasites and host programs will develop an ecological equilibrium. Another kind of program which will emerge is the so-called hyperparasite. These programs are immune to certain parasites, and even manage to put parasitical programs to good use: the parasites contribute to the reproduction of the hyperparasites.

In this way, many sorts of programs come into existence. Some are shorter than the original programme of 80 instructions, and others are longer. Moreover, programs of a certain length are also varied in composition.


Living Organism -

The various kinds of emerging programs also create various forms of cohabitation. Besides parasitism, we also see the emergence of commensalism and programs which cohabit within a social context and cannot multiply independently. The parallels between Tierra and conditions in nature are certainly conspicuous. This suggestion is enhanced by the fact that, in some cases, a kind of 'punctuated equilibrium' can be observed. Periods in which no, or hardly any, new programs develop alternate with surge periods in which a great many different, new programs develop rapidly within a short time. For natural evolution, this phenomenon has been described by Stephen Gould and Nils Eldridge.

In Ray's opinion, the suggestion that the processes unfolding in Tierra have their pendant in nature is justified. According to him, the programs inhabiting Tierra must be regarded as living organisms. Even though these are digital organisms, and can therefore not be compared in detail to natural organisms, they meet Ray's criteria for calling something a living organism. Namely, they can reproduce themselves, and are capable of what he calls 'open ended evolution': there are no a priori restrictions with regard to the life forms which could develop. Of course, Ray's criteria are contestable, but on the other hand, no one has yet come up with an uncontestable definition of 'life'. The suggestion that these are living organisms remains very strong.
Ray now has ambitions to generate a kind of Cambrian explosion in Tierra. The Cambrian explosion took place at the end of the Cambrian Era, some 550 million years ago. Within a relatively short period of time, all the now known basic forms of multicellular organisms came into existence. Ray believes that a similar, sudden development of diversity and complexity would also be possible in Tierra. Up until now, the programs generated, digital organisms, are in fact very simple. According to Ray, they are comparable to the first life forms on earth (Ray, 1994a). In order to cause the kind of development that took place in the Cambrian Era, Tierra should now be put through a long-lasting session. The underlying thought is that long-term effects will occur, which remain invisible if a Tierra session is terminated after a short time (a day, a week). Ray has a network in mind, on which Tierra could be constantly active, and has his eye specifically on the Internet. Because Tierra is a virtual computer, it can move from one site on the network to another. In this set-up, Tierra would have to explore the net for the sites in which a (real) computer is active. The network version of Tierra could develop into a sort of reserve for digital organisms. Here, Ray's scientific work synchronizes with his work as an activist for the preservation of the tropical rainforest in Costa Rica (Ray, 1994b).

The importance of Tierra is manifold. Firstly, of course, there is its importance to biology. Ray emphasizes that we have to realize that Tierra generates digital organisms, which have a different constitution from natural ones, and possibly also a kind of biology of their own. In spite of this, he believes that Tierra could enable a better understanding of, for example, the Cambrian explosion, and the role of parasitism in evolution. Secondly, Tierra demonstrates the importance of such phenomena as self-organization and emergence. Based on a program of 80 instructions and a number of pre-conditions, such as the degree in which mutations occur, a whole gamut of new digital organisms and relationships between these organisms will develop. Nowhere in Tierra's set-up do we find the merest reference to parasitism or social organization. There is no fixed design, the development is spontaneous. //Local rules, global order:// this central concept in the research into artificial life is pre-eminently applicable to Tierra. Apart from purely scientific significance, Ray also envisages a direct practical use for the network version of Tierra. The programs generated in the digital reserve could actually turn out to be useful programs. That is the irony of the project. Initially criticized because of the potential dangers involved in the random generation of computer programs, now, based on a bottom-up organization, it could possibly become a generator of useful software.

translation OLIVIER / WYLIE


M.A. Boden (ed.), The Philosophy of Artificial Life Oxford 1996

J. Dibbel, Viruses are Good for You 1996

K. Gerbel en P. Weibel (eds.), Ars Electronica 93 - Genetische Kunst/Kunstliches Leben Wenen 1993

S.J. Gould, Wonderful Life - The Burgess Shale and the Nature of History London 1989

C.G. Langton (ed.), Artificial Life - An overview Cambridge (Mass.) 1995

R. Lewin, Complexity - Life at the Edge of Chaos'' New York 1992

T.S. Ray, A Proposal to Create Two Biodiversity Reserves: One Digital and One Organic / A Proposal to Consolidate and Stabilize the Rain Forest Reserves of the Sarapiqui Region of Costa Rica 1994