Life as an Abstract Phenomenon:
Is Artificial Life Possible?

Claus Emmeche

[a version of this paper appeared as pp. 466-474 in: Francisco J. Varela and Paul Bourgine (eds.):Toward a Practice of Autonomous Systems. Proceedings of the First European Conference on Artificial Life. The MIT Press, Cambridge, Mass., 1992. ]

Abstract

Is life a property of the material structure of a living system or an abstract form of organization that can be realized in other media; artificial as well as natural? One version of the Artificial Life research programme presumes, that one can separate the logical form of an organism from its material basis of construction, and that its capacity to live and reproduce is a property of the form, not the matter (Langton 1989). This seems to oppose the notion of a cell within contemporary molecular biology, according to which "form" and "matter" do not represent separate realms. The information in a living cell is intimately bound to the properties of the material substrate. This condition may represent a restriction on the validity of formal theories of life.

1 Introduction

The new field of research called "Artificial Life" (henceforth referred to as AL) is under establishment as a respectable domain of scientific inquiry. Like "Artificial Intelligence," it brings together people from a lot of disciplines and provokes new questions and approaches to the study of complex phenomena. The purpose of this paper is to discuss one basic notion of this research programme - the claim of medium-independence of life from any specific material substrate - and by implication, to discuss the models that are claimed to realize genuine lifelike properties.

The strong version of Artificial Intelligence is based on the assumption that cognitive functions are computational and thus in principle independent of the specific material substrate supporting computational processes. In the same way, the proposed AL research programme seems to presume that one can separate the logical form of an organism from its material base, and that its "aliveness," its capacity to live and reproduce, is a property of the form, not the matter (Langton 1989). Therefore it is possible to synthesize life (genuine living behaviour) on the basis of computational principles. The claim that life, or lifelike behaviour is possible to realize in the computational medium, I will (cf. Sober 1991) call the "strong version" of AL, as opposed to a weaker version that only claims to model aspects of living behaviour by computer simulation techniques. (Cybernetic/robotic approaches to the construction of lifelike devices with sensors and effectors (animats) or biochemical approaches to prebiotic life through in vitro experiments will not be considered though sometimes included in the broad AL research programme). Strong version AL seems to contradict an immediate intuition of molecular biologists, that "form" and "matter" do not represent separate realms (at least on the intracellular level), and that information is intimately bound to the properties of the material substrate. In this paper, I will explicate this intuition by an example to give an idea of the problems facing computational attempts to synthesize life and give formal descriptions of living behaviour. In a second paper I will discuss the semiotic problem of describing the type of sign-relations that is often presumed to exist within a cell seen through the perspective of molecular biology, and the sign-processes that characterizes the relation between an observer, an organism and a model of an organism.

2 Artificial Life as a Contribution to Theoretical Biology

AL research may evoke a new dialogue between computer scientists and experimentalists about a set of related questions:

1. Can we construct universal theories of life as a phenomenon independent of the specific media that life on Earth is made of, as described by biochemistry and molecular biology? (I think we can, and I think that the theory of Maturana & Varela (1980) is one such example).

2. Are there any necessary relations between the material components and the formal processual structures that characterise living systems? (I'll guess that the physical constraints on form discussed by D'Arcy Thompson (1942) and Vogel (1987) are a candidate for one set of such necessary relations).

3. Is the computational approach to biology - and the idea of synthesizing not only molecules, but whole organisms - counterintuitive to biologists only because of a prejudice that experimental intervention in Nature by the standard methods of chemistry and physiology is the only way to assess the structure of living reality?

4. In what sense may life be a computational, medium-independent phenomenon? Is life a multi-media-realizable phenomenon because it is intrinsically computational, or because the form of movement of any specific natural phenomenon (that can be described by an algorithm) can be realized by a computational set-up.

5. What kind of concept (or set of related concepts) of computation is presupposed in ALife discussions?

Part of the historical background of these questions is the way one has conceived of the relation between life and inorganic nature, and by implication, biology and physics. Traditionally, biology is seen as an empirical science concerned with local and contingent phenomena formed by natural selection, and often too complex for detailed basic explanation. Physics is seen as a science of universal processes, ranging from the smallest particles to the evolution of cosmos. This picture is highly simplified. On the one hand, some types of physical processes only occur very rarely at highly specific circumstances, and are thus equally local. On the other hand, life, or lifelike processes, may be a much more global phenomenon than the picture of present biology can tell. Considering the size of the universe, we should expect on probabilistic grounds that other forms of life have evolved on other planets, yet too far away to give access to empirical investigation (Papagiannis 1985).

The specific earthly forms of life we know about are the result of a vast succession of historically frozen accidents constrained by some general principles of biological evolution and morphogenesis, which in turn depend on the mechanisms of heredity and biochemistry. Unfortunately, neo- and postdarwinian evolutionary theory has been unable to give any satisfying account of the nature of developmental and evolutionary constraints (Webster and Goodwin 1982, Ho and Fox 1988). Thus, we cannot by the present theory of biology distinguish between possible and impossible forms of life (but for a small section of the biological possibility space very close to actual forms of life, such as the sets of lethal mutants). The genetic code, for example (specifying the transcription of DNA sequences into protein sequences), might have been differently composed. How ever, its presumed arbitrarity might not be due exclusively to historically frozen accidents and various external and (with respect to the living system) contingent causes; rather, some general biochemical constraints on possible forms of protein synthesis and regulation not yet understood may have acted lawfully in the process of creation of this specific code, disallowing the formation of other code tables (Crick 1968, Orgel 1968). Thus, theore tical biology could benefit from new approaches to its subject matter in order to make progress as to the general aspects of living systems.

3 Definition of Artificial Life

In this context, it is stimulating that Chris Langton proposes to characterise the field of AL as the study of "life as it could be" - so that other forms of life than those actually evolved on earth until present fall within the proper realm of bio-research. However, this extension of theoretical biology may pose some problems.

The claim of empirical extension: the biology of the possible

The first point in Langton's definition of the subject of AL goes like this: "Artificial Life is the study of man-made systems that exhibit behaviors characteristic of natural living systems. It complements the traditional biological sciences concerned with the analysis of living organisms by attempting to synthesize lifelike behaviors within computers and other artificial media. By extending the empirical foundation upon which biology is based beyond the carbon-chain life that has evolved on Earth, Artificial Life can con tribute to theoretical biology by locating life-as-we-know-it within the larger picture of life-as-it-could-be." (Langton 1989, p.1) This is a step towards a more comprehensive view of biology, although one can discuss if present biology is merely analytic. Biologists have recognized the synthetic organizational complexity of their objects ever since the word organization in the late 18th century was used by French and German naturalists to emphasise that the distinction between living, organized bodies and brute, inanimate ones was more essential than the earlier division of Nature into the kingdoms of animals, plants, and minerals.

It is not clear what kind of criteria can be used to evaluate theories and models of "life as it could be" in a non-trivial subset of possible worlds (a general science of life in natural and artificial systems should be delimited from mere game construction and computer animation for science fiction movies). As anything is possible in pure imagination, AL has to take recourse to the earthly biology to see if a particular instance of an artificially constructed model of life has a plausible behaviour. Physically interesting models of reality should not represent violations of known physical laws (which we believe to be both universal and well defined in terms of contemporary physics). So biologically interesting models of some aspect of a living process should violate neither physical laws nor what is conceived presently to be a possible behaviour of a living system. The problem is, that the latter is ill-defined, and the computational paradigm of AL does not by itself provide a better description of the universal phenomena of biology.

Methodology and emergence

With respect the computational approach of AL and the crucial property of emergent behaviour in AL models, Langton rightly dissociates the AL research programme from classical research in Artificial Intelligence, where models are built "top-down" (general specifications of behaviour are recursively decomposed into simple algorithms), inference is sequential, and the global control of behaviour allows no emergence of really new patterns of behaviour. The computational paradigm of connectionism within cognitive science has an approach to modelling complex behaviour which is essentially the same as in AL, at least with res pect to parallelism, "bottom-up" specification (recursive rules apply to local structures only) and emergence. However, the concept of emergence is ambiguous, and neither neural nets nor cellular automata models of dynamic systems may prove to be emergent under more strict definitions of the term (Cariani 1991).

The claim of medium-independence of life

The third point in Langton's definition of AL is more troublesome. It is true that bottom-up models of emergent properties can be said to synthesize lifelike behaviours within a computer, but Langton intensifies this view and postulates that, in fact, by these means one can realize lifelike properties. Thus we can have genuine life in artificial systems (p.32, ibid.). For example, in a model of flocking behaviour of birds by Craig Reynolds, the individual simulated birds (or "boids") are not real birds, but their emergent behaviour in the model is for Langton as genuine and real as the behaviour of their natural counterparts. Thus Langton does not claim that computers themselves will be alive, but that the informational universes they can support eventually can be alive (p.39). Accord ingly, the "artificial" in AL refers only to the component parts, not the emergent processes: "If the component parts are implemented correctly, the processes they support are genuine - every bit as genuine as the natural processes they imitate." (p.33)

The reason for this claim can be seen as a somewhat Platonic conception of life, according to which "...the dynamic processes that constitute life - in whatever material bases they might occur - must share certain universal features - features that allow us to recognize life by its dynamic form alone, without reference to its matter. This general phenomenon of life - life writ-large across all possible material substrates - is the true subject matter of biology." (p.2, ibid.) And thus "...the principal assumption made in Artificial Life is that the `logical form' of an organism can be separated from its material basis of construction, and that `aliveness' will be found to be a property of the former, not the latter." (p.11, ibid.)

Thus, while the simulated birds have no cohesive physical structure but only exist as information structures within a computer (as Langton admits), the phenomenon of flocking birds and the flocking of simulated birds should be two instan ces of the same phenomenon: flocking. To a hunter or an orni tho logist, or even a bird, it may seem a little strange. To a logician or a computer scientist used to hand ling abstract symbolic structures, it is quite obvious. The problem of theoretical biology, I think, is to deal with the living material structures of organisms and their inherent "logic" at the same time.

But even if that can be done, our models of the logic of living systems are not necessarily instances of the true logic inherent in the very systems themselves. First, as data are theory-ladden, so are models; a scientific model do not represent nature in any direct or iconographic way; what it represents is a theory of nature (in this instance, a computational theory with many unspecified presuppositions). Furthermore, the logic of life is a many level affair, spanning in time and space from molecular to ecological and evolutionary relationships. The physical/chemical causal processes within an organism are of a different kind, described by a different set of theories, than the processes within a computer running some programme. Their functions may be similar on some level of description, but the inherent logic of the processes, on the physical/chemical level (and probably on higher levels as well), is likely to be different.

4 Why "Strong AL" is Biologically Counterintuitive

Why does strong AL seem to be so counterintuitive from a biological point of view? Hardly any biologists can disagree with Langton when he says that "Neither nucleotides nor amino acids nor any other carbon-chain molecule is alive - yet put them together in the right way, and the dynamic behavior that emerges out of their interactions is what we call life." (p.41) But we also learn that "Life is a property of form, not matter, a result of the organization of matter rather than something that inheres in the matter itself." (ibid.) Though this is a purely philosophical claim rather than a scientific proposition, it appears to be incompatible with an intuition nourished by the current paradigm of molecular biology. This intuition says that real life is both form and matter, and that the proper object of life science is to study both aspects and their dynamic interdependence.

All living organisms on Earth happen to be made up of cells. A cell is analyzed in terms of its materials, structures, and processes, and can be seen as a bag of chemicals (each of which has its own form), and as a complex self-organizing structure containing a sequence of digital non-complete self-description (Pattee 1977) and embodying a web of structural informational relationships in time and space (Løvtrup 1981, Alberts et al. 1989). In a material world, life cannot be pure form. Molecular biology is often accused of only taking interest in the material components of the cell. The obvious answer is that, on the intracellular level, molecular biologists cannot separate form and matter, because the behaviour of the cell and its constituent molecules depends crucially on the form of the individual macromolecules (proteins, ribosomes, messenger-RNAs, etc.).

The "form," or biological information, of an organism is bound to the properties of the material substrate to such an extent that attempts such as von Neumann's to give "the logical form of the natural self-reproduction problem" (see Langton 1989, p.13) will encounter severe problems. This interdependence of form and matter can be illustrated by an example:

In the bacterium Escherichia coli, the synthesis of the aminoacid tryptophan from chorismatic acid occurs in three steps, each of which is cata lysed by a specific enzyme. These enzymes are themselves synthesized from aminoacids by the ribosomes. This process of polymerizing aminoacids into protein chains is termed translation, because the sequence of the 20 different aminoacids in proteins is specified by a DNA sequence (gene) on the chromosome. In the case of the three tryptophan synthesis enzymes, the DNA regions specifying their aminoacid sequences are grouped together in a common control unit, the tryptophan operon. For the enzymes to be synthesized, the DNA regions have to be transcribed into mRNA chains. This is done by RNA polymerase which starts transcribing from the tryptophan promoter in the first part of the operon and terminates after all the genes have been transcribed. Since transcription and translation are very energy consuming processes, the regulation of these processes in the response to the need for the enzymes is highly advantageous to the cell. A sophisticated system for the control of transcription of the tryptophan operon is found in E.coli. The attenuation control system involves both the protein coding function and the physical nature of the RNA chain. The mechanism is thoroughly studied and shall only be briefly described here (see Landick and Yanofsky 1987).

Attenuation control is based on the tight coupling of transcription and translation in prokaryotes. The RNA leader region contains a ribosome binding site followed by a short gene encoding 13 aminoacids, a stop codon, and a non-coding region including the attenuator sequence. The latter consists of a transcription terminator, i.e., a short C-G-rich palindrome (that forms a "hairpin" secondary structure by base pairing) followed by eight U residues. The region can exist in alternative base-paired conformations (fig.1); only one of these allows the formation of the terminator. A substantial body of evidence has established a model according to which the tryptophan level determines the ability of the ribosome to proceed through the leader region, and this in turn controls the formation of mRNA secondary structure: a) When tryptophan is present, the ribosome can continue along the leader (synthesizing the leader peptide) to the stop codon between region 1 and 2. The ribosome now extends over mRNA region 2, preventing it from base pairing. Thus region 3 can pair with region 4, generating the terminator hairpin, that causes the RNA polymerase to terminate (it dissociates the RNA from the DNA so that no genes are transcribed). b) When the cells are starved for tryptophan, the ribosome stalls (because of deficiency of charged Trp-tRNA) when it encounters the two tryptophan codons (i.e., the RNA code specifying the insertion of tryptophan in the growing chain) in immediate succession within region 1 of the leader. By stalling, the ribosome sequesters region 1, so that regions 2 and 3 will base pair before region 4 has been transcribed by the polymerase. Thus, mRNA region 4 remains single-stranded, no termination hairpin is formed, and RNA polymerase will read through the attenuator and transcribe the remainder of the operon. In summary, the attenuation control responds directly to the need of the cell for tryptophan in protein synthesis.

The right timing of the events is essential. Evidence suggests that after initiation at the promoter, the RNA polymerase proceeds to a position (after the leader peptide sequence) where it pauses; this may be necessary to give time for ribosomes to bind to the ribosome binding site of the leader transcript before regions 3 and 4 are synthesized. The 1:2 mRNA secondary structure may function as the transcription pause signal.

The example shows several things. The "linguistic mode" of the cell (i.e., the instructions in the DNA) and the "dynamic mode" (the workings of the machinery) are so closely connected in the prokaryote cell that the "logic" that describes the behaviour of the cell is time-dependent and for some part implicitly represented in the machinery that reads the instructions (pace Pattee 1977). The argument of von Neumann that it is possible to abstract the logical form of some feature of an organism's performance (such as self-reproduction) runs into difficulties when one attempts to "realize" this form in another medium. To describe logical aspects of biological systems in order to formalize them may be complicated, though possible (and if possible, it may only be trivial aspects of self-reproduction that are formalized, as argued by Kampis and Csányi (1987)). But attempts to realize these formal descriptions in a second medium may be much harder if the implementation of the formal description does not take into account the interdependence of form and matter at the cellular level. What is realized is our formal theory, not a duplication of the original living system.


Fig.1. The alternative base-paired conformations of the mRNA trp leader region. The four regions that can base pair are shown. Region 1 contains the last five codons of the leader peptide. Region 4 and the last part of region 3 contain the attenuator sequence. Left: The 1:2 and 3:4 secondary structures. The pairing of region 3 and 4 generates the terminator (the 3:4 hairpin followed by the U residues). Right: The 2:3 secondary structure allows no formation of terminator hairpin. (After Lewin, 1983, and Landick and Yanofsky, 1987).

One might object to the example above, arguing that it only shows that AL models must be adapted to another level of detail to encompass the mechanisms described. One could in principle make a cellular automaton model of the interactions of the attenuation control. But still, this would be a simulation; the model might be formally similar to the operon of the E.coli but would have no physical or causal similarity to the real system. It is often claimed (e.g., Burks 1975) that von Neumann's 1948 kinematic model of self-reproduction (the components of which are a constructor, a duplicator, a controller, and a written instruction) was verified by the discovery of DNA struc ture and functioning (the analogous components being the ribosomes, the poly merases, the repressor + derepressor control molecules, and the DNA). But the analogy is in no way complete, because the functions of the biological components are not separated in the real system and depend on the specific physical structure of the constituents. The dynamic information (Burks and Farmer 1984) stored in the 3D structure of DNA and in the rest of the cell's components is not represented by the formal model of self-reproduction (a central instance of the symbol-matter problem described by Pattee 1989).

5 The Implicit Functionalism in "Strong AL"

The strong version of AL is in one respect very similar to the strong version of Artificial Intelligence, or the functionalistic stance within cognitive science. They both embrace the philosophical idea of medium-independence: The characteristics of life and mind are independent of their respective material substrates. Genuine living behaviour can be realized in the computer because life basically is (or belongs to) a class of complex beha viours that could haunt other media than the biochemical.

To a molecular biologist, functionalism may seem rather peculiar: a phi losophical doctrine of "a person who believes that study of the functioning of a person or animal is all important and that it can be studied, by itself, in an abstract way without bothering about what sort of bits and pieces actually implements the functions under study." (Crick 1989) Functionalism in cognitive science has a background in psychology, developed in reaction to behaviourism (that did not allow psychologists to look into the black box of the brain), and in philosophy of mind, put forth as an alternative to a problematic materialistic theory of identity between mental states and neural states. Though one can distinguish between functional analysis as a research strategy (Cummings 1975), explanatory functionalism within psychology appealing to computation by representations within a "language of thought" (Fodor 1975), and metaphysical functionalism as a philosophical theory of mind (Putnam 1960, Block 1980), there are some main features shared by all forms of functionalism:

a) A more or less explicit notion of functional equivalence, where x is functionally equivalent with y, if x has capacities to contribute to the capacities of the whole in a similar (or the same) way as y.

b) A more or less strict reliance on the concept of a Turing Machine.

c) The assumption that the causal structures postulated to be identical with the mental states can be realized by a vast variety of physical sy stems.

ad.b) Mental states are often identified with Turing machine table states; and to give a true explanation of some psychological phenomenon is seen as something like providing a computer program for the mind - or some of its subroutines. One should therefore attempt to give a functional analysis of mental capacities broken down into their component mechanical processes. If these processes are algorithmic (which is often assumed without justification), then they will be Turing-computable as well.

ad.c) It is well known that a digital computer, in principle, can be of many different kinds of components; valves, transistors, chips, neurons, or jets of water. This multiple-realization argument may be true for any formal system, given the right interpretation of the structure that implements it. But three hurdles should be noted:

1. It does not guarantee that our formalization of specific systems - whether mental, biological, or physical - can catch all the essential factors that govern such a system. There might even be aspects of the system that are in principle unformalizable. For instance, the meaning of the symbols manipulated by a cognitive process is context-dependent, and the ultimate context of human language is the natural and cultural world - that may be hard to formalize.

2. The construction (of any material kind) that implements the formal structure (a model of speech, for example) is still in need of our interpretation in order to give any meaning - a thing we might easily forget in the case of a system based on purely syntactical rules appearing to instantiate semantically meaningful beha viour. The semantics is not intrinsic to syntax but depends on our conscious interpretation of the system (Searle 1980).

3. The functioning of a construction implementing some formal structure may well be functionally equivalent to other implementations (or realizations) on one chosen level of description, while on another level it may show dissimilar properties that from a biological point of view may seriously effect its chances of survival in a realistic environment. This fact shows another problem with the property of functional equivalence: that it basically is a logical property; that it is level-dependent; and that it may not cope with "real life" situations where dependence on time-consumption and energetic efficiency on several levels of organization may be crucial for the proper functioning of a system.

Though some elements of functionalism are shared by the strong version of the AL programme, it does not follow that all problems facing functionalism in cognitive science will be the same in AL. I think, however, that two parallels can be drawn.

I. Some psychologists and phenomenalistic philosophers have objected to computational accounts of mind and cognition, arguing that cognitive activity is intimately related to a living human being (or animal) situated in a specific environment and cannot be abstracted from the sensuous, bodily actions of the organism without losing some crucial aspects of this activity, as, e.g., the view from inside (or the Umwelt, i.e., the species-specific subjective universe (Uexküll 1926)). Only in theory is cognition guided by the formal rules of logic; in practice, it is subjected to a sub ject's specific bodily desires, feelings, material needs, interests, purposes, etc. Thus, one cannot separate cognition from volition and emotion, and these "psychical" properties are features of genuine biological processes. As the "psyche" of man or animal in this sense is medium-dependent, so is a living organism's teleonomic orientation and relation to its environment. Therefore, as we cannot have machines that "think" in the same way as humans or intelligent animals think, we cannot have machines that act and react, self-organize and reproduce (and sustain their "autopoiesis") in the same way as real organisms do. To generalize the concept of cognition to include machine as well as personal thinking leaves unanswered the question of the real nature of (the human type of) thinking. In the same way, though we could generalize the concept of life to include lifelike behaviour of machines, and postulate that wet organisms just instantiate some of the same abstract properties of reproduction, metabolism, irritability (or what might be selected as important features of an organism-machine), this would not reveal the specific constraints on the way life has evolved or could have evolved on Earth. And it does not tell us much new about life as it could be - not even in a silicon valley on a foreign planet. The processual characteristics of life will always be higher level phenomena constrained by specific lower level properties. The general phenomenon of emergence is probably a universal feature of life, but one must also look at the set of possible material substrates that can "support" emergence.

II. A second parallel between problems with functionalism in cognitive science and AL concerns the notion of computation and the relationship between the pattern generating properties of the physical functioning of a computer model and our specific interpretations of these patterns. Much research in what Haugeland (1985) dubs "Good Old Fashioned Artificial Intelli gence" relies on the formalists' motto: "If you take care of the syntax, the semantics will take care of itself," i.e., if the system modelled is well formalized and the rules sufficiently strong, the automation of that system guarantees that any output when interpreted makes sense. However, this presupposes that such rules can be found, but many cognitive skills and topics (such as common sense and natural language use) resist formalization. Furthermore, the interpretation is still not intrinsic to the formal system itself but imputed by somebody ascribing meaning to the output symbols: semantics is not intrinsic to syntax. Although the computational paradigm of AL (and connectionism in cognitive science) is different, there seems to be a parallel computationalists' motto at stake: "If you take care of the computational setup, living behaviour will emerge by itself." Again, this presumes that the component units can be formalized appropriately and that "aliveness" exclusively is a property of a formal or computational system. But what if computation is not intrinsic to physical or biochemical systems? We normally conceive of computation as mathematical operations with numbers performed by man or manmade machines (interpreted by man). One may talk of the lac operon in digital-mechanical terms as a "chemical computer" and express one's amazement about DNA metaphorically, calling it "certainly the most sophisticated computer of which we are aware" (Burks and Farmer, 1984), but that does not by itself render the physical or biological world a computer or its processes computational. Anything that obeys physical laws can be simulated on a computer (with limitations on accuracy and speed), but that does not substantiate a computational viewpoint of physical processes.

One could argue that, in contrast to cognitive science that claims it possible to synthesize intelligence because the brain is information processing, AL is not committed to the view that life (e.g., a cell) is information processing; it may be, but that is not central to the possibility of AL. What is central is that the parallel, bottom-up computational approach allows the computer to support emergence of complex behaviour in the same way as a prebiotic chemical system allowed the self-organization of matter into living cells. However, that does not make the model an instance of the thing modelled. A cellular automaton (CA) model of some physical system such as weather, constructed by the same computational approach, is still not to be thought of as an (artificial) instance of weather, realizing the very causal phenomena of thunderstorms. To use a distinction of Kant in this context (as re-introduced by Sober 1991), one recognises that the model system follows a rule (or a set of rules, namely the ones represented by the CA state transition function table), but the natural system acts in accordance with a rule (i.e., physical laws). The natural system does not consult representations in order to "update" its state but behaves as if it had consulted a set of rules. Thus, even if the possibility of strong AL is not committed to the view that life is information processing (that key features of life are governed by intrinsic representations within the living system), admitting this distinction should moderate the claims of computational realizability of life.

6 Conclusion

If AL or some other kind of "empirical mathematics" should have any bearing on the way biologists conceive of their subject matter, the question about the reality-status of the models will inevitably spring up - from both sides: What is the biological content of AL models, and what logical or com pu tati onal lessons can be drawn from the biologists' empirical garden of model species? In biology, many theoretical generalizations have often been made on the basis of a small set of model species such as the fruit fly or Eschericia coli. (The question whether AL models are simulations of biological processes or essentially realize lifelike properties has an analogy within the field of complex dynamics. Here, the relationship between mathematical properties and measurable real behaviour of the systems described is not straightforward. It is not always clear whether real systems actually realize deterministic chaos or if their behaviour "simulates" instances of quasi-periodicity and noise.)

I am not really convinced that computationally based "real" artificial life is impossible; on the other hand, I'm far from persuaded that it's inevitable. I am dubious, because difficult questions about the nature of life and computation remain open. With respect to the latter, one could ask if a concept of computation presupposes a symbolic representational relation of reference between the physical level of propagating signals and a conceptual level of mathematical entities (such as numbers) - such relation of reference eventually being constituted by interpreting organisms or special devices with some minimal complexity (or, to use a technical term of Peirce, interpretants) - or if the process of computation is an intrinsically physical process, that require no "organic" or referential instance in addition to the mere physical functioning of the computing system. This question is beyond the scope of the present note. The following points are not meant to be conclusive, but to express the limits of my doubt about the AL research programme. There are reasons to believe that:

1. Life is not medium-independent, but shows an interdependence of form and matter.

2. Life may be realized in other media than the carbon-chain dominated as a result of a long, natural evolutionary process.

3. AL research may contribute to theoretical biology by:
(i) simulating developmental and evolutionary phenomena of life on Earth,
(ii) simulating life as it could have evolved in non-earthly environments given some set of realistic boundary conditions,
(iii) providing new concepts and models of emergent phenomena belonging to a general set of complex systems of which biological systems (under particular kinds of descriptions) may be a subset.

4. AL may inspire attempts to realize life artificially in other media by in vitro experiments. Such prospects include the experimental approach of molecular biology and protobiology research. However, this is not yet the centre of interest in the present AL research programme.




Acknowledgments

A version of the argument above was presented at the NATO Conference on Complex Dynamics and Biological Evolution at Hindsgavl, Denmark, August 1990, arranged by Erik Mosekilde whom I thank for comments on an earlier version of this paper. I also thank Jesper Hoffmeyer, Mogens Kilstrup, Jakob Skipper, Benny Lautrup, Chris Langton, and Peter Pruzan for stimulating discussions.

Note Added in Proof
I can only recommend the treatise of Kampis (1991) which adds to and in several ways explicate the argument of the present paper.

References

Alberts, B., Bray, D., Lewis, J., Raff, M., Roberts, K., Watson, J.D., 1983, Molecular biology of the cell, Garland, New York.

Block, N., 1980, What is functionalism?, pp. 171-184 in: Readings in the Philosophy of Psychology, Vol.I, N.Block, ed., Methuen, London.

Burks, A.W., 1975, Logic, biology and automata - some historical reflections, Int.J.Man-Machine Studies 7: 297-312.

Burks, C., and Farmer, D., 1984, "Towards modeling DNA sequences as automata," Physica 10D: 157-167.

Cariani, P., 1991, "Emergence and artificial life", pp. 775-797 in C. G. Langton et al., eds., Artificial Life II, Addison-Wesley, Redwood City, California.

Crick, F., 1968, "The origin of the genetic code," J.Mol.Biol. 38: 367-379.

Crick, F., 1989, What mad pursuit, Weidenfeld & Nicolson, London.

Cummings, R., 1975, "Functional Analysis," Journal of Philosophy 72:741-764.

Fodor, J.A., 1975, The Language of Thought, Crowell, New York.

Haugeland, J., 1985, Artificial Intelligence: the very idea, MIT Press, Cambridge, Mass.

Ho, M.-W. & Fox, S., 1988, Evolutionary processes and metaphors, Wiley, New York.

Kampis, G., 1991, Self-modifying Systems in Biology and Cognitive Science, Pergamon Press, New York.

Kampis, G., and Csányi, V., 1987, "Replication in abstract and natural systems," BioSystems 20: 143-152.

Landick, R., and Yanofsky, C., 1987, "Transcription attenuation," pp.1276-1301 in: Escherichia coli and Salmonella typhimurium. Cellular and molecular biology, Vol.2, F.C.Neidhardt, ed., American Society for Microbiology, Washington.

Langton, C.G., 1989, "Artificial life," pp.1-47 in: Artificial Life (Santa Fe Institute Studies in the Sciences of Complexity, Vol. VI), Langton, ed., Addison-Wesley, Redwood City, California.

Lewin, B., 1983, Genes, 2nd.ed., Wiley, New York.

Løvtrup, S., 1981, "Introduction to evolutionary genetics," pp.139-144 in: Evolution Today, G.G.E. Scudder & J.L.Reveal, eds., Hunt Institute for Botanical Documentation, Pittsburgh, Pennsylvania.

Maturana, H.R. and Varela, F.J., 1980, Autopoiesis and Cognition. The Realization of the Living, D.Reidel, Dordrecht.

Orgel, L., 1968, "Evolution of the genetic apparatus," J.Mol.Biol. 38: 381-393.

Papagiannis, M.D., 1985, "Recent progress and future plans on search for extraterrestrial intelligence," Nature 318: 135-140.

Pattee, H.H., 1977, "Dynamic and linguistic modes of complex systems," Int.J. General Systems 3: 259-266.

Pattee, H.H., 1989, "Simulations, realizations, and theories of life," pp. 63-77 in: Artificial Life (Santa Fe Institute Studies in the Sciences of Complexity, Vol. VI), Langton, ed., Addison-Wesley, Redwood City, Calif.

Putnam, H., 1960, "Minds and machines," in: Dimensions of Mind, S. Hook, ed., New York University Press, New York.

Searle, J., 1980, "Minds, brains, and programs," Behavioral and Brain Sciences 3: 417-458.

Sober, E., 1991, "Learning from functionalism - prospects for strong AL," pp. 749-765 in C. G. Langton et al., eds., Artificial Life II, Addison-Wesley, Redwood City, California.

Thompson, D'Arcy W., 1942, On Growth and Form, 2nd ed., Cambridge at the University Press, Cambridge.

Uexküll, J. von, 1926, "Theoretical Biology," Kegan Paul, London.

Vogel, S., 1988, Life's Devices, The Physical World of Animals and Plants, Princeton University Press, Princeton.

Webster, G., and Goodwin, B.C., 1982, The origin of species: a structuralist approach, J. Social Biol. Struct. 5: 15-47.

*