IS METABOLISM NECESSARY? Margaret A. Boden School of Cognitive and Computing Sciences ßÏßÏÊÓƵ Brighton, BN1 9QH, UK Email: maggieb@cogs.susx.ac.uk January 1998 CSRP 482 A shorter version of this paper will be published in the British Journal for the Philosophy of Science, 1999. Abstract Metabolism is a defining criterion of life. Three senses are distinguished. The weakest allows the possibility of strong (virtual) A-Life. The second allows that many A-Life programs are (crude) simulations of real metabolism, and might allow also that (some) non-biochemical A-Life robots are genuinely alive. The third, which stresses energy-budgeting and equilibrating energy-exchanges of some (necessary) complexity, excludes strong A-Life. Even A-Life simulations of metabolism are not alive by this criterion. Nor are A-Life robots, as currently envisaged. The third sense resembles, but also differs from, "autopoiesis." This concept, too, allows (biochemical) artificial life but not strong A-Life. Key words: metabolism, self-organization, autopoiesis, life, A-Life IS METABOLISM NECESSARY? When Norbert Wiener and his colleagues founded the cybernetic movement, they made much of the distinction between information and energy, and declared that the former alone was their proper concern [Wiener, 1948]. Information may supervene on some material base: the "Animal" and the "Machine" of Wiener's sub-title. But the abstract features of control and communication were the focus of cybernetics. Analogous remarks apply to functionalist approaches in the philosophy of mind, and also to AI and computational psychology. All these focus on abstractly defined processes, or software, rather than neurophysiology, or hardware. (We must avoid punning: the "functions" in functionalism are abstract mappings like mathematical functions -- not underlying physical processes, nor biological "functions" defined by reference to evolution.) What of the philosophy of life? Is (abstract) functionalism also prominent here? And is it acceptable? In particular, are energy and matter essential to life? -- If so, is that equivalent to saying that metabolism is essential? Or is there more to metabolism than energy-usage? There is no universally agreed definition of life. But certain properties are normally mentioned. They are: self-organization, emergence, autonomy, growth, development, reproduction, adaptation, responsiveness, and metabolism. These seem to capture our everyday intuitions about the nature of life. Many biologists, and some philosophers [e.g. Berdua, 1996], would add evolution. This has several difficulties. One is that individual living things, such as lions or oak-trees, could no longer be counted as paradigm cases of life. Rather, the basic example of life would be an evolving, multi-generational, population. This is not only counterintuitive as regards our normal usage, but also causes difficulty (as we shall see) with respect to another vital property, metabolism. Another drawback is that a breeding population that for some reason had stopped evolving would not count as living: at best, it would be "dormant." Someone might counter both these objections by allowing individuals to remain the paradigm case, but insisting that a necessary condition for an individual thing to be alive is that it is a member of a self-reproducing species that has evolved. Even this formulation, however, besides casting some doubt on the status of sterile hybrids such as mules, counter- intuitively locates a necessary condition for vitality outside the organism itself. A third difficulty of taking evolution to be a criterion of life is that this would make creationism, the view that each living species was specially created by God, self-contradictory. It would be ruled out of court by semantics, so the question of its biological credibility would not arise. Admittedly, there are philosophical problems attached to the notion of an immaterial God creating a material universe. But these apply across the board: there is no special inconceivability about God's creation of biological organisms. Finally, there is a "humanized" version of the third objection. If we were to create some radically new organism by genetic engineering, and if for some deep reason it were incapable of variation and therefore of evolution, it could not be counted as alive. Perhaps there could be no such creature: variation, or replication errors, may in fact be unavoidable. Indeed, this may follow from the second law of thermodynamics. But a faithfully self-replicating life-form does seem to be conceivable. Certainly, evolution appears to be a universal feature of the terrestrial life we know about. But someone might admit this as a matter of fact, even as an inevitable consequence of thermodynamics, while insisting that the MEANING of "life" does not include evolution. This would leave it open for there to be non-evolved life somewhere in the universe. For example, some philosophers argue that the concept of evolution is dependent on that of (the relevant sort of) self-organization, and that life can -- and probably once did -- occur without it [Maturana & Varela, 1980: see below]. Even more important here than the apparent ubiquity of evolution is its explanatory force. Evolutionary theory plays an exceptionally strong, and widely integrative, explanatory role in biology. Most biologists who resist the reductionist temptations of molecular biology, taking the form of whole organs and organisms as their EXPLANANS, see it as fundamental. But a minority do not. For instance, Brian Goodwin [1990; Webster & Goodwin, 1996, part II] and Stuart Kaufmann [1992] argue that biological self-organization is a more fundamental explanatory concept than evolution -- and that the two processes can sometimes pull in different directions [see also Wheeler, 1997]. But even these theoretical mavericks allow that Darwinian evolution selects, and so (superficially) shapes, the range of living things that survive, given the (deeper, wider) potentialities afforded by self-organization. In short, all serious biologists -- I exclude the tiny handful of creationists -- acknowledge that evolution has considerable explanatory force. This fact may understandably affect our definition of "life". We know, from various examples in other fields, that scientific advance sometimes leads us to redefine familiar concepts in initially counter-intuitive ways. It is hardly surprising, then, if the theoretical importance of evolution is reflected in the way we conceive of life itself. So, despite the difficulties we have noted let us provisionally allow evolution to be included in the list of vital criteria. If there is general agreement on anything in this controversial area, it is that the central feature of life is self-organization. Self-organization involves the emergence (and maintenance) of order, out of an origin that is ordered to a lesser degree. It concerns not mere superficial change, but fundamental structural development. The development is spontaneous, or autonomous, in that it results from the intrinsic character of the system (often in interaction with the environment), rather than being imposed on it by some external force or designer. Indeed, the concept of self-organization covers all the others on the list. That's not to say that every self-organized system is a living thing. In the Belousov- Zhabotinsky reaction, for example, mixing two liquids results in the spontaneous emergence of order (visible whorls and circles), but no-one would regard this as life. The point, rather, is that each of the other vital properties, if and when they occur, can be seen as an aspect of self-organization. Emergence, autonomy, and development form part of the definition given above. Metabolism and growth are aspects of self-maintenance, another part of the definition. Responsiveness falls under "interaction with the environment," which also is explicitly mentioned. And reproduction, evolution, and adaptation are forms of self-organization. Organization in general, and self-organization in particular, is usually conceived of as an abstract, functionalist notion. Sometimes, however, it is defined in terms of energy. For instance, it is said that (all) self-organizing systems have to be energetically open systems, and that "it is [the] continual flux [of energy and matter] that is the wellspring of new forms" [Thelen & Smith, 1993, p. 54]. For reasons that will become clearer below, it is more helpful to define self-organization in abstract terms (as is usually done), allowing that it can be realized in living systems only by means of various types of "energy-flux." Functionalism is prominent in A-Life. A-Life researchers typically think of vital properties in terms of information and computation, not matter or energy. For example, Von Neumann defined the general requirements of reproduction in logical terms, and pointed out that copying-errors (an informational notion) could result in adaptive evolution [Burks, 1966, 1970]. Similarly, Christopher Langton, in his Call for Papers for the first conference identifying Artificial Life as a unitary project, said: "The ultimate goal of A-Life is to extract the logical form of living systems" [Levy, 1992, p. 113]. Nevertheless, none of these scientists doubts that living things are material entities. Langton makes this explicit in his statement that life is "a property of the organization of matter, rather than a property of the matter which is so organized" [Langton 1989, p. 2] So far, then, our question "Are matter and energy essential to life?" seems to be answered with a guarded "Yes." Some matter is organized, somehow. But the nature of the material stuff is philosophically irrelevant to the status of the physical system as a living thing. In this, A-Life scientists resemble functionalist philosophers of mind. Putnam's original definition of functionalism could in principle be satisfied by squads of angels jumping on and off immaterial pinheads [Putnam, 1967/1975]. But functionalists normally do assume a material base, whether wetware or hardware, on which mental properties somehow supervene. Indeed, this is why functionalism was welcomed by scientifically-inclined philosophers of mind as an advance on, as opposed to a wholesale rejection of, the identity theory and central state materialism. However, Langton [1986] also says: "The ultimate goal of the study of artificial life would be to create 'life' in some other medium, ideally [sic] a virtual medium where the essence of life has been abstracted from the details of its implementation in any particular model." Such 'life' would inhabit cyberspace, a virtual world existing only in computers. Whether genuine life could exist within a purely virtual medium -- in other words, whether "strong A-Life" is in principle possible -- is controversial. It is disputed even by many A-life researchers, so Langton's stated aim cannot be ascribed to A-life in general. One A-Life researcher who does agree with Langton is Tom Ray, a tropical botanist whose computer models of co-evolution in the virtual world "Tierra" have led to the foundation of the "Digital Reserve" [Ray, 1992, 1994]. This is a virtual space spread across a worldwide network of computers, which allow their spare space to be used at idle times. The creatures (Ray's word) inhabiting the Digital Reserve, like those within Tierra itself, are strings of self-replicating computer code. They can mate (exchange genetic instructions), compete, and evolve. For example, some code-strings evolve which lack the instructions responsible for self-replication, but which can "parasitize" the code of other creatures in order to replicate themselves. This is a successful evolutionary strategy because "fitness" is defined in terms of access to computer-memory -- and a "species" with shorter strings can fit more individuals into a given memory- space. The creatures let loose in the Digital Reserve move from one computer to another in their search for unused memory-space. (Because of the way they are implemented, the software creatures cannot "escape" into computers not on the Reserve network, nor infest the everyday workings of those that are included.) Ray insists that the Digital Reserve is an experiment in the creation of new forms of life. Those who, like Langton and Ray, regard strong A-Life as a real possibility defend their view by making two interconnected claims. First, that the virtuality is limited: computers, after all, are material things, which need energy in order to function. Second, that the criteria for life are essentially abstract, saying nothing whatever about the nature of its (admittedly necessary) material grounding. If we consider the list of vital properties given above, we can argue (perhaps incorrectly: see below) that all but one of them are abstract, informational concepts. This applies not only to the core concept of "self-organization," but also to reproduction (self-copying) and evolution (adaptive change by means of reproduction, variation, and selection). The one obvious exception is metabolism. But what, exactly, is metabolism? It locates life in the physical world (no angels on pinheads). But it does not denote mere materiality. A volcano is a material thing, and so is a grain of sand, but neither of these metabolises. Rather, metabolism (in the weakest sense of the term) denotes energy dependency, as a condition for the existence -- and the persistence -- of the living thing as a unitary material system. If energy-dependency were all there was to it, then strong A-Life would be possible. For, as both Langton and Ray are quick to point out, virtual life satisfies this criterion. Strong A-Life is utterly dependent on energy. Pull the plugs on the computers, and cyberspace is not merely emptied, but destroyed. Strong A-life, having once existed, would have died. However, "metabolism" is normally used to mean more than mere energy-dependency. Two further senses of the term can be distinguished, associated with notions of using, collecting, spending, storing, and budgeting energy. These activities are characteristic of life. (Active volcanoes involve huge amounts of energy, without which they would not exist. But they don't "use" it, "collect" it, "store" it, or "spend" it, except in a weakly metaphorical sense -- and they certainly don't "budget" it.) A stronger sense of metabolism supplements mere energy dependency with the idea of individual energy packets, used to power the behaviour of the system. "Behaviour," here, means the creature's bodily (and mental) activities, the existence of its body being taken for granted. (Individual creatures are here treated as the paradigm of life, so this sense of metabolism sits uneasily with evolution on the list of vital criteria.) Each living system has assigned to it, or collects for itself, a finite amount of energy. This is used up as it engages in its various activities. When the individual's energy is spent, either because it is no longer available in the environment or because the system can no longer collect or use it, the energy-dependent activities must cease and the system dies. Much A-Life work models metabolism in this sense. Examples abound of programs that simulate individual animals with distinct energy-levels, raised by eating and rest, and reduced by activities such as food-seeking, fighting, and mating. Some of these even assign different sub-packets of energy to various drives, so that at a particular time a creature might have energy available to mate, but not to fight. (For a very early example, where a simulated rat has to choose between seeking warmth and food, see [Doran, 1968].) However, the "packets" and "sub-packets" here are not actual energy-sources or energy-stores, but mere simulations. So, although the criterion of energy-dependency is still satisfied (via the electric plug or battery), metabolism in the second sense is not in fact achieved. Some very early efforts in A-Life (around mid- century) already involved the idea -- and the reality -- of individual energy-packets. Grey Walter's [1950, 1951] mechanical "tortoises," Elmer and Elsie, were simple robots that used their energy to engage in physical activities. They moved around the floor by means of electric power, every so often abandoning their current activity in order to recharge their batteries. This example is not strictly germane to the question whether strong A-Life is possible. For "strong A-Life" does not refer indiscriminately to any A-Life artefacts (including A- Life robots, and systems grounded in exotic biochemistries), but only to virtual creatures inhabiting virtual worlds. However, the second definition of metabolism would cover A-Life robots broadly comparable to Grey Walter's tortoises. Such robots could, therefore, be termed alive according to this criterion. The second sense of metabolism, clearly, is not the biologist's concept of it. For no biologist takes the existence of a creature's body for granted. On the contrary, one of the prime puzzles of biology is to explain how living bodies come into existence, and how they are maintained until the organism dies. We therefore need a third, stronger, definition of metabolism to capture what biologists normally mean by the term. The third sense of metabolism refers to the use, and budgeting, of energy for bodily maintenance as well as behaviour. Metabolism is here seen as a type of material self-organization which (unlike the Belousov-Zhabotinsky reaction) involves the autonomous use of matter and energy in building, growing, developing, and maintaining the bodily fabric of a living thing. The matter is needed as the "stuff" of which the body is made. And the energy is needed to organize this matter into something that persists in its existence despite changes in external conditions. Metabolism, in this strong sense, generates and maintains the distinction between the physical matter of the individual organism and that of other things, whether living or not. Metabolism in this third sense necessarily involves closely interlocking biochemical processes. Living matter cannot be created "de novo," so new molecules must be synthesized by the organism -- which molecules themselves make up the organism. Moreover, the living system (subject, like every physical thing, to the second law of thermodynamics) continuously tends to disorder and the dissipation of energy. Hence metabolism must involve constant energy exchanges between organism and environment. The simplest conceivable living things would take their energy directly from the environment whenever they needed it. But this would leave them vulnerable to situations in which no "new" energy was immediately available. (Analogously, many computer models rely on the plug in the wall, so are vulnerable to power cuts.) If, by chance, the organism became able to store even small amounts of excess energy for later use, its evolutionary fitness would be enormously increased. Inevitably, then, metabolic systems typically not only exchange energy with the outside world but also do internal energy budgeting. Excess energy is stored, so that reliance on direct energy-collection is avoided. Living organisms convert external energy into some substance ("currency") that can be used to provide energy for any of the many different processes going on inside the organism. (This is "the first fundamental law of bio-energetics," and it turns out that only three convertible energy currencies -- one of which is ATP, or adenosine tri-phosphate -- are used by terrestrial life [Moran et al., 1996, para. 4.2].) Further, purely internal, energy-exchanges are required as the collected energy is first converted into substances suitable for storage and then, on the breakdown of those substances, released for use. Very likely, these processes will produce waste materials, which have to be neutralized and/or excreted by still other processes. In short, metabolism necessarily involves a nice equilibrium between anabolism and catabolism, requiring a complex biochemistry to effect these vital functions. Bodily maintenance is normally continuous. But the underlying metabolic processes are more active at some times -- of the day, year, and life-cycle -- than at others. Sometimes, they are drastically slowed down, or (perhaps) even temporarily suspended. In hibernating animals, for instance, metabolism is kept to a minimum: respiration and excretion occur at a very low rate. Even in the case of seeds or spores frozen, or entombed, for centuries, some minimal metabolic activity may have been going on. But what if it has not? It's not clear that this strong concept of metabolism assumes that self-maintenance must be absolutely continuous, allowing of no interruptions whatsoever. If biochemical research were to show that metabolism is occasionally interrupted, in highly abnormal conditions (such as freezing), so be it. Indeed, we already speak of "suspended animation": a spore may be currently inactive, but if it retains the potential to metabolise in suitable conditions we don't regard it as "dead." What counts as the body is not always unproblematic. We've noted that it must be a material unity supporting various vital properties. Normally, each and every part of the bodily fabric is built and maintained by metabolism. And normally, many parts of the human body are sources of perceptual information and/or under voluntary control. But what of physical prostheses? These include a wide range of examples. Some, such as cardiac pacemakers, are "involuntary" muscle-controllers of whose (successful) activities the human host is unaware. Some are artificial sensory organs, such as retinas and cochlear implants, whose activities cannot be controlled (except to be turned off) but furnish information that the person can consciously use. Others, from simple peg-legs to the many types of jointed artificial limbs, are replacements for motor organs; these involve varying levels, and methods, of voluntary control. Yet others are implanted electrical circuits, integrated with the neuromuscular anatomy so that (for example) a paraplegic person can excite their leg-muscles at will. None of these is part of the body according to the criterion of metabolic emergence. But some involve close connections to specific aspects of metabolic function. And many are crucial to the dynamics of interaction between the person and their environment. Moreover, motor and sensory prostheses, like ordinary tools, may come to feel like part of the body, from the user's point of view. Admittedly, an artificial hand will not hurt if it is pinched. But its amenability to voluntary control may be continuous with, even to some extent experientially indistinguishable from, that of genuine bodily parts. Those philosophers who stress the embodiment of cognition, glossing ordinary tools -- such as chisels, walking-sticks, and microscopes -- as extensions of our phenomenological world, might want to say that prostheses are, or with practice can become, part of the body. Nevertheless, lacking metabolism they are not strictly alive. Clearly, the third sense of metabolism rules out strong A-Life. Metabolism involves material embodiment. It also requires a complex equilibrium of biochemical processes. It cannot be adequately modelled by a system's freely helping itself to electricity by plug or battery, or even by assigning notional "parcels" of computer power to distinct functions within the program. Virtual creatures may have individual energy-packets, and some form of energy-budgeting, but these are pale simulations of the real thing. Even "biochemical" A-Life models are excluded from the realm of the living, at least while they are still confined to cyberspace. This forbids us to regard as truly living things a "species" of A-Life that has recently attracted considerable attention -- from the general public, as also from the A-Life community. I am thinking of the cyber-beings conjured up by running "Creatures." "Creatures" is a computer-game, or more accurately a computer-world, built by the use of A-Life techniques [Grand et al., 1996]. It is a far richer virtual world than that of other computerized "pets" -- such as "Dogz", "Catz," and the electronic Tamagochi-chick that the user must rest, exercise, and clean. What is of special interest here is that "Creatures" includes a (crude) model of metabolism, as well as of behaviour. Moreover, its main designer, Steve Grand, views its virtual denizens as primitive forms of life [e-mail correspondence]. The human user of Creatures can hatch, nurture, aid, teach, and evolve apparently cuddly little creatures called "norns". Up to ten norns can co-exist in the virtual world (future increases in computer power will make larger populations possible). But even one solitary individual will keep the person quite busy. (In the description below, I have refrained from peppering the text with scare-quotes. Strictly, terms like eat, food, poison, hungry, and so on should be in inverted commas. But the word learn should not: these software creatures really do adapt their behaviour as a result of their individual history.) One of the user's tasks is to ensure that all the norns are able to find food when they are hungry, and that they learn to eat the right food and avoid poisons. Another is to teach them to respond to simple linguistic inputs (proper names, categories, and commands), different norns receiving different lessons. Yet another is to help them learn to cooperate in various simple ways. These lessons are reinforced by administering rewards and punishments to each individual (using the computer-mouse to tickle its nose or slap its bottom, respectively), as appropriate to its current actions and perceptions -- which have to be carefully monitored accordingly. In addition, the user can nurture the norns more directly, for instance by leading them towards food or away from danger. Also present in the virtual world are a number of "grendels," fierce and predatory creatures who are bad news for norns. The user must protect the norns from the attentions of grendels, if they are not to suffer an early death. Finally, the human can evolve new norns likely to combine preferred features of appearance and behaviour, since mating two individuals results in (random) recombinations of their "genes". A norn's genes determine its outward appearance and the initial state of its individual neural-network "brain" (initially 1,000 neurones and 5,000 synapses) whose specific connection-weights change with experience. The genes also determine its idiosyncratic "metabolism". Each creature's behaviour is significantly influenced by its biochemistry. This models global features such as widespread information-flow in the brain, hormonal modulations within the body, the norn's basic metabolism, and the state of its immune system. The virtual biochemistry is defined in terms of four types of biochemical object. First, there are 255 different "chemicals," each of which can be present in differing concentrations. (These are not identified with specific biochemical molecules: the functions of the 255 substances are assigned randomly.) Second, various biochemical "reactions" are represented. These include fusion, transformation, exponential decay, and catalysis (of transformation and of breakdown). Third and fourth, there are a number of "emitter" and "receptor" chemicals, representing various processes in the brain and body (for example, activity in the sense-organs). Taken together, these biochemical categories are used to build feedback-paths modelling phenomena such as reinforcement-learning, drive-reduction, synaptic atrophy, glucose-metabolism, toxins (from plants or bacteria), and the production of antibodies. Clearly, this general architecture offers significant potential for advances in A-Life modelling. Its largely untapped complexity, including its ability to model global features of information-processing, makes it a promising test-bed. It could be advanced, for example, by incorporating recent AI-ideas on the computational architecture underlying motivation and emotion [Sloman, 1990; Wright et al., 1996; Beaudoin, 1994], which have as yet been modelled only in very preliminary ways [Wright, 1997]. Moreover, "Creatures" is undeniably seductive. It commands not only our time and attention, but our animistic tendencies. All but the most hard-headed of users find themselves spontaneously addressing the little creatures as though they were alive. And some users sadly mourn the demise of individual norns, despite knowing that they can hatch others at the touch of a button. This projection of individual personality onto the norns is encouraged by the fact that each one is unique. Besides superficialities such as hair and eye-colour, each one hatches with an idiosyncratic "brain" and each goes through (enjoys? suffers?) a unique experiential history. For all that, "Creatures" is a simulation, not a realization. The norns don't enjoy their history, nor suffer it either. There is no actual glucose, and no actual metabolism. The system is not even a chemically plausible simulation of real molecular processes. If we regard metabolism as -- literally -- vital, we must reject the claim that norns, and their cyber-cousins, are simple forms of life. We must also reject any claim that A-Life robots, at least as currently envisaged, are living things. A-Life robots are typically "situated" robots, engineered (or evolved) to respond directly to environmental cues. Some don't look at all life-like [Cliff, Harvey, & Husbands, 1993]. Others resemble insects in their physical form, and may have control systems modelled on insect neuroanatomy [Brooks, 1986, 1991; Beer, 1990]. Certainly, such robots are in a significant sense autonomous, especially if they have been automatically evolved over many thousands of generations [Boden, in press]. And they undoubtedly consume real energy as they make their way around their physical environment. Unlike classical robots, they are embedded in the world, in the sense that they react directly to it rather than by means of a complex internal world-model. But being embedded does not necessitate being (truly) embodied. I argued above that a body is not a mere lump of matter, but the physical aspect of a living system, created and maintained as a functional unity by an autonomous metabolism. If that is right, then these robots do not have bodies. Conceivably, some future A-Life robots might be self-regulating material systems, based on some familiar or exotic biochemistry. Just how exotic that biochemistry might be is unclear. In principle, it need not even be carbon-based. However, it may be that carbon is the only element capable of forming the wide range of stable yet complex molecular structures that seem to be necessary for life. And Eric Drexler [1989] has argued that even alien biochemistries would have to share certain (relational) properties with ours. They would have to employ general diffusion, not channels devoted to specific molecules; molecular shape-matching, not assembly by precise positioning; topological, not geometric, structures; and adaptive, not inert, components. Whatever the details, artefacts grounded in exotic biochemistries might well merit the ascription of life: not "strong A-life" confined to cyberspace, but real, metabolising, life. At present, however, there is nothing in A-Life that presages such alien creatures. The third sense of metabolism distinguished above is reminiscent of the concept of "autopoiesis" (etymologically: "self-making") developed by Humberto Maturana and Francisco Varela [Maturana and Varela, 1980]. More precisely, since autopoiesis as such is a purely abstract concept, which can be applied (for example) to inorganic chemistry, to business organizations, or to whole societies, it recalls the type of autopoiesis they take to be characteristic of life. Autopoietic systems are defined in terms of their organization, not of their components nor even the properties of their components. What is crucial is "the processes and relations between processes realized through components" [op. cit., p. 75]. An autopoietic system (or "autopoietic machine") in general is formally defined as follows: [...] a network of processes of production (transformation and destruction) of components that produces the components which: (i) through their interactions and transformations continuously regenerate the network of processes (relations) that produced them; and (ii) constitute it (the machine) as a concrete unity in the space in which they (the components) exist by specifying the topological domain of its realization as such a network. [op. cit., p. 79]. So defined, the concept can be applied (for example) to inorganic chemistry, to business organizations, or to whole societies. None of these are living organisms, although the last two are higher-level systems whose lower-level components are living things. For life, embodiment is required: "autopoiesis in the physical space [is] a necessary and sufficient condition for a system to be a living one" [op. cit., p. 84]. In plainer English, living things are of necessity physical, with a bodily fabric produced and maintained by themselves. Their outstanding feature is a form of self-organization termed autopoiesis. This is a special case of homeostasis, where what is preserved is not one feature (such as blood-temperature), but the organization of the system itself as a unitary whole [op. cit., pp. 78-79]. In other words, the very boundaries of the system as a unity, as well as its bodily components, are continuously produced by its own activities. A human body, or a tree, is an autopoietic unity. But they are higher-level autopoietic systems, made up of many such systems at a lower level [op. cit., pp. 107ff]. The basic phenomenon here is not the formation of a body with arms and legs, or leaves and boughs, but the self-organization of a single cell. The generation of the cell-membrane both bounds and constitutes the cell as an autonomous vital entity, distinguishable from its environment. Explaining how this can happen is universally acknowledged to be one of the core problems of biology (see [Maynard-Smith & Szathmary, 1995, ch. 7]). Maturana and Varela unequivocally identify it as THE fundamental problem. They addressed it in the early 1970s by outlining a formal theory of autopoiesis [Varela et al., 1974; see also McMullin & Varela, 1997]. This was further developed by Milan Zeleny a few years later [Zeleny, 1977]. Zeleny's computer model simulates the emergence of a self-repairing membrane in a two-dimensional cellular automaton. (He compared it with John Conway's "Game of Life" [Zeleny, 1977, p. 18]). That is, it shows how a "cell" can be created, and simultaneously differentiated from its "environment." For understanding how such a system could in principle arise, one does not need to specify the chemicals involved in practice. Accordingly, Zeleny's molecules (as in "Creatures," two decades later) are not identified as specific chemicals, but are defined in relational terms. That is, they are classed as substrate molecules, catalysts, and three types of molecular link (free, single-bonded, and fully-bonded). Likewise, the chemical processes modelled are abstractly defined events such as molecular production, bonding, disintegration, and diffusion -- whose rates can be varied by adjusting parameters in the rules. The movements and interactions of the various components are governed by about twenty simple rules, although some molecular "meetings" occur partly by chance. The three molecular links have differing stability, but any link can (under certain conditions) disintegrate into two units of substrate. They also have varying potential for bonding. A free link can bond with a chain of bonded links; two chains of bonded links can be bonded into one, or re-bonded after their connecting link has disintegrated; and two free links can be bonded together to start a chain formation. A free link (and an empty "hole") is produced whenever a catalyst interacts with two units of substrate. Disintegration and bonding can take place without catalysis, although bonding can occur only outside the catalytic neighbourhood. The role of the catalyst, given substrate molecules nearby, is to produce more complex component-linkages, which in turn are capable of bonding. Ultimately, they polymerize to generate a membrane whose self-closure both creates and bounds an enclosed space [Zeleny, 1977, p. 14]. This system is not only self-generating, but also self-maintaining. The membrane is semi-permeable. It permits the diffusion of substrate molecules, because these are allowed to pass through a bonded link. Substrate can enter the space from the environment if there are holes in the space, adjacent to the membrane. By contrast, neither catalyst molecules nor free links can pass out through the membrane. These highly active (and, in the case of catalysts, relatively rare) components therefore stay "trapped" within the space. Moreover, the re-bonding rule cited above enable spontaneous repair of the membrane if it is ruptured at any point by disintegration. In other words, this is an autopoietic system. More accurately, it is one of a large set of possible autopoietic systems, each produced by some variation in the rules determining potential interactions, or in the variables used in instantiating those rules [Zeleny, 1977, pp. 21-25]. For instance, one can vary the number -- and operational properties -- of catalysts, their concentrations, or the size and shape of their "neighbourhoods." One can also vary the rules regarding their interactions, with the result that catalytic activity in the space as a whole changes (perhaps rhythmically) through time. Similarly, one can introduce a wider variety of bonds, or vary the properties, and amount, of the substrate(s). Different types of membrane, and different "cellular" behaviours, will be generated accordingly. (Some rule-variations, of course, will make autopoiesis less long-lasting, or even prevent its development altogether.) Neither Maturana and Varela, nor Zeleny himself, regard this computer simulation of life as real life. It is indeed an autopoietic system, but the autopoiesis does not take place "in the physical space." The same applies to Varela's more recent simulation of autopoiesis by means of an artificial chemistry [McMullin & Varela, 1997]. That's not to say that Maturana and Varela deny the possibility of any sort of artificial life. On the contrary, they explicitly allow that one could in principle "design" and "make" a living system [Maturana & Varela, 1980, p. 114]. They even remark that we may, unwittingly, already have done so. In saying this, however, they are not thinking of virtual creatures in computer memory, but of self-maintaining biochemical systems. Nor are they thinking of some marvel of nanotechnology, whereby individual bio-molecules are directly arranged "by hand." Rather, they have in mind the human scientist's "creating the conditions under which [autopoietic biochemical systems] form themselves" [Zeleny, 1977, p. 27]. The similarity between the concept of autopoiesis and our third, strongest, sense of metabolism is clear. In particular, both refer to self-organized physical processes that generate, and maintain, the bodily fabric of living things. And the computer model of autopoiesis just described could well have appeared in an A-Life journal, alongside other simulations of metabolic processes. (A paper co-authored by Zeleny on autopoiesis in INORGANIC chemistry was published in the proceedings of the first A-Life conference [Zeleny et al., 1989].) The main differences between metabolism and autopoiesis, and between accounts of life based on these concepts, result from the emphasis put by Maturana & Varela on the self-production of the organism's boundaries as a unitary system. This is the core of their theory. They speak, for instance, of the "total subordination of [all the processes of change within] the system to the maintenance of its unity" [op.cit., p. 97]. This perspective leads them to say a number of things about metabolism (which they call "energy-relations"), and about the concept of life in general, which most other researchers would not. For example, because autopoietic systems are defined not by their components, but by their processes and the relations between their processes, the biochemical questions asked by Maturana and Varela are subtly different from those asked by most researchers into metabolism [Moran et al., 1996]. They insist, for instance, that questions about the origin of life -- the formation of the first cells -- should be focussed not on molecules, but on the relations that molecules can have with one another [Maturana & Varela, 1980, p. 93]. This is why the cell-membrane model described above does not specify any particular chemicals. Admittedly, some other theorists also ask "relational" questions about the metabolism involved in the origin of life. One example is Drexler's [1989] work, mentioned above. Another is Stuart Kauffman's [1992] research on autocatalytic networks. The latter research, widely cited in A-Life, is generally regarded as focussing on the metabolic origin of life. Maturana and Varela interpret it differently [Maturana & Varela, 1980, pp. 93-94]. Seeing the self-organization of boundaries as an all-or-none phenomenon, they allow no intermediate stages between non-autopoietic (non-living) and autopoietic (living) systems. The first living system is the cell, brought into being by the formation of the cell-membrane. Autocatalytic networks, they insist, do not qualify as autopoietic systems, because they do not determine their own topology. The network's boundaries are set by something external, namely, the walls of the container in which the relevant chemical processes are taking place. If actual metabolic processes were to be reproduced in vitro, they would not constitute an autopoietic system either. (This is not implausible: few people would say that a metabolic network functioning in a test-tube was an example of life.) I argued above that we can allow the possibility, in principle, of non-continuous metabolism in living things. But no possibility of interrupted autopoiesis is granted by Maturana and Varela [op. cit., p. 98]. This is not primarily because they believe (as most prononents of metabolism probably do also) that, as a matter of fact, the physical processes concerned are continuously dynamic. Rather, their conception of autopoiesis as the fundamental source of the unity of the living thing forbids any suggestion that it might be interrupted without thereby destroying the integrity of the system in question. As they put it: "in a living system, loss of autopoiesis is disintegration as a unity and loss of identity, that is, death" [op. cit., p. 112]. More generally, metabolism for Maturana and Varela is not part of the definition of autopoiesis. Rather, it is constitutive of its material structure when it is realized as a living thing [op. cit., 88]. Accordingly, all life depends on metabolism -- but autopoiesis is the more fundamental concept. As for the concept of life itself, Maturana and Varela refuse to accept reproduction or evolution as part of its definition [op. cit., pp. 105-107]. They argue that the notion of self-reproduction assumes the pre-existence of an identifiable unity. That unity, for them, is explained by autopoiesis. This is not a merely semantic point, but a substantive biological hypothesis. The first living things (cells) would have satisfied the criterion of autopoiesis, but might have been incapable of reproduction. The earliest reproduction could have happened accidentally. If autopoietic powers were distributed across the whole system, and the cell were to be mechanically broken into two by some external force, then each part would be capable of persisting as a self-coherent unit. If, by chance, any structural change were to happen which made this accidental breakage easier, or which made some non-accidental (internally generated) breakage probable, then systems capable of reproduction would gradually evolve so as to exclude non-reproducing systems. Hence, they say, the ubiquity of reproduction in life as we know it. As for evolution, this concept presupposes the existence of reproduction, so evolution also is excluded from the definition of life as such. Maturana and Varela's emphasis on self-organized unity as the constitutive principle of the living system also leads them to reject informational concepts in biology. Such concepts apply, they say, only when describing the passing of messages ("the reduction of uncertainties") between two independent unities, where the messenger acts as "an arbitrary non-participant link" [op. cit., p. 102]. Nor do they allow that processes within a living body, as opposed to processes in a computer, can really have a function. "Function," they say, applies only to the domain of purposeful human design; moreover, function defined (metaphorically) in relation to evolutionary fitness is, like evolution itself, necessarily secondary to autopoiesis. This puts them at odds not only with many biologists, but also with most workers in cognitive science and A-Life. They would certainly reject the suggestion (above) that reproduction can be glossed as an informational notion. They distinguish between "self-reproduction" and "self-copying," saying that only the first involves autopoiesis. They even describe the genetic "code" as a fundamentally misleading metaphor [op. cit., p. 102]. Since DNA is a constitutive part of the autopoiesis of the organism, not a mere arbitrary non-participant link, its role cannot be informational. Admittedly, it may be "arbitrary" in the sense that some other biochemical carrier of heredity might have evolved (as ATP might not have been the intra-cellular energy currency used by all known life). But, once DNA has taken on this autopoietic role, it is integral to the self-reproduction of the organism. Maturana and Varela's avoidance of informational concepts leads them to deny also that organisms have any inputs from or outputs to the environment. They speak only of "perturbations" of the system itself. They grant that an observer may find it useful to distinguish between "internal" and "external" perturbations, but insist that for the autopoietic system itself these are indistinguishable. A state of the system is merely a state of the system: it does not carry a label announcing its causation. In reality, then, they are all internal perturbations. (Moreover, they are all perturbations in the present tense: an observer may say that the system has "learnt" something, but the system merely does what its state at that moment leads it to do.) Equally, they refuse to describe bodily processes, even the activities of nervous systems, as involving internal representations of the environment. Such language, they say, is "metaphorically useful, but inadequate and misleading [in revealing] the organization of an autopoietic system" [op. cit., p. 99; cf. pp. 22- 26]. Only human autopoiesis can produce representations. For only human beings can act as observers of their own cognitive interactions, and treat (primarily linguistic) representations of them as though they were independent entities [op. cit., p. 14]. (Likewise, human beings can sometimes distinguish between "inputs" and "outputs" in their own case, because they can reflect on the causal history of their own states.) The eschewal of terms such as input, output, information, function, and representation -- and the insistence that "the neuron cannot be considered as the functional unit of the nervous system" [op. cit., p. 19] -- is especially interesting, given that Maturana was a co-author of the famous paper "What the Frog's Eye Tells the Frog's Brain" [Lettvin et al., 1959]. He and his colleagues interpreted their research as having found single-cell feature detectors in the frog's retina. Their experiments provided the first neurophysiological evidence of such things. But the idea that perceptual feature detectors might exist had come from an information-theoretic context. At the end of their paper, the authors specifically acknowledged their intellectual debt to Oliver Selfridge, whose pioneering model "Pandemonium" simulated a series of increasingly complex visual feature detectors [Selfridge, 1959]. Maturana's seminal work encouraged David Hubel and Torsten Wiesel to look for feature detectors in the visual cortex of cats and, later, monkeys [Hubel & Wiesel, 1959, 1962, 1968]. Since then, many others have been identified in the brains of humans and other species. Some of these appear to detect fairly complex inputs, such as faces, hands, or paws. One, found in certain wading birds that use their beaks to search for food in the mud, fires when the lower surface of the upper beak-half and the upper surface of the lower beak-half are stimulated simultaneously [Pettigrew & Frost, 1985]. Most neuroscientists and cognitive scientists would say that the cell in question carries the information that, or even represents the fact that, the bird has picked up something potentially edible. Indeed, Maturana and his co-authors made similar remarks about the "fly-detectors" they found in the frog's retina. Now, however, Maturana avoids such language. Rather, he says that the nervous system -- of which the relevant cell is an integral part -- continuously couples the organism with its environment so that it (the organism as a whole) is perturbed in certain ways, given certain conditions in the environment. These perturbations sometimes involve specific types of behaviour (conduct), such as nicely directed movements of a frog's long, sticky tongue. An observer may find it convenient to describe this behaviour as an appropriate "output" caused by "input" to the relevant "feature detector," and may regard this causation as part of the "function" of the cell concerned. But for Maturana and Varela, the proper description of this sensory-motor circularity as a biological phenomenon is in terms of the ongoing autopoiesis of the whole animal. As Maturana puts it: "Only conduct itself [not the neuron, nor any fixed group of neurons] can be considered as the functional unit of the nervous system" [op. cit., p. 19]. It is not only creatures with nervous systems that are credited by (most) biologists with internal representations. The widespread phenomenon of the "body- clock" is often seen as involving internal representations of diurnal or seasonal rhythms. Maturana and Varela would say, instead, that it is an autopoietic phenomenon, enabling the organism to preserve its unity by making internal compensations for rhythmically varying conditions of light and temperature. If Maturana and Varela are niggardly in their use of informational language, they are prolific in their ascriptions of cogniton. Knowledge, for them, has no need of a nervous system. Human knowledge is more complex (self-reflexive) than the knowledge embodied in oak trees and amoebae. But all living things, without exception, are cognitive systems: "living as a process is a process of cognition" [op. cit., p. 13]. This is equivalent to saying that metabolism is not only a biochemical phenomenon, but a cognitive one, too. I demur. It is unnecessary and confusing to widen the scope of cognition so as to include all living things [Boden, forthcoming]. One can express the biologically important fact that different autopoietic systems have evolved closely coupled with different environments without using tendentious terms such as knowledge and cognition. For that reason, among others (not least the difficulty in interpreting Maturana and Varela's often tortuous prose), I prefer to think in terms of metabolism rather than autopoiesis. One thing, however, is clear. If life is defined in terms of metabolism (in the third sense), or in terms of autopoiesis, strong A-Life is out of the question. A final objection: Someone might protest that too much fuss is being made about metabolism. If artificial systems cannot metabolise, so what? Who's to say that metabolism really is essential to life? After all, we were prepared to consider adding evolution to the list of vital criteria, despite its decidedly counterintuitive air. Moreover, some A-Life workers hope to discover further properties to add to the list. For instance, Langton [1990, 1992] suggests that all living things satisfy a narrow range of numerical values of the "lambda parameter," a simple statistical measure of the degree of order and novelty in a system. It's not obvious that this sort of discovery is impossible. In short, the list of vital properties can change. So perhaps we can delete metabolism? We could retain a commitment to physicalism: no angels on pinheads allowed. And metabolism would still be recognized as a universal characteristic of the sort of (biological) life we happen to know about. But it would no longer be seen as essential. This would not be a sensible move. There are strong scientific reasons for opposing everyday intuitions about the non-criteriality of evolution. That's not to say that these reasons are strong enough. For Maturana and Varela, as we have seen, they are not. Moreover, respecters of the coherence of creationism, not to mention respecters of the concept of metabolism itself, may wish to continue taking individual creatures (not evolved species) as the paradigm of life. But there are no good reasons for rejecting our intuitions about the necessity of metabolism. On the contrary, scientific advance in biology and biochemistry reinforces our everyday assumption that metabolism is crucial -- indeed, inevitable. In sum, metabolism is necessary and strong A-Life is impossible. Understood in the third sense distinguished above, metabolism is essential to life. It is not merely part of our concept of life. It is also a necessary feature of self-organizing bodily creatures. There is no reason for removing metabolism from the list of vital properties, nor for weakening the term to mean mere energy dependency. Or rather, there is no reason independent of a (question-begging) "angels-on-pinheads" functionalist account of life in general, and of strong A-Life in particular. Acknowledgments: I am grateful to Graham Oddie and Michael Wheeler for helpful comments. REFERENCES Beaudoin, L. P. [1994] Goal-Processing in Autonomous Agents. PhD thesis, School of Computer Science, University of Birmingham. Birmingham, England. Bedau, M. A. [1996] "The Nature of Life." In M. A. Boden (ed.), The Philosophy of Artificial Life. Oxford: Oxford University Press, pp. 332-357. Beer, R. D. [1990] Intelligence as Adaptive Behavior: An Experiment in Computational Neuroethology. New York: Academic Press. Boden, M. A. [forthcoming] "Life and Cognition." In J. Branquinho (ed.), The Foundations of Cognitive Science at the End of the Century. Lisbon. Brooks, R. A. [1986] "A Robust Layered Control System for a Mobile Robot," IEEE Journal of Robotics and Automation, 2, 14-23. Brooks, R. A. [1991] "Intelligence Without Representation," Artificial Intelligence, 47, 139-159. Burks, A. W. [1966] Theory of Self-Reproducing Automata. Urbana: University of Illinois Press. Burks, A. W. [1970] Essays on Cellular Automata. Urbana: University of Illinois Press. Cliff, D. I. Harvey, & P. Husbands. [1993] "Explorations in Evolutionary Robotics," Adaptive Behavior, 2, 71-108. Doran, J. E. [1968] "Experiments With a Pleasure-Seeking Automaton." In D. Michie (ed.), Machine Intelligence III, Edinburgh: Edinburgh University Press, pp. 195-216. Drexler, K. E. [1989] "Biological and Nanomechanical Systems: Contrasts in Evolutionary Complexity." In C. G. Langton (ed.), Artificial Life. The Proceedings of an Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (held September 1987). Redwood City, CA: Addison-Wesley. Pp. 501-519. Goodwin, B. C. [1990] "Structuralism in Biology," Science Progress, 74, 227-44. Grand, S. D. Cliff, & A. Malhotra. [1996] Creatures: Artificial Life Autonomous Software Agents for Home Entertainment. Research report CSRP 434. Brighton: School of Cognitive and Computing Sciences, Sussex. Grey Walter, W. [1950] "An Imitation of Life," Scientific American, 182 (5), 42-45. Grey Walter, W. [1951] "A Machine that Learns," Scientific American, 185 (2), 60-63. Hubel, D. H., & T. N. Wiesel. [1959] "Receptive Fields of Single Neurones in the Cat's Striate Cortex," Journal of Physiology, 148, 579-591. Hubel, D. H., & T. N. Wiesel. [1962] "Receptive Fields, Binocular Interaction and Functional Architecture in the Cat's Visual Cortex," Journal of Physiology, 160, 106-154. Hubel, D. H., & T. N. Wiesel. [1968] "Receptive Fields and Functional Architecture of Monkey Striate Cortex," Journal of Physiology, 195, 215-243. Kauffman, S. A. [1992] The Origins of Order: Self-Organization and Selection in Evolution. Oxford: Oxford University Press. Langton, C. [1986] "Studying Artificial Life with Cellular Automata," Physica D, 22, 1120-49. Langton, C. G. [1989] "Artificial Life." In C. G. Langton (ed.), Artificial Life: The Proceedings of an Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (held September 1987). Redwood City, CA: Addison-Wesley. Pp. 1-47. (Reprinted, with revisions, in M. A. Boden (ed.), The Philosophy of Artificial Life, Oxford: Oxford University Press, pp. 39-94.) Langton, C. J. [1990] "Computation at the Edge of Chaos: Phase-Transitions and Emergent Computation," Physica D, 42, 12-37. Langton, C. G. [1992] "Life at the Edge of Chaos." In C. G. Langton, C. Taylor, J. D. Farmer, & S. Rasmussen (eds.), Artificial Life II. Redwood City, CA: Addison-Wesley. Pp. 41-91. Lettvin, J. Y., H. R. Maturana, W. Pitts, & W. S. McCulloch. [1959] "What the Frog's Eye Tells the Frog's Brain," Proceedings of the Institute of Radio Engineers, 47, 1940-1959. Levy, S. [1992] Artificial Life: The Quest for a New Creation. New York: Pantheon. McMullin, B., & F. J. Varela. [1997] "Rediscovering Computational Autopoiesis." In P. Husbands & I. Harvey (eds.), Fourth European Conference on Artificial Life. Cambridge, Mass.: MIT Press. Pp. 38-47. Maturana, H. R., & F. J. Varela. [1980] Autopoiesis and Cognition: The Realization of the Living. London: Reidel. (First published in Spanish in 1972.) Maynard Smith, J, & E. Szathmary. [1995] The Major Transitions in Evolution. Oxford: W. H. Freeman. Moran, F., A. Moreno, E, Minch, & F. Montero. [1996] "Further Steps Towards a Realistic Description of the Essence of Life," Proceedings of the Fourth International Conference on Artificial Life (May 1996). Nara, Japan. Pettigrew, J. D., & B. J. Frost. [1985] "A Tactile Fovea in the Scolopacidae," Brain, Behavior, and Evolution, 26, 185. Putnam, H. [1967/1975] "The Nature of Mental States." In H. Putnam, Mind, Language, and Reality: Philosophical Papers, II. Cambridge: Cambridge University Press. pp. 429-40. (First published in 1967.) Ray, T. S. [1992] "An Approach to the Synthesis of Life." In C. G. Langton, C. Taylor, J. D. Farmer, & S. Rasmussen (eds.), Artificial Life II. Redwood City, CA: Addison-Wesley. Pp. 371-408. (Reprinted in M. A. Boden (ed.), The Philosophy of Artificial Life, Oxford: Oxford University Press, pp. 111-145.) Ray, T. S. [1994] "An Evolutionary Approach to Synthetic Biology: Zen and the Art of Creating Life," Artificial Life, 1, 179-210. Selfridge, O. [1959] "Pandemonium: A Paradigm for Learning". In D. V. Blake & A. M. Uttley (eds.), Proceedings of the Symposium on Mechanization of Thought Processes. London: H. M. Stationery Office. Pp. 511-529. Sloman, A. [1990] "Motives, Mechanisms, and Emotions." In M. A. Boden (ed.), The Philosophy of Artificial Intelligence. Oxford: Oxford University Press, pp. 231-247. Thelen, E., & L. B. Smith. [1993] A Dynamic Systems Approach to the Development of Cognition and Action. Cambridge, Mass.: MIT Press. Varela, F. G., H. R. Maturana, & R. B. Uribe. [1974] "Autopoiesis: The Organization of Living Systems, Its Characterization and a Model," Biosystems, 5 (4), 187-196. Webster, G., & B. C. Goodwin. [1996] Form and Transformation: Generative and Relational Principles in Biology. Cambridge: Cambridge University Press. Wheeler, M. [1997] "Cognition's Coming Home." In P. Husbands & I. Harvey (eds.), Fourth European Conference on Artificial Life. Cambridge, Mass.: MIT Press. Pp. 10-19. Wiener, N. [1948] Cybernetics, or Control and Communication in the Animal and the Machine. New York: Wiley. Wright, I. P. [1997] Emotional Agents. PhD thesis, School of Computer Science, University of Birmingham. Birmingham, England. Wright, I. P., A, Sloman, & L. P. Beaudoin. [1996] "Towards a Design-Based Analysis of Emotional Episodes," Philosophy, Psychiatry, and Psychology, 3 (2), 101-137. Zeleny, M. [1977] "Self-organization of Living Systems: A Formal Model of Autopoiesis," International Journal of General Systems, 4, 13-22. Zeleny, M., G. J. Klir, & K. D. Hufford. [1989] "Precipitation Membranes, Osmotic Growths, and Synthetic Biology." In C. G. Langton (ed.), Artificial Life: The Proceedings of an Interdisciplinary Workshop on the Synthesis and Simulation of Living Systems (held September 1987). Redwood City, CA: Addison-Wesley. Pp. 125-139.