Go to Semiotic Review of Books Home Page
Go to SRB Highlights
Go to SRB Archives
This editorial appeared in Volume 3 (2) of The Semiotic Review of Books.
One of my less imaginative colleagues recently remarked that he had "never come across any new questions in philosophy." But that just shows he reads no science. This old sci-fi chestnut is now both science and philosophy: If we found life on another planet, how would we know it?
A science fiction fantasy raises a real new philosophical question, when speculation about it, however inconclusive, is serious and informed. Earth bound life is, perforce, what our biologists know. Can we make sense of some deeper notion of life that would recognize something as living even though it were made of stuff wholly different from our own? That is the project which loosely links the exploding interdisciplinary web of questions known as Artificial Life.
When art recapitulates life, it usually does it backwards. Thus, in Artificial Intelligence, we first made machines do what we humans had most lately mastered: make inferences on the basis of syntactic form. We still haven't managed to mechanize what our older ancestors knew long ago: how to walk, jump or see. Again, intelligence presumably added a late bloom to an already teeming variety of living forms. But it is only five years now, after a long half century of Artificial Intelligence research, that Artificial Life got its first Conference. Others are following at a quickening pace. (Details can be obtained without human intervention from a computer by electronic mail.)
Like the ambitions of AI those of AL are either weak or strong (Sober 1992). The weaker aspiration is merely to study the with models and simulations; the strong is actually to create life -- but in a different matter. The word 'artificial' fits equally well either way: artificial flowers are not flowers, but artificial light is real light.
A simulation is not the thing it simulates. Rainstorms simulated in a computer are really neither wet nor windy. How then could there be real life or intelligence in a computer simulation?
Alan Turing, father of Artificial Intelligence, devised a famous test forascribing intelligence to a machine: see if it can fool someone long enough into thinking it is human. Many people are sceptical of this test: a machine might so behave as to convince observers that it was really thinking, when in fact its responses were either "canned or just the outcome of purely syntactic processes lacking "intrinsic intentionality": whatever it did the machine would never mean anything (Searle 1980) But this problem does not seem seriously to threaten AL. If we could agree on an objective test for the, it is hard to imagine someone seriously complaining that something might pass it and yet lack essential 'vitality'.
So here AL seems better off than AI. On the other hand, there is one possible way out of the problem for AI that is not obviously sailable for AL. It consists in claiming that the matter of thought is information. If that is so, then information, rightly manipulated by an artificial system, might qualdy as genuinely intelligent. Is the same comfort available to AL? It is not obviously plausible to claim that the stuff of live is inflation. Yet this is, in a way, just what the proponents of "strong AL" are boldly endeavouring to establish. In a way undreamt of by himself, prehaps Pythagoras was right after all: the world is made of numbers. Something like that is the claim of the bolder advocates of "strong" AL.
Before we take up such breathtaking claims, however, it must be acknowledged that what goes on at Artificial Life conferences is pretty much business as usual for a lot of practitioners of biological or Artificial Intelligence modelling. To call this weak AL is not to slight it: it is vital indeed. A rich variety of sophisticated quantitative models is on offer, illuminating the mechanisms of selection, evolution, embryological development, genetics, and selforganization or "autopoiesis". In this domain, there is no sharp line between Artificial Life and Artificial Intelligence. And that's as it should be, surely, for though ants and even amoebas aren't intelligent in the full, Semiotic Review sense of the word, they are still too smart in their own way to be replaced by any machine in the wild. To make this much life, then, we already need to know how to make a fair bit of intelligence. Moreover, what vague boundary there is gets straddled by the use of biological models in aid of non-biological ends. An example is the work of Daniel Hillis, in which sorting algorithms were set to evolve by natural selection by being pitted against also evolving candidates for sorting. There is no thought here of emulating life itself, but only of enlisting some of its mechanisms in the service of other ends.
But how can a computer model even illuminate life, let alone produce it?
All modelling, of course, is intellectually dangerous. To make a model is to abstract some features from reality, and incorporate them into a structure some features of which, in turn, must be ignored. When the Planetarium shows you a model of the solar system, you don't need to look at it through UV filter glasses; nor, on the other hand, are you induced to think that the real Solar System is driven by an electric motor. How then, ask critics who march under the banner of Holism, can we model things biological without distorting them to the point of uselessness?
To those committed to the ineffability of their favourite mysteries, this objection will no doubt be reason enough to ignore the whole field. Among the rest of us, some who have a taste for sadistic irony we wil wish upon AL's holy naysayers an instructively destructive encounter with a real five computer virus. As self reproducing, sometimes evolving, parasitical entities, viruses are the strongest candidates so far for the status of artificial living thing.
But actually one can make a more constructive point: for some of the models studied by practioners of Artificial Life enable us to move beyond the silly squabbles of reductionism and holism.
The claim to transcend an ancient quarrel all too often turns out to be lust a bid by one side to preempt further debate in our own favour But in this case, I think the claim has all the more merd for the fact that the AL practioners themselves have generally failed to make R. Let me explain.
Among AL's favourite toys are cellular automata: collections of cells in a visual space optically represented as a grid on a computer screen), each of which can be in any of a finite number of states. The state of each cell is revised according to rules that refer exclusively to its own state and the state of its neighbours. In this sense, all effective causality resides entirely at the local level. Some of their effects, however some of the patterns and behaviour that result from the sum of those interactions -- look remarkably like globally designed ones. Thus Langton and others often speak of those effects as "emergent" but in fact the term is not used here in anything like the sense it used to have in arguments for the autonomy of biology or mentality. That sense implied that information about parts and their organization could never suffice to predict the behaviour of the total configuration on as a whole. In these models, by contrast we know from the start that the parts and their organization are sufficient to make such predictions purely mechanically, since making such purely mechanical predictions is all that computer are empowered to do.
Actually, the idea of global systems locally controlled is not new. Adam Smith's "invisible hand", Darwinian evolution, and more recently connectionist systems are all instances of such systems. But Cellular Automata now show us especially clearly that the issue between holism and reductionism was a false problem. There were things we couldn't predict about whole systems but only because we didn't have computers to work it out fast enough: so we got a surprise. That, it turns out, is all that Langton et al. mean by 'emergence'. Emergence in his sense is hoist by the petard of reductionism.
But what of the strong claim? Is it possible not merely to define life beyond the earthbound, but actually to make a specimen? Some say we can, and suggest that the world in which such specimens will live is a world of information structures. Even more boldly, Langton seems to suggest that our own world of matter and the world of information are, if not identical, at least both structured by the same deeper laws.
We may be softened up for that bold claim by remembering how the notion of entropy links physics and information. In both physics and information theory, entropy is a measure of disorder The connection is quite concrete. Disorder is more probable than order, because there are more permutations of elements that count as "random" than ones that count as "organized". Toss a coin a dozen times: there is only one way to get a straight string of Heads, compared to 685,280 ways of getting six of each. This translates directly into the physical meaning of entropy, as a state in which no usable energy exists. Picture a gas in a vessel with two chambers of equal sim. If most of the molecules are in one chamber, then any molecule taken at random is more likely to flow from the fuller into the emptier chamber and that Now might be harnessed as energy. The rate at which at this takes place depends on how active the particles are -- that is, on the gas's temperature.
Langton's computer studies have revealed a fascinating analogue to temperature and entropy in the world of cellular automata. Using a measure of the rate of activity of his cellular "molecules", based on the quantity of local change introduced by the rules governing local actively, Langton discovered that the consequences of all possible rules divide into three types, matching the three fundamental states of ordinary matter. In the 'periodic' or 'solid' state, repeated application of low-activity rules results in rapid stabilization into unchanging or cyclically repeating patterns. In the 'chaotic' or 'gaseous' state, the actively of the system looks like the chaotic behaviour of the molecules of a gas. In other words, nothing happens that is of much interest in either extreme state. The transitional region, corresponding to computations that are typically undecidable, is the one where we find complex and meaningful patterns: patterns that look, in short, like those of life as we know it.
Langton's claim could hardly be more ambitious. If he is right, he has uncovered the ultimate structure of a deep level of really, more fundamental than either information or matter, but underlying the essential characteristics of both.
Does this amount to showing that life, in a sense worthy of that name, can be sustained in a "programmable matter" which is essentially abstract mathematical pattern?
There is room for doubt. Patterns, after all, are only patterns: Platonic objects that need spatiotemporal niches H they are properly to come to life. For us living things, the flesh is all; or at least, there is no the without individual, spatiotemporal perishable flesh. Langton may be fight, that the level of analysis he has uncovered is indeed the most profound there is. But might not the only the worthy of the name be, in a sense that Oscar Wilde didn't anticipate, essentially superficial?
Here for the benefit of my jaded colleague, may be genuinely new philosohpical questions.
Bedau, M.A. (1992) Philosophical aspects of arfificial life. In P. Bourgine & F. Varela (1992), (pp. 10.
Bourgine, P. and Varela, F. (Ed.) The First European Conference on Arrtificial Life Proceedings. Cambridge, MA: MIT Press . A Bradford Book.
Hillis, W.D. (1992). Co-evolving parasites improve simulated evolution as an optimization procedure. In C.G. Langton et al (Eds.) Artificial Life II (pp. 313-24).
Langton, C.G.(l989) Studying arfificial life with cellular automata. Physica D (1-2) 120-49.
--- (Ed.) (1989). Artificial Life (Proceedings of an interdiscipfinaw Workshop on the Synthesis and Simulation of Living Systems Held September 1987 in Los Alamos, NM Santa Fe Institute Studies in the sciences of complexity. Aedwood City, CA: AddisonWesley.
--- (Ed). (1992) Artificial Life II (Proceedings of the Workshop on Artificial Life Held February, 1990, in Santa Fe, NM) Santa Fe lnstitute Studies in the sciences of complexity. Redwood City, CA: Addison-Wesley.
--- (Ed.) (1992) Life on the edge of chaos. In Artificial Life II Langton et. al., (Eds.) (pp. 41-92)
Searle, J.R. (1980). "Minds, Brains and Programs." Behavioral and Brain Sciences 3, 417-67.
Sober E. (1992). Learning from functionalism prospects for strong arfificial life. In Langton et al, (Eds) Artificial Life II (pp.749-97).
Turing, A.M. (1950) Computing Machinery and Intelligence. Mind, 59, 533-60.
1. Information can be obtained electronically from: email@example.com from which one can learn of over a dozen forthcoming conferences from Brussels to Hawaii. For the various conference proceedings, see Bourgine and Varela; Langton (1986; 1989;1992). For a good general introduction to the philosophically interesting issues, see Bedau.