Copyright © 1999 Massachusetts Institute of Technology
Artificial life (A-Life) uses informational concepts and computer modeling to study life in general, and terrestrial life in particular. It aims to explain particular vital phenomena, ranging from the origin of biochemical metabolisms to the coevolution of behavioral strategies, and also the abstract properties of life as such ("life as it could be").
It is thus a form of mathematical biology -- albeit of a highly interdisciplinary type. Besides their presence in biology, especially ETHOLOGY and evolutionary theory, A-Life's research topics are studied also (for instance) in artificial intelligence, computational psychology, mathematics, physics, biochemistry, immunology, economics, philosophy, and anthropology.
A-Life was named by Christopher Langton in 1986 (Langton 1986 and 1989). Langton's term suggests (deliberately) that the aim of A-Life is to build new living things. However, not all A-Life scientists share this goal. Even fewer believe this could be done without providing some physical body and metabolism. Accordingly, some A-Life workers favor less philosophically provocative terms, such as "adaptive systems" or "animats" (real or simulated robots based on animals) (Meyer and Wilson 1991).
The claim that even virtual creatures in cyberspace could be genuinely alive is called strong A-Life, in analogy to strong AI. Most A-Lifers reject it (but see Langton 1989 and Ray 1994). Or rather, most reject the view that such creatures can be alive in just the same sense that biological organisms are, but allow that they are, or could be, alive to a lesser degree. Whether life does require material embodiment, and whether it is a matter of degree, are philosophically controversial questions. Proponents of autopoiesis (the continual self-production of an autonomous entity), for example, answer "Yes" to the first and "No" to the second (Maturana and Varela 1980). Others also answer the first question with a "Yes," but for different reasons (Harnad 1994). However, these philosophical questions do not need to be definitively answered for A-Life to progress, or be scientifically illuminating. Using artifacts to study life, even "life as it could be," is not the same as aiming to instantiate life artificially.
The theoretical focus of A-Life is the central feature of living things: self-organization. This involves the spontaneous EMERGENCE, and maintenance, of order out of an origin that is ordered to a lesser degree. (The lower level may, though need not, include random "noise.") Self-organization is not mere superficial change, but fundamental structural development. This development is spontaneous, or autonomous. That is, 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.
In SELF-ORGANIZING SYSTEMS, higher-level properties result from interactions between simpler ones. In living organisms, the relevant interactions include chemical diffusion, perception and communication, and processes of variation and natural selection. One core problem is the way in which self-organization and natural selection interact to produce biological order over time. Some work in A-Life suggests that whereas self-organization generates the fundamental order, natural selection (following on variation) weeds out the forms that are least well adapted to (least fit for) the environment in question (Kauffman 1993).
The higher-level properties in living organisms are very varied. They include universal characteristics of life (e.g., autonomy and evolution); distinct lifestyles (e.g., parasitism and symbiosis); particular behaviors (e.g., flocking, hunting, or evasion); widespread developmental processes (e.g., cell differentiation); and bodily morphology (e.g., branching patterns in plants, and the anatomy of sense organs or control mechanisms in animals).
A-Life studies all these biological phenomena on all these levels. A-Life simulations vary in their degree of abstractness or idealization. Some model specific behaviors or morphologies of particular living things, whereas others study very general questions, such as how different rates of mutation affect coevolution (Ray 1992). They vary also in their mode of modeling: some A-Life work concentrates on programs, displaying its creatures (if any) only as images on the VDU, while some builds (and/or evolves) physical robots. The wide range of A-Life research is exemplified in the journals Artificial Life and Adaptive Behavior, and in international (including European) conference proceedings of the same names. Brief overviews include Langton (1989) and Boden (1996, intro.). For popular introductions, see Emmeche (1994) and Levy (1992).
A-Life is closely related to -- indeed, it forms part of -- cognitive science in respect of its history, its methodology, and its philosophy.
Historically, it was pioneered (around the mid-twentieth century) by the founders of AI: Alan TURING and John VON NEUMANN. They both developed theoretical accounts of self-organization, showing how simple underlying processes could generate complex systems involving emergent order. Turing (1952) showed that interacting chemical diffusion gradients could produce higher-level (including periodic) structures from initially homogeneous tissue. Von Neumann, before the discovery of DNA or the genetic code, identified the abstract requirements for self-replication (Burks 1966). He even defined a universal replicator: a cellular automaton (CA) capable of copying any system, including itself. A CA is a computational "space" made up of many discrete cells; each cell can be in one of several states, and changes (or retains) its state according to specific -- typically localistic -- rules. Von Neumann also pointed out that copy errors could enable evolution, an idea that later led to the development of EVOLUTIONARY COMPUTATION (evolutionary programming, evolution strategies, genetic algorithms, etc.).
Even in relatively simple CAs, (some) high-level order may emerge only after many iterations of the relevant lower-level rules. Such cases require high-performance computing. Consequently, Turing's and von Neumann's A-Life ideas could be explored in depth only long after their deaths. Admittedly, CAs were studied by von Neumann's colleague Arthur Burks (1970) and his student John Holland, who pioneered genetic algorithms soon after CAs were defined (Holland 1975); and more people -- John Conway (Gardner 1970), Steve Wolfram (1983 and 1986), Stuart Kauffman (1969 and 1971), and Langton (1984), among others -- became interested in them soon afterward. But these early studies focused on theory rather than implementation. Moreover, they were unknown to most researchers in cognitive science. The field of A-Life achieved visibility in the early 1990s, largely thanks to Langton's initiative in organizing the first workshop on A-Life (in Los Alamos) in 1987.
Methodologically, A-Life shares its reliance on computer modeling with computational psychology and AI -- especially connectionism, situated robotics, and genetic algorithms (evolutionary programming). These three AI approaches may be integrated in virtual or physical systems. For instance, some A-Life robots are controlled by evolved NEURAL NETWORKS, whose (initially random) connections specify "reflex" responses to specific environmental cues (e.g., Cliff, Harvey, and Husbands 1993).
A-Life's methodology differs from classical (symbolic) AI in many ways. It relies on bottom-up (not top-down) processing, local (not global) control, simple (not complex) rules, and emergent (not preprogrammed) behavior. Often, it models evolving or coevolving populations involving many thousands of individuals. It commonly attempts to model an entire creature, rather than some isolated module such as vision or problem-solving (e.g., Beer 1990). And it claims to avoid methods involving KNOWLEDGE REPRESENTATION and PLANNING, which play a crucial role in classical AI (Brooks 1991). The behavior of A-Life robots is the result of automatic responses to the contingencies of the environment, not preprogrammed sequences or internal plans. Each response typically involves only one body part (e.g., the third leg on the right), but their interaction generates "wholistic" behavior: the robot climbs the step, or follows the wall.
Philosophically, A-Life and AI are closely related. Indeed, if intelligence can emerge only in living things, then AI is in principle a subarea of A-Life. Nevertheless, some philosophical assumptions typical of classical AI are queried, even rejected, by most workers in A-Life. All the philosophical issues listed below are discussed in Boden 1996, especially the chapters by Bedau, Boden, Clark, Godfrey-Smith, Hendriks-Jansen, Langton, Pattee, Sober, and Wheeler; see also Clark (1997).
Much as AI highlights the problematic concept of intelligence, A-Life highlights the concept of life -- for which no universally agreed definition exists. It also raises questions of "simulation versus realization" similar to those concerning strong AI. Problems in A-Life that are relevant also to the adequacy of FUNCTIONALISM as a philosophy for AI and cognitive science include the role of embodiment and/or environmental embeddedness in grounding cognition and INTENTIONALITY.
A-Life in general favors explanations in terms of emergence, whereas AI tends to favor explanation by functional decomposition. Moreover, many A-Life researchers seek explanations in terms of closely coupled dynamical systems, described by phase-space trajectories and differential equations rather than computation over representations. Although A-Life does avoid the detailed, "objective," world-modeling typical of classical AI, whether it manages to avoid internal representations entirely is disputed. Also in dispute is whether the "autonomy" of environmentally embedded A-Life systems can capture the hierarchical order and self-reflexiveness found in some human action (and partly modeled by classical AI). Many philosophers of A-Life justify their rejection of representations by criticizing the broadly Cartesian assumptions typical of classical, and most connectionist, AI. They draw instead on philosophical insights drawn from Continental philosophy, or phenomenology, sometimes using the concept of autopoiesis.
Besides its theoretical interest, A-Life has many technological applications. These include evolutionary computation for commercial problem solving, environmentally embedded robots for practical use, and computer animation for movies and computer games. The "Creatures" computer environment, for example, employs A-Life techniques to evolve individual creatures capable of interacting, and of learning from their "world" and the human user's "teaching."
Beer, R. D. (1990). Intelligence as Adaptive Behavior: An Experiment in Computational Neuroethology. New York: Academic Press.
Boden, M. A., Ed. (1996). The Philosophy of Artificial Life. Oxford: Oxford University Press.
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.
Clark, A. J. (1997). Being There: Putting Brain, Body, and World Together Again. Cambridge, MA: MIT Press.
Cliff, D., I. Harvey, and P. Husbands. (1993). Explorations in evolutionary robotics. Adaptive Behavior 2:71-108.
Emmeche, C. (1994). The Garden in the Machine: The Emerging Science of Artificial Life. Princeton: Princeton University Press.
Gardner, M. (1970). The fantastic combinations of John Conway"s new solitaire game "Life." Scientific American 223(4):120-123.
Harnad, S. (1994). Levels of functional equivalence in reverse bioengineering. Artificial life 1:293-301.
Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. Ann Arbor: University of Michigan Press.
Kauffman, S. A. (1969). Metabolic stability and epigenesis in randomly connected nets. Journal of Theoretical Biology 22:437-467.
Kauffman, S. A. (1971). Cellular homeostasis, epigenesis, and replication in randomly aggregated macro-molecular systems. Journal of Cybernetics 1:71-96.
Kauffman, S. A. (1992). The Origins of Order: Self-Organization and Selection in Evolution. Oxford: Oxford University Press.
Langton, C. G. (1984). Self-reproduction in cellular automata. Physica D 10:135-144.
Langton, C. G. (1986). Studying artificial life with cellular automata. Physica D 22:1120-1149.
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.)
Levy, S. (1992). Artificial Life: The Quest for a New Creation. New York: Pantheon.
Maturana, H. R., and F. J. Varela. (1980). Autopoiesis and Cognition: The Realization of the Living. London: Reidel.
Meyer, J.-A., and S. W. Wilson, Eds. (1991). From Animals to Animats: Proceedings of the First International Conference on Simulation of Adaptive Behavior. Cambridge, MA: MIT Press.
Ray, T. S. (1992). An approach to the synthesis of life. In C. G. Langton, C. Taylor, J. D. Farmer, and 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.
Turing, A. M. (1952). The chemical basis of morphogenesis. Philosophical Transactions of the Royal Society: B 237:37-72.
Wolfram, S. (1983). Statistical mechanics of cellular automata. Review of Modern Physics 55:601-644.
Wolfram, S. (1986). Theory and Applications of Cellular Autom ata. Singapore: World Scientific.
Copyright © 1999 Massachusetts Institute of Technology