Introduction

Artificial life (Alife) is a field of study devoted to understanding life by attempting to derive general theories underlying biological phenomena, and recreating these dynamics in other physical media - such as computers - making them accessible to new kinds of experimental manipulation and testing. This scientific research links biology and computer science. Artificial life is an alternative life-forms - literally "life made by Man rather than by Nature." Artificial cells are used rather than living cells.

While biological research is essentially analytic, attempting to break down complex phenomena into their basic components, Alife is synthetic, trying to construct phenomena from their elemental units - put together systems that behave like living organisms. In addition to providing new ways for studying biological phenomena associated with life on Earth, life-as-we-know-it, Alife lets us extend our studies to the larger domain of "biologically" possible life, life-as-it-could-be[Langton 1992].

The Importance of Evolution

Evolution is central to Alife research. It offers the possibility of adaptation to a dynamic environment - when an unforeseen event occurs, the system can evolve, in analogy to nature. Genetic programming has been successfully applied by Koza to tackle several problems [Koza 1992]. Genetic programming is now widely recognized as an effective search paradigm in artificial intelligence, databases, classification, robotics and many other areas, such as financial forecasting[Chopard 1996] and image discrimination [Tackett 1993].

An evolutionary method is advantageous not only in solving difficult problems but also in offering better adaptability. Current computer programs nowadays are well known for their "brittleness" - in some unanticipated situations, the programs tend to fail, and the results are unpredictable, may even be disastrous. This is one of the major causes of high software development and maintenance costs.

The Importance of Emergence

Emergence is a process where phenomena at a certain level arise from interactions at lower levels. This emergent properties is created when something becomes more than the sum of its parts. Alife systems consist of a large collection of simple, basic units whose interesting properties emerge at higher levels. Examples are von Neumann's universal constructing machine and Craig Reynolds' work on flocking behaviour[Reynolds 1987]. Reynolds' algorithm has been used to produce photorealistic imagery of bat swarms for the feature motion pictures Batman Returns and Cliffhanger.

Differences between Artificial Intelligence and Alife

Brooks' method [Brooks 1991] for building sophisticated robots demonstrates the Alife approach, which is basically different from that of traditional artificial intelligence (AI). AI employs a top-down methodology, where complex behaviours (for example, chess playing) are identified, to build a system that fulfills all the specifications. Alife operates in a bottom-up manner, starting from simple elemental units, gradually building its way upwards through evolution, emergence, and development.

Moreover, AI has traditionally concentrated on complex human functions, such as chess playing, text comprehension, medical diagnosis. Alife focuses on basic natural behaviours, emphasizing survivability in complex, dynamic environments.

Genetic Algorithms and Artificial Life

Genetic algorithms (GAs) are currently the most prominent and widely used computational models of evolution in artificial-life systems. Overview of GAs is found in the article of my partner, Hiu Man. GAs have been used both as tools for solving practical problems and as scientific models of evolutionary processes.

However, it can be very difficult to relate the behaviour of a simulation quantitatively to the behaviour of the given system. This is because the level at which artificial-life models are constructed is often so abstract that they are unlikely to make numerical predictions. In GAs, all of the biophysical details of transcription, protein synthesis, gene expression, and meiosis have been stripped away. Nevertheless, useful Alife models may well reveal general conditions under which certain qualitative behaviours arise, or critical parameters in which a small change can have a drastic effect on the behaviour of the system.

Applications of Artificial life

Artificial life, which attempts to explain existing life and recreate biological phenomena in alternative media, results in both better theoretical understanding of the phenomena under study, and practical applications of biological principles in the technology of computer hardware and software, synthetic chemistry to model new compounds, IBM "immune system" to protect computers from viruses mobile robots [Brooks 1986, JMD 1994], spacecraft, medicine, anotechnology, industrial fabrication and assembly, and other vital engineering projects.

Learning

There are extensive interconnections between the fields of neural networks and artificial intelligence [FH 1987, Watson 1991, Dorffner 1994]. Learning in a multi-agent setting of Alife provides numerous challenges for theories originally developed to explain learning in isolated individuals [LS 1987, SW 1989, Grefenstette 1991].

The Alife systems can be designed to model interactions between learning and evolution [AL 1992, HN 1987]. Biological phenomena can be studied with controlled computational experiments whose natural equivalent (for example, running for thousands of generations) is not possible or practical. Moreover, when performed correctly, these experiments can provide new insight into these natural phenomena. The potential benefits of the work are not limited to understanding natural phenomena. A GA researchers are studying ways to apply GAs to optimize neural networks to solve practical problems [SWE 1992].

Virtual Reality

Virtual Reality (VR) is a burgeoning field of computer science with widespread practical applications and tight connections with artificial life [BC 1993]. Both VR and Alife practitioners seek to use the computer to represent life-like processes operating in artificial, but life- like worlds. There are marked differences between the two fields: the user of a VR simulator is often involved in the activities of the artificial world, while this is seldom the case in a Alife simulator. Verisimilitude is typically found in a VR simulator, while Alife simulators typically allow for radical departures from natural rules.

Ecosystems and Evolutionary Dynamics

Alife is used as ecological monitoring tools, modelling ecosystem behaviour and the evolutionary dynamics of populations. Echo system allows a large range of ecological interactions, and Strategic Bugs system measures evolutionary activity.

Echo system models ecologies in the same sense that the GA models population genetics [Holland 1993]. It abstracts away virtually all of the physical details of real ecological systems and concentrates on a small set of primitive agent-agent and agent-environment interactions. The goal of Echo is to study how simple interactions among simple agents (creatures) lead to emergent high-level phenomena, such as the flow of resources in a system or cooperation and competition among agents.

The Strategic Bugs world is a two-dimensional lattice, containing only adaptive agents ("bugs") and food [BP 1992]. The evolutionary activity is defined and measured.

Robotics/Psychology

The study of action selection [Maes 1990, Maes 1991], the mechanisms by which an organism (real or artificial) selects which among a variety of (often mutually incompatible) behaviours to execute at a given moment, has practical implications for robots, as well as providing an experimental platform for the evaluation of psychological theories.

In traditional robotics, programmers tried to anticipate and explicitly control every aspect of the action of the robot. These control systems tend to fail when an unforeseen event occurs. By contrast, decentralized, adaptive control of robot motion in Alife is achieved through the robotic controllers which continuously learn and adapt to changing environments [Brooks 1989, Dorigo 1995]. The main idea is that intelligent autonomous agents cannot be built but should evolve in a process similar to the way that intelligence evolved in nature: using a combination of evolution by natural selection, adaptivity and development [Steels 1995].

Traffic Control

A typical application for distributed artificial intelligence is found in the control of traffic. The traffic control may concern physical vehicles [MTD 1993], or simply the flow of informations packets in a network [FD 1993a, FD 1993b].

Air traffic control, in particular, has been intensively studied [Ndovie 1993]. It is focused on cooperation among air traffic controllers themselves, and between the controllers and aircraft. Smooth cooperation is required in order to achieve a safe, orderly, prompt and efficient movement of traffic in airspace.

Intelligent Manufacturing

Distributed systems of agents take over monolithic, centralized control mechanisms. In this approach, each machine or process has an antonomous-agent controller. The agent monitors the state of its machine, tries to satisfy its needs in terms of raw material etc., possibly competing with other agents for resources [KD 1993].

Education

Artificial intelligence has been used extensively for computer-aided instruction. A number of Alife simulators have been developed to teach biology, especially to children [RM 1990, Resnick 1994]. Some of these programs are commercially available as educational games, such as SimLife, for learning about artificial life, the management of resources and so on.

Computer Viruses

A computer virus can be viewed as a kind of Alife [Spafford 1991] . It is a computer program which attempts to satisfy its purpose without human intervention. Typically, its aim is to reproduce and spread copies of themselves to many computers through communication links or disk exchange. These viruses often integrate their code directly into that of other programs, such that execution of the host program causes execution of the viral program.

Viruses often have destructive effects on their host computers, for example in personal computers. However, viruses may be designed to have constructive effects. For example a viral program, which seek out and destroy anomalies in a database, would be useful for maintaining the integrity of the database. Viruses hold particular interest for Alife since they have properties very similar to those of biological viruses.

Virtual World

An open-ended evolution can be constructed within a computer, proceeding without any human guidance. This virtual world is achieved by the Alife simulators. The ideal general-purpose Alife simulator would allow the user to choose from a variety of fundamental algorithms (neural networks, evolutionary algorithms, cellular automata), to easily design populations of creatures, to easily collect and analyze data. Although this ideal simulator is not existed, the closest match is the Swarm Simulation System.

A virtual world, called Tierra, can undergo evolution. The Tierra creatures (programs) compete for the natural resources of their computerized environment, namely CPU time and memory. The virtual world's natural resources are limited, as in nature, serving for competition between creatures. Tierra is used for the study of the evolution of artificial organisms. Other Alife simulators can be found in the Artificial-Life Simulators and their Applications.

Let's have a look at

References

[AL 1992] In C.G.Langton, C.Taylor, J.D.Farmer, and S.Rasmussen, editors, Artificial Life II, (Interactions between learning and evolution) pages 487-507, by D.H.Ackley and M.L.Littman, in 1992.
[BC 1993] The VEOS project, Technical report, Human Interface Technology Laboratory, University of Washington, (Connections between Virtual Reality and Alife), by William Bricken and Geoffrey Coco, in 1993.
[BP 1992] In C.G.Langton, C.Taylor, J.D.Farmer, and S.Rasmussen, editors, Artificial Life II, (Measurement of evolutionary activity, teleology, and life) pages 431-461, by M.A.Dedau and N.H.Packard, in 1992.
[Brooks 1986] In IEEE Journal of Robotics and Automation , (A robust layered control system for a mobile robot) pages 14-23, by Rodney A.Brooks, in April 1986.
[Brooks 1989] Neural Computation, (A robot that walks: Emergent behaviour from a carefully evolved network), by Rodney A.Brooks, in 1989.
[Chopard 1996] Parallel Genetic Programming: an application to Trading Models Evolution, by B.Chopard, M.Oussaidene, M.Tomassini and O.Pictet, in 1996.
[Brooks 1991] Science, (New approaches to robotics) pages 1227-1232, by Rodney A.Brooks, in September 1991.
[Dorffner 1994] Neural Networks and a New AI, by Georg Dorffner, in 1994.
[Dorigo 1995] Special issues of IEEE Transactions on Systems, Man and Cybernetics (IEEE-SMC), (Learning Approaches to Autonomous Robots Control), by Marco Dorigo, in 1995.
[FD 1993a] In S.M.Deen, editor, Proceedings of the CKBS-SIG Workshop 1992, (Design considerations for optimal intelligent network routing) pages 19-42, by Martyn Fletcher and S.M.Deen, in 1993.
[FD 1993b] Technical report no. DAKE/-/TR93010.0, Data and Knowledge Engineering Centre, Keele University, (Multi-agent design issues in congestion management, by Martyn Fletcher and S.M.Deen, in September 1993.
[FH 1987] IEEE Computer, (Connectionist architectures for artificial intelligence) pages 100-109, by Scott Fahlman and Geoffrey Hinton, in January 1987.
[Grefenstette 1991] In Proceedings 1991 Conference on Genetic Algorithms, (Lamarckian learning in multi-agent environments) pages 303-310, by J.J.Grefenstette, in 1991.
[HN 1987] Complex Systems, (How learning can guide evolution) pages 495-502, by G.E.Hinton and S.J.Nowlan, in 1987.
[Holland 1993] Technical report 93-04-023, Santa Fe Institute, (Echoing emergence: Objectives, rough definitions, and speculations for Echo-class models, by J.H.Holland, in 1993.
[JMD 1994] In Rodney Brooks and Pattie Maes, editors, Artificial Life IV, (Evolving visual routines) pages 198-209, MIT Press, by Michael Patrick Johnson, Pattie Maes, and Trevor Darrell, in 1994.
[KD 1993] In Proceedings of the IEEE Systems, Man and Cybernetics Conference, (A cooperative search scheme for dynamic problems), by Y.Kitamura et al and S.M.Deen, in 1993.
[Koza 1992] Genetic Programming, by John R.Koza, in 1992.
[Langton 1992] In C.G.Langton, C.Taylor, J.D.Farmer, and S.Rasmussen, editors, Artificial Life II, volume X of SFI Studies in the Sciences of Complexity, (Preface) pages xiii-xviii, by C.G.Langton, C.Taylor, J.D.Farmer, and S.Rasmussen, in 1992.
[LS 1987] In Proceedings IEEE International Symposium on Intelligent Control, (Multiple agent cooperative problem solving with axiom-based negotiation), by S.Lee and Y.G.Shin, in 1987.
[Maes 1990] Connection Science, (How to do the right thing) pages 291-323, by Pattie Maes, in 1990.
[Maes 1991] Designing Autonomous Agents: Theory and Practice from Biology to Engineering and Back, by Pattie Maes, in 1991.
[Ndovie 1993] Multi-agent cooperation in air traffic control , by Baird Ndovie, in 1993.
[Reynolds 1987] Computer Graphics, (Flocks, herds, and schools: A distributed behavioural model) pages 25-34, by C.W.Reynolds, in July 1987.
[Resnick 1994] Artificial Life, (Learning about life) , by Mitchel Resnick, in 1994.
[RM 1990] E&L memo 10, Discussion of Alife in education , (Children and artificial life), by M.Resnick and F.Martin, in November 1990.
[Spafford 1991] In C.Langton, C.Taylor, J.D.Farmer, and Rasmussen, editors, Artificial Life II, volume XI, (Computer viruses - a form of artificial life) pages 371-408, by Eugene H.Spafford, in 1991.
[Steels 1995] The Homo Cyber Sapiens, the Robot Homonidus Intelligens, and the 'artificial life' approach to artificial intelligence, by Luc Steels, in January 1995.
[SW 1989] In Distributed Artificial Intelligence 2, (Learning and adaptation in distributed artificial intelligence systems), by M.J.Shaw and A.B.Whinston, in 1989.
[SWE 1992] In L.D.Whitley and J.D.Schaffer, editors, International Workshop on Combinations of Genetic Algorithms and Neural Networks, (Combinations of genetic algorithms and neural networks: A survey of the state of the art) pages 1-37, by J.D.Schaffer, L.D.Whitley and L.J.Eshelman, in 1992.
[Tackett 1993] In Stephanie Forrest, editor, Proceedings of the Fifth International Conference on Genetic Algorithms, ( Genetic programming for feature discovery and image discrimination , by Walter A. Tackett, in 1993.
[Watson 1991] Common Lisp Modules - Artificial Intelligence in the Era of Neural Networks and Chaos Theory, by Mark Watson, in 1991.

Back to my Homepage