Dateline: 03/27/00
Artificial Life, or ALife, is a rapidly growing scientific field at the intersection of biology and computer science. While it promises to illuminate some fundamental biological questions, it could also provide a means for addressing complex engineering problems that were previously beyond our reach. Proponents of "strong" ALife would even claim that it will fulfill the dream of every AI researcher: a computer that can learn on its own.
Artificial Life simulators can be described as virtual worlds exhibiting the emergence of life-like behaviors. Such simulators offer more possibilities than standard Genetic Algorithms, in which ALife "organisms" have no interaction with each other. For casual users, they are just cool things to watch. However, many of these systems are much more than that - serious research tools with many possible applications. Let's explore the most popular among them, starting with the simple on-line simulations and continuing with the advanced systems for exploring digital evolution.
Boids
Boids are computer models of coordinated animal motion such as bird flocks and fish schools, first created by Craig Raynolds. Each "creature", or boid, has direct access to the whole scene's geometric description, but reacts only to flockmates within a certain small radius of itself. The basic flocking model consists of only three simple steering behaviors: separation, alignment and cohesion. This system is important and often cited example of principles of Artificial Life: flocking is a particularly evocative example of emergence, where complex global behavior arise from the interaction of simple local rules. There have been other applications of the Boids model in the realm of Behavioral Animation (used in movies Batman Returns and Cliffhanger). While we are at that topic, you may also visit Craig's excellent pages on Evolutionary Computation and its application to art and design.
Java Floys belong to the flocking ALife creatures variety, sharing with them the social tendency to stick together, and the lifelike emergent behavior which is based on a few simple, local rules. They differ from most other ALife flocking implementations by being territorial animals that defend their territory against intruders.
CoreLife
CoreLife is another artificial-life environment based on - and compatible with - Tierra, which displays open-ended evolution using a world of assembly-language creatures running on simulated parallel processors. The ease of use and a full-featured context-sensitive help system are its main advantages - it can run on any old MS-DOS PC.
Gene Pool and Darwin Pond
Gene Pool is an artificial life simulation where populations of physics-based organisms evolve over time. It is best described in the words of its author as a virtual Darwinian aquarium in which you initiate a primordial soup, and then occasionally check up on what Virtual Mother Nature is up to. You can interact with the creatures in basic ways, such as moving creatures you want to mate close to one another. Another software from the same author, Darwin Pond, has more advanced user interface and more neat features, but lacks the degree of flexibility in the simulation Gene Pool has. These applications may look like games, but they are really some of the best ALife demonstrations on a PC platform.
Swarm
Swarm is a software package for multi-agent simulation of complex systems, originally developed at the Santa Fe Institute. Swarm is intended to be a useful tool for researchers in a variety of disciplines. The basic architecture of Swarm is the simulation of collections of concurrently interacting agents: with this architecture, users can implement a large variety of agent based models. Several sample applications are available, including a synchronous, time-stepped model of local agents interacting through the environment to create a complex global behaviour and a simple discrete event simulation of fission. It runs under various flavours of Unix and Windows 95/98/2000 and NT.
Gecko
Gecko is a spatial individual-based simulator for modeling ecosystem dynamics. Individual agents are modeled as free-moving spheres. Within the overall Gecko framework, many different ecosystem models (2D and 3D) can be realized: these include GrassWorld, ScaredyWorld, Woods and BacSim. The latest version of Gecko is written in Java, built on CourseWare, a Java 1.1 toolkit for Web-based interactive simulators. Gecko can run either over the Web, or as a local graphical or batch application. New simulator models are written by adding or extending a few Java agent and landscape classes.
Echo
Echo is a simulation tool developed to investigate mechanisms which regulate diversity and information-processing in systems comprised of many interacting adaptive agents, or complex adaptive systems (CAS). Echo agents interact via combat, trade and mating and develop strategies to ensure survival in resource-limited environments. The existing implementation of Echo runs on most Unix and Linux systems.
Avida
Avida group's main research tool is inspired by Tierra, implementing self-reproduction combined with a versatile genetic basis. The population adapts to the combination of an intrinsic fitness landscape (self-reproduction) and an user-provided (extrinsic) fitness function, so evolutionary adaptation, general traits of living systems, and other issues pertaining to theoretical or evolutionary biology and dynamic systems can be studied in detail. The system creates a virtual environment consisting of a 2D grid of virtual processors which execute a limited assembly language that supports self-reproduction. Programs are stored as sequential strings of instructions in the system memory, and every program is associated with a processor. The virtual environment is initially seeded with a human-designed program that self-replicates. This program and its descendants are then subjected to random mutations of various types which change instructions within their memory: any mutation which results in an increased ability to reproduce in the given environment is considered favorable. Each "creature" can interact with others on a local scale, as opposed to the global interactions in Tierra. Other features include detailed flexibility in configuring evolutionary experiments (time-slicing methods, mutation schemes, methods for placement of newborn creatures...) and the ability for precise measurements and the collection of statistical information. Currently, Avida has been compiled on most Unix platforms, Windows 95 and NT.
LEE
The Latent Energy Environments (LEE) package of Fillipo Menzcer and Rik Belew is an Artificial Life model and simulator of controlled complexity, using endogenous fitness. It's environment is shaped as a two-dimensional torus, populated by atoms of various types that can be combined by organisms to release energy. Organisms in this model feature very simple "digestive system", contact and ambient sensors and backpropagation neural network brain that coordinates the interaction between the sensory and motor systems. Using much more general approach than Tierra class of simulators, LEE can be used to study the connection between the processes of learning and evolution. It has an advanced multi-window interface and runs on Unix and Macintosh platforms.
Tierra
Tierra is the most advanced and popular platform for the study of the evolution of artificial organisms at the level of the genome. It was developed by Tom Ray of the University of Delaware and the ATR Human Information Processing Research Laboratories in Kyoto, and represents an extreme bottom-up approach to general ALife simulation. Tierra aims to provide an environment in which Darwinian evolution can proceed within a computer, without explicit direction or intervention from a human operator.
The Tierra C source code creates a virtual computer and its Darwinian operating system, whose architecture has been designed in such a way that the executable machine codes are evolvable. The code resulting from the operations like mutating and recombination must remain fully functional so that natural (or artificial) selection could improve the code over time. This is in direct contrast with the machine-language instruction sets used for traditional programming. If you are interested in more detailed description of this "evolvable" instruction set, see this excellent overview of Artificial Life simulators. Along with the C source code which generates the virtual computer, several programs written in the assembler code of the virtual computer are provided. Some of these were evolved from the code written by a human and included to illustrate the power of natural selection. Besides providing usual memory management and timesharing service, the operating system of the virtual computer allows for a variety of factors that affect the course of evolution: three kinds of mutation rates, disturbances, the allocation of CPU time to each creature, the size of the soup, etc. An elaborate observational system keeps a record of births and deaths, sequences the code of every creature, and maintains a genebank of successful genomes.
The system results in the production of synthetic organisms based on a computer metaphor of organic life in which CPU time is the "energy" resource and memory is the "material" resource. Evolution proceeds by natural selection as different genotypes compete for CPU time and memory space. One may argue that implementation of higher-order functions such as intelligence and communication is very difficult in such system: however, unique advantages of Tierra are simply too attractive if you are interested in serious Artificial Life research. Recent introduction of the Beagle graphical front-end implemented in Tcl/Tk offers the greatest cross-platform portability.
There are many more ALife simulators available on the Web. Some of us would argue that Creatures (now in its third release) are the most sophisticated Artificial
Life simulator available commercially. The comprehensive list of ALife Games, Simulators & Demos is available at Luigi Pagliarini's site. Please do not hesitate to send your questions, comments or additions to the list to my e-mail address.