(Animated Artificial Life)
3.1 Introduction
Life is motion. We see it stirring under blankets, shifting in a crowd, and sauntering down the street. We see it darting beneath leaves in a pond, scurrying around branches of a tree, and lumbering across a field. We can't always see what a living thing looks like in all its visual detail, but we can usually recognize its motion as something distinctly alive, as opposed to something moved by a machine or by the wind. Sometimes it's not just a matter of photoreceptors sewing together pixels to construct an image, but something incorporating brain and time, sensing a unique spatiotemporal event: life. What makes a thing appear alive? I think it might have something to do with a combination of apparent physical laws and some apparent adaptive behavior within that physical system, which is goal-directed. Living things have the quality of being tightly coupled with their environments - adapted, situated, immersed. This quality is achieved through recursive processes such as evolution and learning. The acquisition of lifelike behavior in software can be achieved through similar means, and visualized as animated artificial life.
The discussions in this paper are founded on the following premise: in designing real-time animated artificial life worlds, it is better not to begin by solving computergraphics rendering problems in the classical way, and then to subsequently fill texturemapped polyhedra with behavior, as an afterthought. It is better to compose an ontology of dynamics first - biological and physical laws; to solve problems of adaptive behavior within this framework of biological and physical laws; and then to render them visually, in a way that speaks to the truth of the underlying dynamic. The simulations I will discuss throughout this paper were built with this in mind.
1.1. Disney Meets Darwin
The art of animation as invented by the Disney animators has been called "The Illusion of Life." Computer Animated cartooning is still catching up to the expressivity, humor, and life-filled motion possible with classic cel-based character animation, although recent animators such as those at Pixar have demonstrated some success in adapting computer technology to the fine art of classic character animation. As film animators refine this marriage of new and old technologies, a new form of animated character emerges, not from within the film industry but from within various research labs. Add to Disney's "Illusion of Life," the "Simulation of Life," and witness a new technology - one which is compatible with the goals of Virtual Reality - future cyberspaces in which characters are not just animated, they are autonomous, reactive agents as well. Imagine further that these characters achieve their behaviors, in all their complexity and subtlety, by adapting to their environments and each other, "designing themselves," not through animation scripting, but through reproduction, crossover, mutation, and selection.
The surprises and novelty arising from such evolutionary systems can indeed be an annoyance and a distraction when the design objective is well understood, and when the animator knows exactly how a motion must look. But when seen as an art form or as a prototyping tool in which creativity, discovery, serendipity, and chance are welcome elements in the design process, these systems can be useful. A whole world of strange and funny animated behavior lies dormant, waiting to be fished out of the primordial soup.
2.1.2. Designing Emergence
According to one account, L. Frank Baum, the author of "The Wizard of Oz," was wandering around in a trance while developing a story, for several weeks. His wife asked what was wrong. He said that his characters wouldn't do what he wanted them to do. Days later, he seemed better and was busy back to work. His wife asked him how he had solved his problem. He said, "I let them do what they want to do" [American Heritage, 1964]. Many creative projects benefit from this approach: a work in progress often gets weighted down by top-down design, too much intentionality from its inventor, and cries out to be set free(to acquire autonomy, to be what it naturally wants to be. While this may have little to do with most problems in scientific investigation, it does touch upon problems in synthesis and simulation, which are tools used in artificial life.
A key objective in artificial life projects is to construct systems in which self-organization and adaptive behaviors can arise spontaneously, not by design, but as emergent phenomena, having meaning in the context of the simulated environment. A duality is observed in the creation of artificial life worlds: while the goal is something emergent like self-organization or adaptability, the system is ultimately a designed artifact. Creating artificial life worlds, in this view, is an activity of designing emergence.
People often describe things within one particular level of detail, such as: the atomic, molecular, biological, psychological, ecological, social, etc. These levels overlap in our deeper pictures of reality. In composing an artificial life simulation, one chooses an arbitrary level as a base. On a "higher" descriptive level, life-like phenomena emerge as a result of the dynamics, when the simulation is run.
For instance, many artificial life simulations model populations of organisms whose spatial positions are represented by locations in a cellular grid, and whose physical means of locomotion (jumping from cell to cell) are not clearly specified, but simply represented as instantaneous relocations in space. The emergent properties in question may not require a deeper level of physical simulation than this. Physical mechanics are not needed. However, the experimenter may be more interested, for instance, in the specific motions of particular bodies than in general issues of population dynamics. A simulation in which an articulated body plan is integral to locomotion (as well as reproduction) requires another level of abstraction, and a deeper physical model becomes necessary, as a different set of emergent descriptors are sought.
In presenting my approach to artificial life, I will be describing a variety of experiments. I consider this body of work to be an exploration and a celebration of autonomously generated motion and form. The original impetus is not biological science, although biological principles have been incorporated into the work. The origins are in the study of aesthetics, perception, and creative design, particularly in the context of time-based media.
Towards the quest for increasing autonomy, and more interesting emergent phenomena, I have been progressively deepening my artificial life models. These include a variety of simply-rendered articulated figures incorporating physics and periodic motor control. The most recent simulation consists of a population of many hundreds of swimming creatures which animate in real-time. The creatures come in a large variety of anatomies and motion styles. They are able to mate with each other, and choose who has the "sexiest" motions, enabling evolution of swimming locomotion and anatomy which is attractive (beauty of course being in the eye of the beholder). The best swimmers reproduce their genes (because they can swim to a mate), and the most "attractive" swimmers get chosen as mates. In this simulation, not only are the aesthetics of motion and form subject to the chaotic nature of genetic evolution, but the creatures themselves partake in the choice of what forms and motions emerge. Emergent phenomena are observed on a number of levels, from population dynamics down to details as small as the idiosyncratic wiggling of a body part for sexual attraction.
2.1.3. Organization of This Chapter
This paper describes an approach to the new science of artificial life [Langton, 1989], stemming from animated art and a design process for creating it. It is an expansion of a paper entitled "Designing Emergence in Artificial Life Worlds" [Ventrella, 1998], which was presented at the Virtual Worlds Conference in Paris in 1998.
The simulations are presented in chronological order, although a bit of shifting around has been done in order to introduce topics in a more logical way. In these simulations, emergent behavior is enabled by progressively designing levels of autonomy into the model. In a very special way, the methodology itself has evolved over time. The simulations are used to illustrate how this methodology was built, with the first examples being quite primitive, and with the most recent simulation incorporating deeper physics and more autonomy.
Section 1 introduces the main concepts of this paper. Section 2 discusses background research and similar approaches to animated artificial life. Section 3 comprises the bulk of this paper, and provides a tour of a series of artificial-life-based experiments, accompanied by commentary. Section 4 introduces the latest set of simulations, in which organisms mate autonomously. Section 5 discusses the issue of rendering. Section 6 discusses various topics pertaining to artificial life and its relation to artificial intelligence. Finally, a conclusion is given in section 7.
next section:
Background
Background
