6: Conclusions
An Animation Tool
The Character Evolution Tool is primarily aimed as being an eventual component in the ensemble of tools for the designer of dynamic media. The future of the art of animation may look more like the direction of live actors in the production of a movie than the drawing of thousands of individual picture frames. The Character Evolution Tool contributes towards this eventual ideal.
Imagining this tool in the context of animation production, here is a make-believe dialogue (interpreted verbally here) which a future animator may have, while attempting to develop a character.
"Bring up the 'Human Species.' Now, load the genotype file, "Paul". He still is not funny enough for the animated commercial I am scripting."
"Paul is already in a sitting position, but I need to make him more upright - a tweak here, a tweak there...that's better."
"OK, now, I want Paul to have the ability to cock his head, or maybe perk his shoulders, in a way to indicate confused intrigue, perhaps." I'll breed this population (multiple versions of Paul) for a while, and try to tease this behavior out ..."
"Well, Paul, I'm not getting exactly the kinds of motions I need, but, you know, this funny motion would be great as an action to use at the end of the scene when you take a bite of the cereal. Let's save the genotype file for this action to be used later, and call it 'Gracious_Yum_Yum!'."
"Now, a second action is needed after the yum yum shot, but I have a specific motion you need to follow, so here, learn to follow this gesture' (draws a cute gesture across the screen indicating the direction and style of Paul's head motion).....etc.. etc...
A Research Environment for Exploring Evolution in Design
A tool which allows a designer to easily represent a design problem as a genotype/phenotype pair could be very useful, if not only to augment the designing, then to at least give illumination to the nature of the design problem and the important variables involved. One facet of the Character Evolution Tool I have developed is that it may be seen as a testbed for a genotype/phenotype methodology of designing. During development, as I continued to explore new graphical domains (adding new species of behavior objects to the system) I began to modularize the process of prototyping new behavior object species - and to give myself the ability to prototype a new behavior object species in a short amount of time. To be able to do this, I found, can be a powerful way to explore theme-and-variation within a particular graphical domain. With more work on the interface, the system could be developed further toward this goal, making it easier to prototype new genotype/phenotype representations. Currently, this is done by making alterations to the C language code. Simplifying this process, or even eliminating the need to write code, would be desirable research development to further this tool.
PPP: Professional Phenotype Programmer
But since eliminating programming altogether is a large task, one might more easily imagine enhancing the modular aspect of this system in that phenotype "plug-ins" could be generated easily in C code or C++. The Character Evolution Tool already comes with a modular skeletal framework for creating new phenotypes. A two-tiered design strategy would be useful in this regard: imagine a design firm or an animation production house in which an official meta-designer, or professional phenotype programmer is employed. The chief artist finds a need for a new class of design objects. He or she then asks the phenotype programmer to design a representation for that class, as a phenotype (genotypes would already be available in a huge library). The main part of the system would already be in place, and the phenotype programmer would write the code which takes a set of genes and turns them into parameters for the new class of design objects. Having created this new phenotype, the chief artist and design team could then plug the phenotype into the main system and perform the many functions which are presented in this thesis. The Phenotype Programmer may also be in charge of designing fitness functions to accompany the phenotype, in accordance with how the design team intends the phenotypes to be critiqued in automatic evolution runs.
A Tool for Exploration of Evolutionary Principles
Another domain in which my work may indirectly offer contribution is the exploration of Artificial Life concepts, Darwinian evolution, and principles of emergence (Resnick, 92). I have observed that people using this system tend to become engaged in the populations of animated characters (more than the other species), to make comments on the way in which they are evolving, and to try their hand at breeding characteristics. This observation has lead me to see this as a tool to think with. The interface to the Character Evolution Tool has an instructional side in that it must bring some of the deeper ideas in GA mechanisms and evolution in general to the understanding of the user in order to be fully exploited as a design tool. For the purposes of exploring emergence and artificial life methods, the behavior objects can be compared to individuals within a single species, each species having its own unique genome and expression into a visual form. This facet of the Character Evolution Tool allows an extended way to explore artificial evolution, via many examples of species.
Design and Evolution
In reference to the uses of artificial life techniques in computer graphics, Steven Levy asks, "How can a bottom-up approach satisfy two seemingly exclusive goals: creating organisms which display novel behaviors, and creating organisms which do what we want them to do?" (Levy, 92). This gets to the heart of the problem in using evolutionary techniques in the design of graphical objects.
A constant philosophical underpinning to this project is a question sounding like the debate on Creationism vs. Evolutionary theory: can a system modeled after Darwinian evolution contribute to the design process? Dawkins (86) reminds us that Darwinism signifies a universe "without design." In this spirit, artificial life research is often about setting up minimal initial states and analyzing self-organizing dynamical systems as they emerge - a form of theoretical biology, a kind of what-if modeling where organization emerges bottom-up style (Langton, 91). It is very distinct from deliberate top-down design methodologies. Yet as we have seen, GA's have been explored as aids to solving complex design problems (Goldberg, 83). Perhaps new paradigms of Design are emerging themselves, which include adaptive systems separate from human problem solving, acting as a collaborator to the human creator. The Character Evolution Tool provides an environment with which to study this notion by bringing these two evolutionary processes together (the designer's designing, and the GA's searching).
User Interfaces to the Genetic Algorithm A designer may ask, "How can I remain in charge of what's going on? If a GA is controlling evolution, I want to have the freedom to tell it when; how much; for how long; and for what purpose. And most importantly, I want it to communicate to me what it is doing at all times". How can this system give a designer as much control as possible in using the GA, while on the other hand, letting it do what it is good at? Should the GA be a black box? According to Darwinism, earthly DNA has been doing its phenomenal work throughout the history of earth life without being explicitly "known" by any organisms for the purposes of generally getting around and reproducing. The exception is the recent disruption of biological evolution by humankind, in which the black box is recently being opened, and genetic information is being used directly by the species for purposes of its advancement. At times the activities of the GA should be transparent to the designer - it must communicate. At other times, it should do the dirty work as the designer thinks on a high level. In summary, one aspect of this thesis may be categorized as design research in GA interfaces. The approach taken in this thesis is to offer the user varying levels of access to the nitty gritty mechanics of the GA dynamics, depending on what the user is interested in knowing at a particular time.
Conclusion
The animation system I have described in this thesis represents an intersection of many inquiries. Among them are: the nature of expressivity in visual objects, autonomy vs. control in animated characters, and the use of evolution as a metaphor and as an actual computational technique in graphic design. On the one hand this thesis research has produced a highly interactive artifact which serves as a prototype for future animation and design systems, but it has also presented an environment for which these inquiries can be explored. And at the same time, it can be fun to use.
Research in simulating natural processes continues to enhance the arts of imaging and animation at a rapid pace. The merging of artificial life with computer graphics has begun to produce images and animations of artificial creatures which are very realistic and which elicit sympathy and emotion in viewers, perhaps owing to a certain amount of autonomy in their behaviors. As techniques for autonomous character animation improve, we will see more examples of real time characters who can respond in increasing ways to the animators who make them and to viewers alike. This being a catalyst for my thesis research, the Character Evolution Tool presents some new ways to interact with autonomous characters, for the development of expressive body language.
Future Work
