The D.E.A.D system evolves architectural forms using a genetic algorithm. Genetic algorithms are pieces of computer code that are capable of cross-breeding, mutating and reproducing. In the case of this system, the algorithm is a set of rules that describe the formation of space and structure, based on a set of 8 parameters (see the algorithm section for more on the algorithmic morphology of the system). The system roughly follows the cycle of life. These algorithms produce architectural forms, these forms are tested for survival, competing against other forms in the virtual bioreserve, the more successful forms survive, and their genes (parameters) are cross-bred to produce a new generation of possible form-generating.

The figure on the right shows a snapshot of 3 generations of evolution. Each generation starts with 4 parents (in the D.E.A.D system they are all 'sexless') that are cross-bred with each other to produce 12 offspring – often mutation occurs during the cross-breeding, ensuring continual variation in new generations. The 12 offspring are tested for fitness, determining their survival in the bioreserve. The 4 most successful off-spring go on to produce a second generation, those are in turn tested and the top four are parents to the next generation. This evolutionary process can continue ad infinitum.

The survival of genetic code from one generation to the next depends upon the fitness of the architectural form generated. In nature fitness is determined by a matrix of factors including food and resource limitations, predators, and innumerable other environmental affects. In the environment of the D.E.A.D system, a set of factors that are more architecturally suited are used to determine fitness, among these are structural feasibility, material efficiency and program models. The criteria for structural feasibility and material efficiency are self-explanatory, testing whether a building will "stand up" and how many elements are used, compared to others in the same generation. The program models for fitness are more complex.

The component of the fitness criteria that tests program uses an array of abstract volumes. These volumes fall into two categories, exclusionary and inclusionary volumes. The inclusionary volumes are used to test the overall size of the structure and the density of elements surrounding an area – which is equated to its level of privacy. The test works by counting all elements of the form that fall within the volume, relative to the number of elements that fall outside the volume; the resulting factor is used to determine a portion of its fitness. The exclusionary tests work by testing how hollow is an area of the form. The system counts how many elements fall within a volume, the higher the number the more 'unfit' the overall form is. An example of program volumes being used to test the fitness of a form is shown on the left.

The genetic algorithm as used in computer science is typically a search algorithm, for finding optimal solutions for a problem. The solution that the genetic algorithm finds is the form that best meets the fitness tests of program, structural efficiency and structural feasibility. In that way, the D.E.A.D system could be thought of as a form-finding system. However, the goal of the system is more than to find a solution for a specific design problem – but to be a generative device, discovering new problems as well as new solutions. To create a system that would work freely, testing new forms as well as generating new tests, the D.E.A.D system uses genetic algorithms to produce the fitness tests as well. The fitness tests are described by a set of parameters, and are evolved from generation to generation. Instead of devising another set of criteria to test the fitness of these evolving algorithms, their fitness is determined by their ability to 'kill' the architectural structure. This puts the fitness tests and the structures in a predator-prey relationship, continually co-evolving. Once an architectural form is evolved that is an optimal solution for one set of fitness criteria (program, structure, efficiency) the fitness test will begin to mutate, because evolution will encourage them to find test that are more successful at destroying the forms. This predator-prey co-evolution can run indefinitely, producing a library of new structures as well as matching models of program. The graph on the right shows the changing size of structures evolved over 100 generations. The black line shows the changing size of structures evolved using a co-evolved fitness test, the red line shows the changing size of structures evolved without co-evolution. The graph shows the evolution of structures without a co-evolved fitness test settles into stasis, after 30 or so generations there is little variety in the forms created by the system without co-evolution, however, the co-evolving system continually changes - producing novel forms.