The Open Border 231
circuits before recombination. After all, such preprogramming might introduce conven-
tional design methods and solutions through the back door and thus fail to fulfi ll ED’s
promise.
The challenges facing such a hands-off approach are understandably enormous and it
is impossible to predict whether this fi eld will bear fruit in the sense described earlier. But
even if, from a practical point of view, ED would be a failed growth on the tree of engi-
neering science, the problems currently identifi ed in the fi eld and some of the methods
used to solve these problems are interesting for my present purpose.
13.2.2 Current State of Development
One key problem in ED is that of scalability. Finding a viable solution to a circuitry design
problem becomes harder with increasing complexity of the circuitry: designing a func-
tional two-bit adder is much easier than designing an evolvable motherboard. Thus many
evolutionary designs for complex electronics fail by leading to solutions that are vastly
inferior to traditional designs. In principle, this problem could be solved by limiting the
search for a functional design to a small portion of the available design space—then, in
many cases, evolutionary processes yield successful, larger-scale designs. However, this
is typically regarded as defeating ED’s purpose of developing functional and innovative
solutions.
A large number of contributions to conference proceedings and of journal papers address
the scalability problem, so there appears to be consensus on the problem. This cannot be
said for the solution, and few proposals amount to more than promissory notes. Still, one
intriguing proposal is to increase ED’s scalability and problem-solving capacity by making
the ED process more similar to natural evolution. The idea behind these proposals appears
to be that since nature has managed to develop solutions to design problems that are even
more complex than those facing engineers, evolutionary electronics can benefi t from imi-
tating nature (Bentley 1999). This does not mean that engineers seek to imitate natural
objects, for example, by making a silicon brain; instead they are interested in modeling
and mimicking the mechanisms by which they think that nature has overcome the scal-
ability problem.
In practice this imitation has at least two levels of intricacy. First, researchers have noted
that complex natural objects typically have various structural features that improve scal-
ability, such as modularity and iteration. Attempts have been made to develop ED processes
that make use of these structural features; one example is the Cellular Encoding approach
(e.g., Koza et al. 1999). In some cases these features are more or less programmed into the
process, roughly speaking by including iteration and modularity rules in a tree of develop-
mental stages where both the confi guration of the artifacts to be designed and the rules for
changing these confi gurations are represented by a “genetic” code. It can be shown,
however, that such so-called static or explicit approaches at best provide very partial