Complexity_ A Guided Tour - Melanie Mitchell [60]
Von Neumann’s interest in computation was not always well received at the elite Institute for Advanced Study. After completing his work on the EDVAC, von Neumann brought several computing experts to the IAS to work with him on designing and building an improved successor to EDVAC. This system was called the “IAS computer”; its design was a basis for the early computers built by IBM. Some of the “pure” scientists and mathematicians at IAS were uncomfortable with so practical a project taking place in their ivory tower, and perhaps even more uncomfortable with von Neumann’s first application of this computer, namely weather prediction, for which he brought a team of meteorologists to the IAS. Some of the purists didn’t think this kind of activity fit in with the institute’s theoretical charter. As IAS physicist Freeman Dyson put it, “The [IAS] School of Mathematics has a permanent establishment which is divided into three groups, one consisting of pure mathematics, one consisting of theoretical physicists, and one consisting of Professor von Neumann.” After von Neumann’s death, the IAS computer project was shut down, and the IAS faculty passed a motion “to have no experimental science, no laboratories of any kind at the Institute.” Freeman Dyson described this as, “The snobs took revenge.”
CHAPTER 9
Genetic Algorithms
AFTER HE ANSWERED THE QUESTION “Can a machine reproduce itself?” in the affirmative, von Neumann wanted to take the next logical step and have computers (or computer programs) reproduce themselves with mutations and compete for resources to survive in some environment. This would counter the “survival instinct” and “evolution and adaptation” arguments mentioned above. However, von Neumann died before he was able to work on the evolution problem.
Others quickly took up where he left off. By the early 1960s, several groups of researchers were experimenting with evolution in computers. Such work has come to be known collectively as evolutionary computation. The most widely known of these efforts today is the work on genetic algorithms done by John Holland and his students and colleagues at the University of Michigan.
John Holland is, in some sense, the academic grandchild of John von Neumann. Holland’s own Ph.D. advisor was Arthur Burks, the philosopher, logician, and computer engineer who assisted von Neumann on the EDVAC computer and who completed von Neumann’s unfinished work on self-reproducing automata. After his work on the EDVAC, Burks obtained a faculty position in philosophy at the University of Michigan and started the Logic of Computers group, a loose-knit collection of faculty and students who were interested in the foundations of computers and of information processing in general. Holland joined the University of Michigan as a Ph.D. student, starting in mathematics and later switching to a brand-new program called “communication sciences” (later “computer and communication sciences”), which was arguably the first real computer science department in the world. A few years later, Holland became the program’s first Ph.D. recipient, giving him the distinction of having received the world’s first Ph.D. in computer science. He was quickly hired as a professor in that same department.
John Holland. (Photograph copyright © by the Santa Fe Institute. Reprinted by permission.)
Holland got hooked on Darwinian evolution when he read Ronald Fisher’s famous book, The Genetical Theory of Natural Selection. Like Fisher (and Darwin), Holland was struck by analogies between evolution and animal breeding. But he looked at the analogy from his own computer science perspective: “That’s where genetic algorithms came from. I began to wonder if you could breed programs the way people would say, breed good horses and breed good corn.”
Holland’s major interest was in the phenomenon of adaptation—how living systems evolve or otherwise change in response to other organisms or