DAILY MAIL & GUARDIAN
15 September 1999
Designed by natural selection
JAMAIS CASCIO advises suits and other Hollywood weasels on what could happen over the next hundred years. He worries about being called a futurist.
t's highly likely that one of the major disputes of the next century will be between the supporters of evolution and the supporters of intelligent design. Evolutionists will argue that the process of generations of trial-and-error leads to subtle and surprising new forms; design supporters will counter that only a guiding hand can create complex structures. Some will argue that both are right, or, at least, both should be accepted. But I'm not talking about school curriculum here -- I'm talking about computers.
Today, computers -- meaning both the hardware and software -- are built by engineers and programmers who can spend years thinking about how a particular chip or application should be constructed. Everything from the circuit pathways to the programming interfaces are thought out in advance. For the most part, this has led to microchips that are powerful, robust, and rapidly gaining speed, and software that is complex and sophisticated.
But some people aren't satisfied.
Scientists and engineers around the world have started looking at evolutionary biology as an example of a new way of designing hardware and software. There are few "market ready" applications yet, but the results so far have been encouraging. Using evolution-type methods, computers have come up with novel approaches for circuit designs, signal-processing software, even antennas.
Evolutionary design -- sometimes called "genetic algorithms" when done just in software -- relies on the same notions of natural selection and mutation that underlie biological evolution. Take a large number of individuals, each slightly different. Introduce some mutation, either by randomizing small bits or mixing elements from individuals (the electronic version of sexual reproduction). Check the resulting generation against the goal -- how well do the various designs accomplish the needed task? Get rid of some of the designs that do very poorly, add more of the designs that do fairly well. Now repeat the process. Many thousands of times.
In nature, repeating thousands of generations can take millions of years; evolution is a very long-term process. In a computer, however, many thousands of generations can go by in a day, letting scientists watch the evolution of new "species" of technology right on their screens. Sometimes, evolutionary systems come up with designs that are quite similar to ones humans came up with -- an evolutionary circuit design system at Stanford University, on its own, developed a layout that was nearly identical to early (1950's-era) human-designed circuits.
Sometimes, however, evolutionary systems come up with surprises. A researcher at the University of Sussex in Brighton, England, used a set of chips called a "field programmable gate array", which can be reprogrammed on the fly, and actually had the circuits evolve (rather than evolving the design in a software simulation). His goal was relatively simple: have the chips be able to recognize two distinct sounds. But the result stunned researchers. Not only did the eventual design use fewer than one-half the circuits that a human-engineered layout used, it did so in a way that took the researcher months to unravel. The evolutionary process had used analog elements (such as heat buildup and circuit pathways) of otherwise digital chips to accomplish the task.
Evolutionary methods are particularly good at coming up with systems that respond well to feedback, that (in essence) learn their way around, and can adapt over time. Engineers are working on artificial limbs that have genetic-algorithm-based controls, so that, rather than an amputee needing to learn how to control the arm (for example), the systems on the arm "learn" what the user wants to do based on nerve signals. The process is remarkably similar to an infant's attempts to use its limbs -- lots of flailing around at first, but visible and relentless improvement with time.
Not all engineers are confident in evolutionary methods. In many cases, current evolutionary systems are fairly fragile. With the field gate array device described above, the final program could not be transferred from one FGA system to another -- it relied too heavily on the unique heat properties of those particular chips. While further generations of evolution would have solved that particular problem -- varying the environment until a design that worked on multiple chips emerged -- this sort of environmental sensitivity is a recurring concern.
Still, the process, which has worked in nature for billions of years, holds much promise. The applications range beyond software and computer hardware. NASA has used evolutionary design techniques to come up with entirely new types of antenna, which have odd shapes looking like bent drinking straws or crumpled up paper. Derek Linden, the researcher who ran the design software, patented the antennas earlier this year. To the best of his knowledge, it was the first patent given to a design by a nonhuman creator.
It will assuredly not be the last.
© Daily Mail & Guardian - 15 September 1999
* Jamais Cascio is a consultant and writer specializing in scenarios of how we may live over the next century. His clients have included mainstream corporations, film and television producers. He has written for many publications, including Wired and TIME, and is currently working on a screenplay. He is an active member of the oldest and most influential online community, The Well, and believes that new technologies are pushing people into new social, economic and political realms.
Join the PCReview mailing list and receive our top stories and regular features once a week, by email.
* Subscribe now
Published weekly by the Electronic Mail & Guardian, Johannesburg, South Africa. Send email comments to the editor, Gavin Dudley