Mechanical Garden


Writing in Z+Blog, Andrew Zolli describes a circuit developed an unorthodox way. Rather than devising it from scratch, researchers "simulated biological evolution by testing, mating and mutating more than 4000 generations of circuit designs. A set of instructions called a 'fitness function' killed off inefficient designs and mated efficient ones, mimicking the effects of natural selection."

One side effect of the method: It took months for the team to figure out how the resulting mechanism worked. "The evolutionary approach exploited what was there in ingenious and intricate ways," Zolli notes, "including properties of the chip its inventor couldn't even measure at the time of the experiment. Ultimately, to figure out what was going on, he had to use the kinds of techniques that biologists use to understand the nervous systems of simple animals."

Programmers have been generating software this way for years. Now, apparently, evolution is coming to hardware. "As engineers act less on planning the specifics of the design," Zolli writes, "and more as artificial selectors of the end product, engineering begins to look less and less like architecture, where one designs from the foundation up, and more and more like gardening, where one directs forces only partially under one?s direct control."

NEXT: Have You Got Ears?

Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Report abuses.

  1. Very interesting article, although I think it may overstate it’s case. Design by evolution may be a handy tool for solving some design problems. Ones where we know the desired end result, but we can’t make all the intermediate steps in a conventional way. However, it is not going to replace engineers or alter our overll approach to design anymore than replacing the drafting board with CAD did.

  2. Hmm. This sounds close to developing “life”. I’ve read, too, where they’re working with little bitty bacteria-like organisms to act as a switch when tickled by an electrical current. Chained together properly, a semi-life form can be envisioned.

    Don’t let the politicians know, though, or they’ll pass a law.

  3. Jarrod, I don’t think Zolli was suggesting that traditional approaches to engineering were going to disappear. Just that this was a radically different way of doing things with a lot of interesting implications.

  4. There may well be applications for this, another tool that can be used. However it is important to understand that evolution does not produce ‘ideal’ designs, as some would think. People assume that Darwin built his case for evolution based on how organisms seemed ideally suited to their environment. However Darwin’s case was actually built equally on the imperfections in the ‘design’, that could only be explained in that they were modifications of something pre-existing to optimize it for a particular circumstance.

    This is an important distinction, as some creationists have thought that evolution is a tautology with no verifiable predictions, or to a lesser extent that the evidence for the theories of evolution or intelligent design is more or less equivalent since both produce ‘ideal’ organisms. This is not in fact what evolution predicts.

    That said, the technological adaptation may not face similar problems with design optimization since it exists in an environment artificially created to favor a certain result, and the options for the input elements can be predetermined by the architecture of the experiment.

  5. Joe,

    My comments about the future of engineers were based partly on this passage:

    “As engineers act less on planning the specifics of the design, and more as artificial selectors of the end product,…”

    This has the effect of taking away our primary function and making us sound a lot more like the despised marketing department.

    I also intended to my objection to the idea that we are replacing a meticulous product development method with a more “mystical” evolutionary approach.

    When one looks at what is actually going on, the researchers are taking existing circuit designs and using a complex algorithm to select the best elements from each. That the algorithm fails to keep track of its steps and requires reverse engineering to understand the finished product is not the miracle of life, or anything like it. It is merely an omission of proper record keeping.

  6. I think it’s misleading to overstate the usefulness of genetic algorithms (the techniques described in this article). While GAs have produced some interesting results over the last 30 years, they are severly limited in their capability. I am a computer scientist, and I have (attempted to) use GAs in a number of projects, but for all but the simplest tasks they pale in comparison to human-created solutions. GAs do not have the power of abstraction, metaphor, or problem subdivision and pattern recognition that are essential in computer science for designing complex systems.
    A GA is only as expressive as its program size, and as the size of the GAs we try to evolve increase, the effort required to evolve a good solution increases exponentially. To evolve a program that functioned even as a basic word processor would likely take millions of years with current technology, and we would be unable to understand the program once we had suceeded. In my experience, any time that we can apply human ingenuity, GAs are a poor solution. Their only successes come in problems which completely defy human analysis.

  7. The biggest problem I have had with GA’s is choosing a good fitness function. It’s often by no means obvious what a good fitness function is for a given problem, and without that the ‘landscape’ you’re optimising on is not well fitted to the design task.

    Having said that, I’ve seen GA’s do some truly remarkable optimisations over large design spaces (most notably in the selection of low cross-correlation bit sequences for medium access control in wireless LAN prototypes).

  8. Whoa! Get me out of here.

  9. EMAIL:

    DATE: 12/11/2003 02:41:52
    God had some serious quality-control problems.

Please to post comments

Comments are closed.