Cheaper than printing it out: buy the paperback book.

Out of Control

The Russian programmer Vladimir Pokhilko reminds me that evolving for beauty alone may be a sufficient goal. Pokhilko and partner Alexey Pajitnov (who wrote the famously addictive computer game Tetris) designed a very powerful selection program that breeds virtual aquarium fish. Pokhilko told me during early work on the game, "When we started we didn't want to use the computer to make something very practical, but to make something very beautiful." Pokhilko and Pajitnov did not set out to make an evolutionary world. "Our starting point was ikebana, the Japanese art of arranging flowers. We wanted to make some kind of computer ikibana. But we wanted something alive, moving. And which never repeats itself." Since the computer screen "looks like an aquarium, we decided to make a customizable aquarium."

Users become artists by filling the aquarium with the right combinations of colored fishes and quantities of swaying seaweeds. Users would need a large variety of organisms. Why not let the aquarist breed their own? Thus "El-Fish" was hatched, and the Russians found themselves in the evolution game.

El-Fish became a monster of a program. It was mostly written in Moscow during a time when smart U.S. entrepreneurs could hire a entire unemployed Russian university math department for the salary of one U.S. hacker. Up to 50 Russian programmers, ignorant of Dawkins, Latham, and Sims, wrote code for El-Fish, rediscovering the power and method of computational evolution.

The commercial version of El-Fish, released by the U.S. software publisher Maxis in 1993, compresses the kind of flamboyant visual breeding done by Latham on large IBMs and Sims on a Connection Machine into a small desktop home computer.

Each El-Fish has 56 genes which define 800 parameters (a huge Library). The colorful fish swim in a virtual underwater world realistically, turning with the flick of a fin as fish do. They weave between strands of kelp (also bred by the program). They pace back and forth endlessly. They school around food when you "feed" them. They never die. When I first saw an El-Fish aquarium from ten paces away, I took it to be a video of a real aquarium.

The really fun part is breeding fish. I got started by dipping a net somewhat randomly into a map of the hypothetical El-Fish ocean, fishing for a couple of exotic parent fish. Different areas shelter different fish. The ocean is, in fact, the Library. I hauled up two fish which I kept: a plump yellow fish spotted in green with a thin dorsal fin and an overbite (the Mama) and a puny blue torpedo-shaped guy with a Chinese junk sail of a top fin (the Pa). I could evolve from either, that is, I could asexually mutate new fish from either the fat yellow one or the tiny blue one, or I could mate the pair and select from their joint offspring. I chose sex.

As in the other artificial evolution programs, about a dozen mutated offspring appeared on the screen. I could slide a knob to adjust the mutation rate. I was into fins. I chose a large-finned one, pushing its shape each generation toward increasingly ornate, heavy-duty fins. I got one fish that seemed to be all fins, top, bottom, side. I moved it from the incubator and animated it before plunking it into the aquarium (the animation procedure can take minutes or hours depending on the computer). After many generations of increasingly weird finny fish, I evolved a fish so freakish that it wouldn't breed anymore. This is the El-Fish program's way of keeping the fish, fish. I had entered the outer boundaries in the Library beyond which the forms are less than fishlike. El-Fish won't render nonfish creatures, and it won't animate unorthodox fish because it's too hard to make a monster move. (The code relies on standard fish proportions to keep a creature's movements convincing.) Part of the game is users trying to figure out where those fishy limits lie and whether there are any loopholes.

Storing full fish consumes far too much disk memory, so only the bare genes of the fish are filed. These tiny seeds of genes are called "roe." Roe are 250 times more compact than the fish they grow into. El-Fish aficionados swap the roe of selected creations over modem lines or stock them in digital public libraries.

One of the programmers at Maxis in charge of testing El-Fish discovered an interesting way to explore the outer limits of the fish Library. Instead of breeding or fishing the pool for sample stock, he inserted the text of his name (Roger) into a roe. Out came a short black tadpole. Pretty soon everyone in the office had a tadpole in their El-Fish tank. Roger wondered what else he could transform into fish roe. He took the digital text of the Gettysburg Address and grew the digits into a ghostly creature -- a pale face trailing a deformed batwing. The wags dubbed it a "Gettyfish." Hacking around they discovered that a sequence of about 2,000 digits of any sort can be shanghaied into serving as roe for a possible fish. Getting into the swing of things, the project manager for El-Fish loaded the spreadsheet of his budget into El-Fish and birthed the bad omen of a fish skull, fangy mouth, and dragon body.

Breeding was once a craft belonging solely to the gardener. It is now available to the painter, the musician, the inventor. William Latham predicts evolutionism as the next stage in modern art. In evolutionism, the borrowed concepts of mutation and sexual reproduction spawn the art. Instead of painting or creating textures for computer graphic models, artist Sims evolves them. He drifts into a region of woodlike patterns and then evolves his way to the exact grainy, knot-ridden piney look which he can use to color a wall in a video he is making.

You can now do this on a Macintosh with a commercial template for Adobe Photoshop software. Written by Kai Krause, the Texture Mutator lets ordinary computer owners breed textures from a choice of eight offspring every generation.

Evolutionism reverses the modern trend in the design of artist's tools that bends toward greater analytical control. The ends of evolution are more subjective ("survival of the most aesthetic"), less controlled, more related to art generated in a dream or trance; more found.

The evolutionary artist creates twice. First, the artist acts as god by concocting a world, or a system for generating beauty. Second, he is the gardener and curator of this made world, interpreting and presenting the chosen works he nurtures. He fathers rather than molds a creation into existence.

At the moment the tools of exploratory evolution restrict an artist to begin with a random or primitive start. The next advance in evolutionism is to be able to begin with a human-designed pattern and then arbitrarily breed from there. Ideally, you would like to be able to pick up, say, a colorful logo or label that needed work (or mind-altering modification) and progressively evolve from that.

The outlines of such a commercial software are pretty clear. Will Wright, SimCity author and founder of Maxis, the innovative software publisher behind El-Fish, even came up with the perfect jazzy title: DarwinDraw. In DarwinDraw you sketch a new corporate logo. Every line, curve, dot, or paint stroke of the image you create is rendered into mathematical functions. When you are done, you have a logo on a screen and a mutable set of functions as genes in the computer. Then you breed the logo. You let it evolve outlandish designs you could never have thought of, in detail you don't have time to do. You jump around randomly at first, just to brainstorm. Then you hone in on an unusual and striking arrangement. You turn the mutation rate down, use multiple marriages and antiparenting to fine-tune it to its final version. You now have an obsessively detailed evolved artwork with cross-hatching and filigrees you wouldn't believe. Because the image is based on algorithms, it has infinite resolution; you can blow it up as large as you like with unexpected detail to spare. Print it!

As a demo of this power of evolutionism, Sims scanned the logo of CM5 into his program and used it as a starting image to breed an "improved" CM5 logo. Rather than the sterile modern look, it had frilly organic lines around the edges of the letters. Folks in the office liked the evolved artwork so much that they decided to make a T-shirt out of it. "I'd really love to evolve neckties," Sims says. His other suggestions: "How about evolving textile patterns, wallpaper designs, or type fonts?"

IBM has been supporting artist William Latham's evolution experiments because the global corporation realizes there is commercial potential here. While Sims's evolution machine is, according to Latham, "a grammar that is more ragged, more uncontrolled," Latham's is more controlled and useful to engineers. IBM is turning the evolutionary tools Latham developed over to automobile designers and having them mutate car body shapes. One of the questions they are trying to answer is whether evolutionary design techniques are more useful in the beginning of rough ideas or later in fine tuning, or both. IBM intends to make a profitable project out of it. And not only for cars. They imagine evolutionary "steering" tools useful for all kinds of design problems entailing large numbers of parameters which require a user to "back up" to a stored previous solution. Latham pictures evolution taking root in packaging design, where the outer parameters are firmly fixed (size and shape of the container), but where what happens within that space is wide open. Here evolution can bring in multiple levels of detail that a human artist would never have the time, energy, or money to do. The other advantage of evolutionary industrial design, Latham has slowly come to realize, is that it is perfectly suited to design by committee. The more people that play, the better.

The copyright status of an artificially evolved creation is in legal limbo. Who gets the protection, the artist who bred or the artist who created the program? In the future, lawyers may demand a record of the evolutionary path an artist followed to arrive at an evolved creation as evidence that such work belongs to him and was not copied, or due to the creator of the Library. As Dawkins showed, in a truly large Library it's improbable to find a pattern more than once. Owning an evolutionary pathway to a particular point demonstrates irrefutable proof that the artist found that destination originally, since evolution doesn't strike twice.