Cheaper than printing it out: buy the paperback book.

Out of Control
Chapter 14: IN THE LIBRARY OF FORM

Latham reported similar experiences while exploring his space. He often ran into what he called a territory of instabilities. In some regions of possible forms, significant changes in genes would effect only insignificant shifts in forms -- Sims's basin of stagnation. He'd have to really push the genes miles around to move an inch in form. Yet, in other regions, minute changes in genes would produce huge alterations in form. In the former, Latham's progress through the space was glacial; in the latter, his tiniest move would send him rapidly careening through the Library at a zoom.

To avoid overshooting a destination of possible form, and to accelerate its discovery, Latham would purposefully twirl a mutation knob as he explored. At first he'd set the mutation rate high, to skip through the space. As the shapes became more interesting, he'd turn the mutation rate down so that each generation sliced thinner, and he'd slowly creep up to a concealed shape. Sims wired his system to perform a similar trick automatically. As the image he was evolving became more complex, his software would crank down the mutation rate for a soft landing on the final form. "Otherwise," Sims says, "things can get crazy as you are trying to fine-tune an image."

These frontiersmen developed a couple of other tricks for traveling through the Library. The most important trick was sex. Dawkins's Biomorph Land was a fertile, but puritanical, place that hadn't a hint of sex. All variation in Biomorph Land occurred by asexual mutations from a single parent. Sims's and Latham's worlds, in contrast, were driven by sex. A major lesson the frontiersmen realized was that you could do sex in an evolutionary system in any number of ways!

There was of course the orthodox missionary position: two parents, with genes from each. But even that plain vanilla mating can be accomplished in several ways. In the Library, breeding is analogous to taking two books and merging their text to form a child-book. You can beget two kinds of progeny: in-betweener books or outsider books.

In-betweener offspring inherit a position in between Mommy and Daddy. Imagine a beeline in the Library bridging Book A and Book B. Any child (Book C) would be found somewhere in the Library on that imaginary line. In-between offspring can be exactly halfway in between as they would be if they inherited exactly half of their genes from Pop and half from Mom. Or, they can be in-between at some other proportion, say 10 percent Mom and 90 percent Dad. In-betweeners can also inherit alternating chapters from Book A and Book B, or alternating clumps of genes from Mommy and Daddy. This method retains genes that may be linked to each other by a proximal function, making it more likely to accumulate "good stuff."

Another way to think of in-betweeners is to imagine creature A morphing -- in the Hollywood term -- into creature B. All the creatures it morphs through on its way from A to B are the pair's possible in-betweener offspring.

Outsider offspring inherit a position outside of the morph-line between Mommy and Daddy. Rather than some random halfway stage between a lion and snake, they are a chimera boasting a lion's head with a snake's tail and forked tongue. There are several different ways to generate chimera, including the pretty basic one of fishing in a potluck stew of random traits possessed either by Mom or Dad. Outsider offspring are wilder, less expected, more out of control.

But that's not the end of the weirdness feasible in evolutionary systems. Mating can also be perverse. William Latham is currently playing around with polygamy in his system. Why limit mating to two parents? Latham coded his system to allow him to choose up to five parents and assign each parent varying weights of inheritability. So he says to his brood of children forms: next time give me something very much like this one, that one, and that one, and somewhat like this one, and a little bit like that one. Then he marries them together and they co-procreate the next brood. Latham can also assign negative values: as in, not like this one. In effect he has made an antiparent. When an antiparent mates in multiple marriages it sires (or not-sires) children as unlike it as possible.

Moving further still from natural biology (at least as far as we know it) Latham hacked a program for Mutator which follows the breeder's progress through the Library. Genes that persist over a particular breeding course, the Mutator assumes the breeder likes. It makes those genes dominant. Genes that keep changing, the Mutator reads as "experimental" and unsatisfying to the breeder, so it reduces their impact by declaring them recessive in any mating.

The idea of tracking evolution in order to anticipate its future course is bewitching. Both Sims and Latham dream about an artificial intelligence module that could analyze a breeder's progress through form space. The AI program would deduce the common element shared by the selections and then reach far ahead into the Library to retrieve a form that encapsulates that trait.

At the Pompidou Center in Paris, and at the Ars Electronica Festival in Linz, Austria, Karl Sims installed a public version of his artificial evolution universe. In the middle of a long gallery space, a Connection Machine hummed on a platform. The jet-black cube was vested in flickering red lights, which syncopated as the machine thought. A heavy cable connected the supercomputer to an arc of 20 large monitors. A footpad on the floor sat in front of each color screen in the crescent. By stepping on a footpad (which covers a switch) a museum-goer chose a particular image out of the row.

I had a chance to breed CM2 images in Linz. To start, I selected what looked like an impression of poppies in a garden. Instantly, Sims's program bred 20 new offspring of the flowers. Two screens filled with gray rubbish, the other 18 displayed new "flowers," some fragmented, some in new colors. At each turn I tried to see how flowery I could push the image. I quickly worked up a sweat running from pad to pad in the computer-heated room. The physical work felt like gardening -- nurturing shapes into existence. I kept evolving more elaborate floral patterns, until another visitor shifted the direction toward wild fluorescent plaids. I was dumbfounded by the range of beautiful images that the system uncovered: geometric still lifes, hallucinogenic landscapes, alien textures, eerie logos. One after another elaborate, brilliantly colored composition would appear on the monitors and then, unchosen, retreat forever.

Sims's installation breeds all day, every day, bending its evolution to the fancy of the passing mob of international museum visitors. The Connection Machine records every choice, and every choice leading up to the choice. Sims now has a database of what humans (at least art museum humans) find beautiful or interesting. He believes that these inarticulate qualities can be abstracted from such a rich trove of data and then used as a selection criteria for future breeding in other regions of the Library.

Or, we may be very surprised to find that nothing unifies the selection criteria. It may be that any highly evolved form is beautiful. We find beauty in all biological creatures, although individual people have individual favorites. Overall, a monarch butterfly is no more or less striking than its host, the milkweed pod. If inspected without prejudice, parasitic beasts are beautiful. My suspicion is that the beauty of nature resides in the process of getting there by evolution and by the important fact that the form must work biologically as a whole.

Still, something distinguishes the selected forms, no matter what they are, from the speckled gray noise that surrounds them. Comparing the chosen to the random may tell us much about beauty and even help us figure out what we mean by "complexity."

continue...