Artificial Life:
The Quest for a New Creation
By Steven Levy
Pantheon, 390 pages, $24
Before you apply your CPU (cerebral processing unit) to this startling book and its controversial subject, I recommend that you reset some switches. First, set your religion switch to ”off.” Your encounter will be more productive if you temporarily lay aside transcendental beliefs in favor of the simpler evidence of your senses. Now set your focus switch from ”local” to
”distant.” The study of artificial life, or a-life, asks us to relinquish a pre-Copernicus view of our species as the center of the universe, the ultimate aim of evolution or the only means to noteworthy behavioral ends-for example, co-operation, learning, consciousness, intelligence, culture. Last, set your mind switch to ”open.” The best way to approach Steven Levy`s book is cordially and credulously, preparing yourself to believe six impossible things before breakfast.
”Artificial Life” is a review of developments across a broad spectrum of research into what might initially be defined as computer simulation. Levy, who wrote the 1983 bestseller ”Hackers” and writes a lively column for Macworld magazine, is a capable and sensitive guide to this terrain. Although conceptually challenging, his approach is kind to the nonspecialist; his subject, in fact, is presented here less for computer buffs and more for general liberal-arts readers who enjoy the humanistic musings of Oliver Sacks and Stephen Jay Gould.
What, exactly, is being simulated in a-life experiments depends on who`s doing the explaining. A half-century ago, the seminal computer thinker John von Neumann launched the enterprise by his theoretical construct of the self- reproducing automaton. In the late 1960s, mathematician John Horton Conway went to work with Von Neumann`s ”cellular automata,” computer entities which follow only a few simple rules but exhibit, through step-by-step iterations,
”interesting behavior.” Two years of tinkering with these rules gave him the famous game of Life, which first-generation computer viewers have followed as intently as children watch bugs in a vacant lot.
Dozens of others joined what became, in Levy`s phrase, a garage-band science. John Holland, the nation`s first Ph.D. in computer science, believed that computers could display adaptive behavior and planted the seeds for what Levy calls ”a new synthesis between biology and information.” Ed Fredkin, an early virtuoso programmer and later head of MIT`s Artificial Intelligence Laboratory, was convinced that the universe is made of information and that life itself is a digital information process. ”Living things may be soft and squishy,” he put it, but the basis of life is digital. ”Nothing is done by nature that can`t be done by computer.”
Take one entry-level example. Starting from an understanding that simple rules can duplicate the behavior of complex nonliving systems in nature-growth in crystals or variety in snowflakes-computer animator Craig Reynolds applied similar rules to mirror the flocking behavior of blackbirds. There`s no one blackbird ”drill instructor” to put the birds through their paces, yet visually complex flocking ”gives the strong impression of intentional centralized control.” After long effort, Reynolds reduced his flocking behavior rules to three: a force to keep the flock together, a counterforce to prevent individuals from getting too close to each other, and an ability to match velocity. He then programmed computer ”boids” (from ”birdoids”) to act individually according to these rules and turned them loose on his computer screen. As Levy tells it:
”He fine-tuned the program over the next few months and eventually . . . began to get precisely the kind of flocking you would see on nature shows. . . . The boids, each using nothing but Reynolds` simple rules, were able to flock in large configurations so convincingly that ornithologists, intuiting that real birds might be performing the same algorithms as Reynolds`s creations, began calling the animator to find out his rules.”
Reynolds even programmed obstacles into his computer environment: thick cylinders resembling Greek columns. The boids reacted like birds: the flock veered, split apart, rejoined and continued on its way, all without specific programming instructions to do so. ”One hapless boid,” reports Levy, ”found itself hemmed in by fellow boids, unable to miss colliding with the column. It slammed into the cylinder, halted for a brief instant-seemingly dazed-and then, as though regaining consciousness and belatedly recalling the rules, sped up to join the flock.”
What`s going on? A complex system is displaying what Levy and his a-life researchers call ”emergent behavior”: unanticipated patterns of action arising from complexity itself, not from any deliberate overall design. Emergent behavior is ”bottom-up,” not ”top-down.” Arising naturally, it can enable the system to accomplish amazing feats. Christopher Langton, organizer of the first interdisciplinary a-life conference in 1987, considers emergence to be at the center of how nature works. ”It`s what makes atoms and what makes molecules. You have these little packets of cooperation, and then packets of packets, and packets of packets of packets.”
What a-life scientists do is create simple artificial organisms, program them with simple rules, set them in simple computer environments and study the complex emergent behavior that results. Levy`s book presents an ingenious zoology of a-life organisms: boids, vants (virtual ants), loops, biomorphs, Ramps, anti-Ramps, animats (artificial animals), GOFERs, GAs (genetic algorithms), agents. Levy describes mysterious artificial worlds: the infinite grid, the theoretical ”genetic space” containing all possible life-forms, the VENUS ”prebiotic soup of information” and other custom universes like PolyWorld, Panspermia, AntFarm, Tierra, AL.
Consider biologist Thomas Ray`s Tierra, an attempt at the first ”open-ended” a-life computer environment. Ray conceived freely evolving
creatures that would reproduce, mutate, compete for computer processing time and memory space. His creatures were digital assembly-language programs that both copied their own codes during reproduction (as ”genotypes”) and performed their competitive functions (as ”phenotypes”) with efficiencies that would determine their individual fitness. The oldest or least successful creatures would die at the instructions of a ”reaper” program, ensuring that Ray`s environment did not stagnate.
Ray kicked off his first test with an organism he called ”the ancestor,” 80 instructions long. He didn`t expect much to happen-but, as he puts it, ”I never had to write another creature.” Over millions of generations, his Tierran organisms showed ability to ”yield the drama, and apparently the dynamics, of an evolutionary biosphere.” Biologist Graham Bell called Tierra ”the first logical demonstration of the validity of the Darwinian theory of evolution.” And a 1991 New York Times article reported that, after the Tierran experiment, ”a new round of debate has developed among scientists as to where the dividing line between life and non-life may lie.”
A-life research might seem like a fascinating diversion, the kind of thing geeks routinely and harmlessly misuse computers for-but don`t bet on it. Basic questions are being asked.
– Can a machine reproduce itself? A-life theorists imagine automated factories one day creating products and producing new factories in a kind of perpetual motion.
– Can a computer program itself? A-life points the way toward independent subroutines combining and mutating to produce robust, ”tank-like” programs immune to glitches.
– Where does life come from? A-life research hints at a ”phase-change,” a radical increase in complexity automatically triggering the flashpoint of self-replication.
– How does evolution proceed? A-life environments demonstrate why sex drive and parasitism increase evolutionary efficiency and keep evolving organisms from winding up stuck at ”local maxima.”
– What`s the best way to create artificial intelligence (AI)? According to the a-life model, by working from the bottom up. Levy and his sources consider classic AI, bent on developing complex programs to duplicate high-order intelligent behavior, as essentially misguided. Instead, we should be
”using emergent processes to make machines behave with the novelty and cunning of nature`s creations.”
Not all a-life modeling is done in cyberspace. On the floor of the MIT Mobot laboratory (whose unofficial motto is ”Fast, Cheap, and Out of Control”), a foot-long mobile robot cockroach named Ghenghis teaches itself to walk, prowl a cluttered room, ”stomp over things,” avoid obstacles, freeze while people pass by and pursue prey-”all without benefit of sight, knowledge of what a room is, or a central brain.”
Don`t be too sure about the harmlessness of all this. One form of a-life, the computer virus, has already proven its fitness for survival by confounding the attempts of creators to control or contain it. A-life researchers worry over the possible consequences of releasing ”real” mutating a-life organisms into our environment. AI robots designed according to a-life principles of emergent behavior might emerge with goals of their own-not necessarily in the interests of their human creators. A-life explorers also contemplate the ethical implications of creating life, artificial or otherwise.
Finally, what is life? The same ”stuff”-atoms, molecules, compounds-can be alive or not alive; according to Chris Langton, life is ”a property of the organization of matter, rather than a property of the matter which is so organized.” Biological life is ”wet.” A-life organisms are ”dry”: made of information, silicon, wheels and pulleys, 1`s and 0`s. The difference, which at first seems self-evident and inarguable, fades in importance as you confront the scientific schemes and dreams expertly presented by Steven Levy in this inspired, exciting survey.
