Essays. FSF Columns

by Bruce Sterling

  How about a frozen human embryo? It can be just as dormant as a dormant virus, and certainly can’t survive without a host, but it can become a living human being. Some people who were once frozen embryos may be reading this magazine right now! Is a frozen embryo “alive” — or is it just the potential for life, a genetic life-program halted in mid-execution?

  Bacteria are simple, as living things go. Most people however would agree that germs are “alive.” But there are many other entities in our world today that act in lifelike fashion and are easily as complex as germs, and yet we don’t call them “alive” — except “metaphorically” (whatever that means).

  How about a national government, for instance? A government can grow and adapt and evolve. It’s certainly a very powerful entity that consumes resources and affects its environment and uses enormous amounts of information. When people say “Long Live France,” what do they mean by that? Is the Soviet Union now “dead”?

  Amoebas aren’t “mortal” and don’t age — they just go right on splitting in half indefinitely. Does that mean that all amoebas are actually pieces of one super-amoeba that’s three billion years old?

  And where’s the “life” in an ant-swarm? Most ants in a swarm never reproduce; they’re sterile workers — tools, peripherals, hardware. All the individual ants in a nest, even the queen, can die off one by one, but as long as new ants and new queens take their place, the swarm itself can go on “living” for years without a hitch or a stutter.

  Questioning “life” in this way may seem so much nit-picking and verbal sophistry. After all, one may think, people can easily tell the difference between something living and dead just by having a good long look at it. And in point of fact, this seems to be the single strongest suit of “Artificial Life.” It is very hard to look at a good Artificial Life program in action without perceiving it as, somehow, “alive.”

  Only living creatures perform the behavior known as “flocking.” A gigantic wheeling flock of cranes or flamingos is one of the most impressive sights that the living world has to offer.

  But the “logical form” of flocking can be abstracted from its “material manifestation” in a flocking group of actual living birds. “Flocking” can be turned into rules implemented on a computer. The rules look like this:

  1. Stay with the flock — try to move toward where it seems thickest.

  2. Try to move at the same speed as the other local birds.

  3. Don’t bump into things, especially the ground or other birds.

  In 1987, Craig Reynolds, who works for a computer-graphics company called Symbolics, implemented these rules for abstract graphic entities called “birdoids” or “boids.” After a bit of fine-tuning, the result was, and is, uncannily realistic. The darn things flock!

  They meander around in an unmistakeably lifelike, lively, organic fashion. There’s nothing “mechanical” or “programmed-looking” about their actions. They bumble and swarm. The boids in the middle shimmy along contentedly, and the ones on the fringes tag along anxiously jockeying for position, and the whole squadron hangs together, and wheels and swoops and maneuvers, with amazing grace. (Actually they’re neither “anxious” nor “contented,” but when you see the boids behaving in this lifelike fashion, you can scarcely help but project lifelike motives and intentions onto them.)

  You might say that the boids simulate flocking perfectly — but according to the hard-dogma position of A-Life enthusiasts, it’s not “simulation” at all. This is real “flocking” pure and simple — this is exactly what birds actually do. Flocking is flocking — it doesn’t matter if it’s done by a whooping crane or a little computer-sprite.

  Clearly the birdoids themselves aren’t “alive” — but it can be argued, and is argued, that they’re actually doing something that is a genuine piece of the life process. In the words of scientist Christopher Langton, perhaps the premier guru of A-Life: “The most important thing to remember about A-Life is that the part that is artificial is not the life, but the materials. Real things happen. We observe real phenomena. It is real life in an artificial medium.”

  The great thing about studying flocking with boids, as opposed to say whooping cranes, is that the Artificial Life version can be experimented upon, in controlled and repeatable conditions. Instead of just observing flocking, a life-scientist can now do flocking. And not just flocks — with a change in the parameters, you can study “schooling” and “herding” as well.

  The great hope of Artificial Life studies is that Artificial Life will reveal previously unknown principles that directly govern life itself — the principles that give life its mysterious complexity and power, its seeming ability to defy probability and entropy. Some of these principles, while still tentative, are hotly discussed in the field.

  For instance: the principle of bottom-up initiative rather than top-down orders. Flocking demonstrates this principle well. Flamingos do not have blueprints. There is no squadron-leader flamingo barking orders to all the other flamingos. Each flamingo makes up its own mind. The extremely complex motion of a flock of flamingos arises naturally from the interactions of hundreds of independent birds. “Flocking” consists of many thousands of simple actions and simple decisions, all repeated again and again, each action and decision affecting the next in sequence, in an endless systematic feedback.

  This involves a second A-Life principle: local control rather than global control. Each flamingo has only a vague notion of the behavior of the flock as a whole. A flamingo simply isn’t smart enough to keep track of the entire “big picture,” and in fact this isn’t even necessary. It’s only necessary to avoid bumping the guys right at your wingtips; you can safely ignore the rest.

  Another principle: simple rules rather than complex ones. The complexity of flocking, while real, takes place entirely outside of the flamingo’s brain. The individual flamingo has no mental conception of the vast impressive aerial ballet in which it happens to be taking part. The flamingo makes only simple decisions; it is never required to make complex decisions requiring a lot of memory or planning. Simple rules allow creatures as downright stupid as fish to get on with the job at hand — not only successfully, but swiftly and gracefully.

  And then there is the most important A-Life principle, also perhaps the foggiest and most scientifically controversial: emergent rather than prespecified behavior. Flamingos fly from their roosts to their feeding grounds, day after day, year in year out. But they will never fly there exactly the same way twice. They’ll get there all right, predictable as gravity; but the actual shape and structure of the flock will be whipped up from scratch every time. Their flying order is not memorized, they don’t have numbered places in line, or appointed posts, or maneuver orders. Their orderly behavior simply emerges, different each time, in a ceaselessly varying shuffle.

  Ants don’t have blueprints either. Ants have become the totem animals of Artificial Life. Ants are so ‘smart’ that they have vastly complex societies with actual institutions like slavery and and agriculture and aphid husbandry. But an individual ant is a profoundly stupid creature. Entomologists estimate that individual ants have only fifteen to forty things that they can actually “do.” But if they do these things at the right time, to the right stimulus, and change from doing one thing to another when the proper trigger comes along, then ants as a group can work wonders.

  There are anthills all over the world. They all work, but they’re all different; no two anthills are identical. That’s because they’re built bottom-up and emergently. Anthills are built without any spark of planning or intelligence. An ant may feel the vague instinctive need to wall out the sunlight. It begins picking up bits of dirt and laying them down at random. Other ants see the first ant at work and join in; this is the A-Life principle known as “allelomimesis,” imitating the others (or rather not so much “imitating” them as falling mechanically into the same instinctive pattern of behavior).

  Sooner or later, a few bits of dirt happen to p
ile up together. Now there’s a wall. The ant wall-building sub-program kicks into action. When the wall gets high enough, it’s roofed over with dirt and spit. Now there’s a tunnel. Do it again and again and again, and the structure can grow seven feet high, and be of such fantastic complexity that to draw it on an architect’s table would take years. This emergent structure, “order out of chaos,” “something out of nothing” — appears to be one of the basic “secrets of life.”

  These principles crop up again and again in the practice of life-simulation. Predator-prey interactions. The effects of parasites and viruses. Dynamics of population and evolution. These principles even seem to apply to internal living processes, like plant growth and the way a bug learns to walk. The list of applications for these principles has gone on and on.

  It’s not hard to understand that many simple creatures, doing simple actions that affect one another, can easily create a really big mess. The thing that’s hard to understand is that those same, bottom-up, unplanned, “chaotic” actions can and do create living, working, functional order and system and pattern. The process really must be seen to be believed. And computers are the instruments that have made us see it.

  Most any computer will do. Oxford zoologist Richard Dawkins has created a simple, popular Artificial Life program for personal computers. It’s called “The Blind Watchmaker,” and demonstrates the inherent power of Darwinian evolution to create elaborate pattern and structure. The program accompanies Dr. Dawkins’ 1986 book of the same title (quite an interesting book, by the way), but it’s also available independently.

  The Blind Watchmaker program creates patterns from little black-and-white branching sticks, which develop according to very simple rules. The first time you see them, the little branching sticks seem anything but impressive. They look like this:

  Fig 1. Ancestral A-Life Stick-Creature

  After a pleasant hour with Blind Watchmaker, I myself produced these very complex forms — what Dawkins calls “Biomorphs.”

  Fig. 2 — Six Dawkins Biomorphs

  It’s very difficult to look at such biomorphs without interpreting them as critters — something alive-ish, anyway. It seems that the human eye is trained by nature to interpret the output of such a process as “lifelike.” That doesn’t mean it is life, but there’s definitely something going on there.

  What is going on is the subject of much dispute. Is a computer-simulation actually an abstracted part of life? Or is it technological mimicry, or mechanical metaphor, or clever illusion?

  We can model thermodynamic equations very well also, but an equation isn’t hot, it can’t warm us or burn us. A perfect model of heat isn’t heat. We know how to model the flow of air on an airplane’s wings, but no matter how perfect our simulations are, they don’t actually make us fly. A model of motion isn’t motion. Maybe “Life” doesn’t exist either, without that real-world carbon-and-water incarnation. A-Life people have a term for these carbon-and-water chauvinists. They call them “carbaquists.”

  Artificial Life maven Rodney Brooks designs insect-like robots at MIT. Using A-Life bottom-up principles — “fast, cheap, and out of control” — he is trying to make small multi-legged robots that can behave as deftly as an ant. He and his busy crew of graduate students are having quite a bit of success at it. And Brooks finds the struggle over definitions beside the real point. He envisions a world in which robots as dumb as insects are everywhere; dumb, yes, but agile and successful and pragmatically useful. Brooks says: “If you want to argue if it’s living or not, fine. But if it’s sitting there existing twenty-four hours a day, three hundred sixty-five days of the year, doing stuff which is tricky to do and doing it well, then I’m going to be happy. And who cares what you call it, right?”

  Ontological and epistemological arguments are never easily settled. However, “Artificial Life,” whether it fully deserves that term or not, is at least easy to see, and rather easy to get your hands on. “Blind Watchmaker” is the A-Life equivalent of using one’s computer as a home microscope and examining pondwater. Best of all, the program costs only twelve bucks! It’s cheap and easy to become an amateur A-Life naturalist.

  Because of the ubiquity of powerful computers, A-Life is “garage-band science.” The technology’s out there for almost anyone interested — it’s hacker-science. Much of A-Life practice basically consists of picking up computers, pointing them at something promising, and twiddling with the focus knobs until you see something really gnarly. Figuring out what you’ve seen is the tough part, the “real science”; this is where actual science, reproducible, falsifiable, formal, and rigorous, parts company from the intoxicating glamor of the intellectually sexy. But in the meantime, you have the contagious joy and wonder of just gazing at the unknown the primal thrill of discovery and exploration.

  A lot has been written already on the subject of Artificial Life. The best and most complete journalistic summary to date is Steven Levy’s brand-new book, ARTIFICIAL LIFE: THE QUEST FOR A NEW CREATION (Pantheon Books 1992).

  The easiest way for an interested outsider to keep up with this fast-breaking field is to order books, videos, and software from an invaluable catalog: “Computers In Science and Art,” from Media Magic. Here you can find the Proceedings of the first and second Artificial Life Conferences, where the field’s most influential papers, discussions, speculations and manifestos have seen print.

  But learned papers are only part of the A-Life experience. If you can see Artificial Life actually demonstrated, you should seize the opportunity. Computer simulation of such power and sophistication is a truly remarkable historical advent. No previous generation had the opportunity to see such a thing, much less ponder its significance. Media Magic offers videos about cellular automata, virtual ants, flocking, and other A-Life constructs, as well as personal software “pocket worlds” like CA Lab, Sim Ant, and Sim Earth. This very striking catalog is available free from Media Magic, P.O Box 507, Nicasio CA 94946.

  “INTERNET” [aka “A Short History of the Internet”]

  Some thirty years ago, the RAND Corporation, America’s foremost Cold War think-tank, faced a strange strategic problem. How could the US authorities successfully communicate after a nuclear war?

  Postnuclear America would need a command-and-control network, linked from city to city, state to state, base to base. But no matter how thoroughly that network was armored or protected, its switches and wiring would always be vulnerable to the impact of atomic bombs. A nuclear attack would reduce any conceivable network to tatters.

  And how would the network itself be commanded and controlled? Any central authority, any network central citadel, would be an obvious and immediate target for an enemy missile. The center of the network would be the very first place to go.

  RAND mulled over this grim puzzle in deep military secrecy, and arrived at a daring solution. The RAND proposal (the brainchild of RAND staffer Paul Baran) was made public in 1964. In the first place, the network would have no central authority. Furthermore, it would be designed from the beginning to operate while in tatters.

  The principles were simple. The network itself would be assumed to be unreliable at all times. It would be designed from the get-go to transcend its own unreliability. All the nodes in the network would be equal in status to all other nodes, each node with its own authority to originate, pass, and receive messages. The messages themselves would be divided into packets, each packet separately addressed. Each packet would begin at some specified source node, and end at some other specified destination node. Each packet would wind its way through the network on an individual basis.

  The particular route that the packet took would be unimportant. Only final results would count. Basically, the packet would be tossed like a hot potato from node to node to node, more or less in the direction of its destination, until it ended up in the proper place. If big pieces of the network had been blown away, that simply wouldn’t matter; the packets would still stay airborne, lateralled wildly acr
oss the field by whatever nodes happened to survive. This rather haphazard delivery system might be “inefficient” in the usual sense (especially compared to, say, the telephone system) — but it would be extremely rugged.

  During the 60s, this intriguing concept of a decentralized, blastproof, packet-switching network was kicked around by RAND, MIT and UCLA. The National Physical Laboratory in Great Britain set up the first test network on these principles in 1968. Shortly afterward, the Pentagon’s Advanced Research Projects Agency decided to fund a larger, more ambitious project in the USA. The nodes of the network were to be high-speed supercomputers (or what passed for supercomputers at the time). These were rare and valuable machines which were in real need of good solid networking, for the sake of national research-and-development projects.

  In fall 1969, the first such node was installed in UCLA. By December 1969, there were four nodes on the infant network, which was named ARPANET, after its Pentagon sponsor.

  The four computers could transfer data on dedicated high-speed transmission lines. They could even be programmed remotely from the other nodes. Thanks to ARPANET, scientists and researchers could share one another’s computer facilities by long-distance. This was a very handy service, for computer-time was precious in the early ’70s. In 1971 there were fifteen nodes in ARPANET; by 1972, thirty-seven nodes. And it was good.

  By the second year of operation, however, an odd fact became clear. ARPANET’s users had warped the computer-sharing network into a dedicated, high-speed, federally subsidized electronic post-office. The main traffic on ARPANET was not long-distance computing. Instead, it was news and personal messages. Researchers were using ARPANET to collaborate on projects, to trade notes on work, and eventually, to downright gossip and schmooze. People had their own personal user accounts on the ARPANET computers, and their own personal addresses for electronic mail. Not only were they using ARPANET for person-to-person communication, but they were very enthusiastic about this particular service — far more enthusiastic than they were about long-distance computation.