Living Machines – Technology and biology are converging fast. The result will transform everything from engineering to art – and redefine life as we know it.

The New Facts of Life
Scientific advances point to a startling conclusion: The nonliving world is very much alive.

Wired.com

By Christopher Meyer, Jason Lohn, Karl Jacob, Dick Morley, Shana Ting Lipton, Marco Dorigo, Avery Pennarun

http://www.wired.com/wired/archive/12.02/machines.html?tw=wn_tophead_5

by Christopher Meyer

Copernicus demoted humanity by removing Earth from the center of the universe. Darwin showed that, rather than being made in God’s image, people were merely products of nature’s experimentation. Now, advances in fields as disparate as computer science and genetics are dealing our status another blow. Researchers are learning that markets and power grids have much in common with plants and animals. Their findings lead to a startling conclusion: Life isn’t the exception, but the rule.

The notion that the inorganic world is alive is as old as mythology; think of Poseidon, the Greek personification of the sea. However, the tools available to examine life at its most essential – DNA sequencing, bioinformatics, gene chips – are new. We’re beginning to discern life processes at their fundamental level, and as we re-create these processes in silico, we’re starting to see how they work in inorganic settings. It turns out that many of life’s properties – emergence, self-organization, reproduction, coevolution – show up in systems generally regarded as nonliving.

EMERGENCE describes the way unpredictable patterns arise from innumerable interactions between independent parts. An organism’s behavior, for instance, is driven by the interplay of its cells. Similarly, weather develops from the mixing of oxygen, carbon dioxide, water, and other molecules.

SELF-ORGANIZATION is a basic emergent behavior. Plants and animals assemble and regulate themselves independent of any hierarchy for planning or management. Digital simulations made up of numerous software agents have demonstrated self-organization in systems ranging from computer networks to tornadoes.

REPRODUCTION was considered strictly the purview of organisms until recently. Now computer programs procreate, too. Genetic algorithms mimic biology’s capacity for innovation through genetic recombination and replication, shuffling 1s and 0s the way nature does DNA’s Gs, Ts, As, and Cs, then reproducing the best code for further recombination. This technique has been used to evolve everything from factory schedules to jet engines.

COEVOLUTION inevitably accompanies evolution. When an organism evolves in response to environmental change, it puts new pressures on that environment, which likewise evolves, prompting further evolution in the organism. This cycle occurs in many social systems – for instance, the interaction between behavioral norms and legal codes.

These life properties are already being built into real-world devices, like Sony’s robotic dog Aibo; put two of them together and their personalities will coevolve. The line between organisms and machines is beginning to blur.

Consider a hypothetical pod of Predator drones. Each unmanned aerial vehicle monitors terrain, weather, and potential threats, and continuously receives target updates and transmits its findings via satellite. The drones are motivated by two rules: hunger for valuable intelligence and repulsion from other drones to minimize redundant observation. These rules enable the UAVs to direct themselves better than any dispatcher could. Other rules help the fleet survive. When a drone observes a hostile signature – missile, rocket-propelled grenade, rifle fire – it executes an evasive maneuver from a stored repertoire. As Predators live and die by these rules, they generate new information about fitness under various conditions. Genetic algorithms use this data to breed more effective rules. Predators are connected, so if one is shot down over Afghanistan, all drones everywhere gain improved responses to that form of attack. This is precisely how bacteria develop resistance to antibiotics, only faster.

This sounds far-out until you realize that something similar is already happening on your desktop, when Norton AntiVirus updates virus definitions automatically over the Internet. In fact, networks could play a critical role as machines come to resemble living creatures. In life, as on the Net, connections matter more than processors. The Internet could allow sensors to interact in emergent ways, forming an autonomic nervous system for the physical world. An early version is taking root in Los Angeles, where sensors at intersections identify approaching buses and ask a central computer whether they’re on time. Late buses get the green light; the system gives crossing traffic extra time in subsequent cycles. The result: 25 percent improvement in transit times without creating congestion.

Oddly enough, our growing knowledge of life processes could have its biggest impact in the social sciences. Social systems, after all, are made up of interacting agents, i.e., people. When we become adept at applying these insights to the social sphere, we’ll be able to run simulations that reveal, say, the conditions under which Iraq would reconstruct itself. At that point, the new science of life will help us not only live better, but live better together.

Christopher Meyer is coauthor of It’s Alive: The Coming Convergence of Information, Biology, and Business.

Design by Darwin
Artificial evolution beats an engineering degree.
by Jason Lohn

No team of engineers, no matter how much time they took or how many bottles of cabernet they consumed, would dream up an antenna that looked like a deer antler on steroids. Yet that’s what a group at NASA Ames Research Center came up with – thanks to a little help from Darwin.

NASA’s Space Technology 5 nanosatellites, which are scheduled to start measuring Earth’s magnetosphere in late 2004, requires an antenna that can receive a wide range of frequencies regardless of the spacecraft’s orientation. Rather than leave such exacting requirements in the hands of a human, the engineers decided to breed a design using genetic algorithms and 32 Linux PCs. The computers generated small antenna-constructing programs (the genotypes) and executed them to produce designs (the phenotypes). Then the designs were evaluated using an antenna simulator. The team settled on the form pictured here.

You won’t find this kind of antenna in any textbook, design guide, or research paper. But its innovative structure meets a challenging set of specifications. If successfully deployed, it will be the first evolved antenna to make it out of the lab and the first piece of evolved hardware ever to fly in space.

Jason Lohn chaired the 2003 NASA/DoD Conference on Evolvable Hardware.

Cracking the Email Genome
Want to stop the spam plague? Try gene therapy.
by Karl Jacob

No one would say that spam is alive, but it might as well be. It infests email servers like cockroaches in a Manhattan apartment and multiplies even more prodigiously. And, like roach motels, most antispam programs barely make a dent in the population. In fact, the best way to block junk email is to treat it as though it is alive – as though it has genes that distinguish it from the kind of email you want to read.

Most spam filters work by blocking email addresses that have been identified as sources of unwanted messages. Many also search for a statistically significant presence of words associated with sales pitches. However, these methods require lots of manual upkeep and fail as often as they succeed, either letting spam slip through or blocking higher-priority communications. Spam can’t be defined by words like Viagra – or V.ia_gr^a – any more than humans can be distinguished from other mammals by how many limbs we have. The essence of spam is structural. Recognize the structure and you can catch the spam.

Like strands of DNA, email messages have a standard data format that amounts to a genome for legitimate email. Spammers exploit and mutate email genes to obscure the origin or content of their messages, creating distinctive spam genes. Just as the Human Genome Project used massive computing power to search for telltale patterns in a bafflingly large and complex data set, my company, Cloudmark, is turning its network loose on spam. Our users identify and submit for analysis 130 million spam messages a day, a gargantuan processing load that’s distributed across their 700,000 desktops. So far, we’ve derived more than 300 genes that enable our software to distinguish spam from nonspam.

What does a spam gene look like? In spam, various iterations of the sender’s address often don’t match, obscuring the message’s origin – a gene known as a forged header. Genes of the character histogram class disguise words by repeating letters or inserting symbolic characters or spaces. The body obscured and HTML comment genes hide a portion of the content; the user sees the spiel, but the filter can’t differentiate it statistically. The base 64 encoding gene translates the message into binary code, which spam filters can’t read but email clients can.

Legitimate email tends to have different genes. Headers match. ID numbers inserted as the message bounces from server to server are consistent and properly formatted. The body often includes quotations from earlier communications.

The genetic approach has made it possible for Cloudmark to identify spam with better than 98 percent accuracy. And our system is continually improving: Whenever it mistakes a legit message for spam, a program called the Evolution Engine mutates the spam genes involved and sends the misidentified message back through the filter until it classifies the message correctly. Result: an increasingly precise definition of the spam genome, and thus increasingly effective filtering.

If spam is a life-form, it falls below slime molds in the phylogenetic chart. But it’s certainly an emergent phenomenon that coevolves with the Internet. As spam blockers develop more sophisticated techniques, selective pressure is bound to produce new strains of superspam. Someday we’ll find ourselves battling not the electronic version of junk mail but the more capable cousin of the computer virus – a vital force with priorities of its own.

Karl Jacob is CEO of Cloudmark.

Beast on Wheels
Don’t kick the tires – They might kick you back.
by Dick Morley

A car takes less care and feeding than any beast of burden, but it’s got drawbacks. It doesn’t adjust to wet roads or rough terrain. It doesn’t repair itself when it’s worn or damaged. It makes little effort to protect itself from catastrophe (imagine trying to jump a horse across the Grand Canyon). And while cars do evolve and reproduce – with the help of parasitic engineers and hivelike factories – one generation passes little evolutionarily useful information to the next.

Research into artificial intelligence aims to make machines more responsive to their environments. The AI method, however, has been to program devices to react to specific events, creating machines that are unable to cope with unexpected circumstances. Complexity theory offers a different perspective. If a car were designed like a living thing – as a collection of components wired to regulate one another in response to external stimuli, like organs mediated by a nervous system – it would act more like a living thing.

How would such a car be put together? Its organs would be low-powered, single-chip devices governed by a small number of rules. The steering column wouldn’t physically connect to the wheels; each wheel would be independent, with its own software agents for steering, braking, and suspension. The nervous system connecting these agents would be wireless, automated, and fully electric, with solenoids and servos rather than gears, levers, and hydraulics. Sensory organs would be embedded in the vehicle’s skin from bumper to bumper. This year’s designs would feed a constant stream of real-world performance data into digital models that would self-evolve into a blueprint for the next-gen fleet.

All this aligns nicely with the evolutionary path automobiles have taken so far. They’re already filled with sensors, processors, and firmware. Vehicles outfitted with OnStar’s wireless information service are networked. GM has thrown its weight behind all-electric, drive-by-wire designs. The market may not be ready for a car with a mind of its own, but many experts in applied complexity agree that the plastic-and-steel beast of burden could be less than a decade away.

[01] Brakes
The braking system is the nexus of the Living Car’s survival instinct. Its software looks at wheel differential, deceleration, front and rear radar, driver responsiveness, and braking effectiveness. It avoids collision by applying the brakes, either automatically or on command.

[02] Suspension
A feedback loop between the CPU and sensors in each wheel governs every aspect of wheel motion. A tire agent adjusts inflation to suit the prevailing road surface and driving style.

[03] Black Box
The black box is more than a diagnostic tool in the event of catastrophic failure. It’s the vehicle’s active memory. Three months of datais always stored on board for use by the CPU; when full, the black box transmits its entire contents back to the factory.

[04] Communications
Ants communicate through scent. The Living Car communicates via satellite. The roof is a virtual satellite dish made up of millions of tiny microchips. Other cars receive location and bearing; the factory gets real-world performance data on which to base future designs.

[05] Steering
There’s no physical link between the steering control and the wheels, so software mediates between them. Code optimizes the driver’s performance by damping oscillation, adjusting the ratio between steering and turning radius, and anticipating driver actions.

[06] Navigation
The car may not be fully able to drive on its own, but it keeps itself on a steady course. The GPS-capable navigation agent downloads maps; optimizes routes depending on traffic, weather, efficiency, and driver taste; and sends travel history to the black box.

[07] Lights
A matrix of adjustable mirrors in the headlights spare oncoming drivers from blinding glare. The rear lights flash and change colors to alert other drivers to potential hazards – from road ice to blowouts – identified by the CPU’s predictive modeling capability.

[08] Engine
Separate agents manage each of the engine’s dynamic components: exhaust and intake valves, fuel injector, alternator, and cooling pump. The goal is to compromise between fuel efficiency and performance, depending on the demands of the driver and control system.

[09] Transmission
The transmission works with the brakes, suspension, and wheels to optimize performance and mileage. It monitors interaction between wheels and road, and provides traction control and braking assistance on rough terrain, steep grades, and the like.

[10] Sensors
A dense web of sensors, including radar and cameras, covers the car’s exterior and monitors its surroundings: humidity, light, temperature, obstacles, other vehicles. The flood of sensing data ensures that the Living Car can cope with whatever happens.

Dick Morley is president of R. Morley Inc., a consultancy specializing in automotive manufacture and design.

Skin Art
Growing tomorrow’s mona lisa in a petri dish.
by Shana Ting Lipton

For most people, taking care of a zit calls for Oxycution. But for biotech artists Oron Catts and Ionat Zurr, it’s about cultivating tissue, not drying it up. Their Semi-Living Wired Zit, shown here, is grown from mouse scar tissue. The duo from Perth, Australia, creates work that blurs the line between art and science, in this and other outlandish creations for their project, Tissue Culture & Art. Catts and Zurr are now sculpting a quarter-size ear from human tissue. "We’re interested in the ear as an object," explains Catts. Performance artist Stelarc wants the pair to grow a life-size version so he can have it surgically attached to his head. Vincent van Gogh, where are you when we need you?

Shana Ting Lipton (www.shanatinglipton.com) writes about art and culture on- and offline.

The Swarmbots Are Coming
Ant algorithms get down to business.
by Marco Dorigo

Ants are simple creatures, yet they can perform complicated tasks. They create highways leading to food, organize the distribution of larvae in the anthill, form cemeteries by clustering dead ants, build living bridges to cross gaps in their way, and assign tasks as needed without any centralized control. Thus, ants provide an excellent model for programming simple devices to achieve complex results.

Boil down ant behavior and what do you get? A new set of business tools known as ant algorithms: basic behaviors that can be programmed into a large number of independent software agents to solve human problems.

Consider the way ants forage. When an ant comes across food, it returns to the nest, leaving a scent trail. Other ants follow the trail, find the goods, and carry them back to the nest, reinforcing the path with their own scent, which attracts still more ants. Shorter routes get more traffic, so the scent becomes stronger along these, while it dies away along lesser-used ones. In this way, ants follow the shortest paths between their nest and nearby food sources.

This route-finding capability is remarkably handy. Colonies of simulated ants laying down digital scent trails can find the best way to send delivery trucks through city streets or data packets through communication networks. More generally, ant algorithms can find minimum-cost solutions to a variety of problems in distribution and logistics. Unilever uses them to allocate storage tanks, chemical mixers, and packaging facilities. Southwest Airlines uses them to optimize its cargo operations. Numerous consulting houses, such as the Swiss firm AntOptima, have embraced them as an indispensable tool.

But logistics are just the beginning. Ant algorithms are also being used to control a class of robots called swarm bots. Typically, a swarm bot is a collection of simple robots (s-bots) that self-organize according to algorithms inspired by the bridge-building and task-allocation activities of ants. For example, if an s-bot encounters an object too heavy to carry on its own, other s-bots will grasp either the object or other s-bots until they get it under control. Two or more can link up to cross a gap that exceeds a single s-bot’s stride. As an ad hoc accretion of simple units, a swarm bot’s form depends on its surroundings and the job it’s doing. Such devices might prove helpful in activities like search-and-rescue and planetary exploration.

The ability to swarm, adapt, and optimize makes ant algorithms a crucial technology for the information age, especially as everyday objects become ever smarter. The rules that insects live by turn out to be perfectly suited to the high tech anthill.

Marco Dorigo is the author of Swarm Intelligence and research director of the IRIDIA lab at the Free University of Brussels.

Your New Immune System
How autonomic computing fights digital infection.
by Avery Pennarun

The human immune system is a remarkably flexible and adaptive piece of technology. It can defeat familiar pathogens as well as bugs it has never encountered before. It remembers the signatures of previous attackers so it can fight them more effectively in the future. Moreover, it balances protection and performance, stepping up its defenses gradually until it confirms an emergency and forces you to bed.

Two decades have passed since a student at the University of Southern California created the first computer virus, yet computer scientists are only just beginning to view operating system design from a biological perspective. In fact, that’s the goal of autonomic computing, an approach that mimics the way the central nervous system regulates the body. At Net Integration Technologies, we’ve created such a nervous system for Nitix, our version of Linux. Our software includes the usual networking services – Web, email, file transfer – but adds a tightly interconnected set of control programs called managers. Think of a digital immune system.

Suppose a hacker tries to infiltrate a Nitix network by attempting to guess a valid email password. A conventional server would reject guesses until the hacker either got it right or gave up. In our system, however, the authentication manager notices these failures and tells the firewall manager to slow down responses to the password-guesser, miring the contagion in a mucuslike "tar pit" of delayed responses. This is better than blocking the guesser outright, since it gives legitimate users who have simply forgotten their password a chance to enter while blunting a possible denial-of-service attack.

Just as a biological immune system combats bacteria, viruses, and parasites alike, Nitix’s protection isn’t limited to thwarting hackers. Say a Windows laptop acquires a virus while traveling, returns to the office, and begins transmitting the infection throughout the network. Windows-based viruses often attack by requesting a URL that contains malicious code. Nitix, based on Linux, isn’t susceptible, so a rigged URL returns a "404 Page Not Found" error. To protect other Windows machines, however, the server forwards all such errors to the WWW manager. When this program sees a suspicious URL, it tells the firewall manager to isolate the offending workstation, much like a killer T cell destroys infected tissue. Later, if it sees the same virus signature coming from other machines, it recognizes an epidemic and decrees that all traffic must pass through the firewall. This degrades network performance somewhat but also keeps the virus from spreading.

The autonomic approach makes Nitix more or less self-healing. Just like the biological system on which it is based.

Avery Pennarun is cofounder and vice president of software architecture for Net Integration Technologies.

A lot can happen in a billion years, but look closely and you’ll see the same dynamic at work in every system, and at every scale. Whether the name of the game is microbiology or geopolitics, it all boils down to the delicate balance between competition and cooperation.