Note- this article was written in 1995 for WIRED magazine, as a follow-on to my Faded Genes piece. Although light-hearted in tone, the subject matter is quite serious. Large, uniform, connected systems can be very efficient and reliable. Until they fail catastrophically.
I wrote the piece while at AT&T Bell Labs, but the article never saw the light of day. Unfortunately, a senior public relations executive withheld publication permission because they didn't want AT&T's name to appear in the same article with a mention of HIV/AIDs...
Still, a glance at the news demonstrates its relevance today. So here it is- unedited and locked in the concerns of the 90's.
Greg Blonder- 2009
Too Much Of A Good Thing
There are many kinds of instabilities in our world. There are the mechanical instabilities a pencil feels, balanced on its tip, before falling slowly to the ground. There are mankind's ageless political instabilities, born of human frailty, which tear apart society. And now there is a new kind of instability-- seductive, powerful, dangerous and swift-- based on the too-rapid dispersal and adoption of good ideas.
Good ideas? We're in trouble because people copy good ideas? Well, if history is any guide, people are getting too smart, too fast for their own good. And this new instability, the instability of continuous improvement, may be upon us sooner than we think.
Instabilities by their very nature amplify a small change until it spirals out of control. Like a gentle breeze collapsing a house of cards. Or the world economy plummeting into depression after the stock market crash of 1928.
Historically, an instability's most potent weapon was its innocuousness: who noticed that last thin straw before it broke the camel's back? Yet society always found ways to restrain the winds of unpredictable change. So what if Joe overloaded his camel? The next driver learned from Joe's mistakes, and made sure to weigh the straw before packing. Others observed the broken camel and took a different tack, keeping a couple of spare camels around "just in case". Learning from each other's mistakes, and borrowing good ideas is society's way to cope- it's the cultural equivalent of an immune system. So why will this immunity fail us in the future?
The problem is time, or rather the absence of time to react to change. On the Internet, the world moves just a little faster, ideas are spread just a little further, and products are adopted just slightly ahead of their time. And just slightly before the whole system has a chance to react. In fact, the Internet instability arises from the confluence of three independent streams, all turbulent on their own, but devastating en mass.
They are instabilities of "connectivity", "uniformity", and "speed", or "c|u|sp".
Living on a c|u|sp may spark global disaster, but a c|u|sp instability can disturb even the most pedestrian of activities. Remember when you discovered that little neighborhood bistro with its wonderful food and great atmosphere? It was a secret you shared with your best friends, but hoped no one else would discover. What happened after publication of a well deserved review, raving about the house special crab cake? Everyone, of course, following the best possible advice, called for a reservation. And, equally well informed, ordered the special. Soon, your quiet little bistro was overwhelmed by customers. The atmosphere and food suffered irreparable harm, and worst of all, ran out of crab cakes by 6 p.m.
The "Hot Restaurant" dilemma illustrates all three factors leading to run away instabilities. The first is "connectivity". If an entire town could not drive to the bistro, the restaurant would not have been overwhelmed by customers. The second factor is "uniformity". Compared to the slow and varied advice generated by word of mouth, all newspaper readers saw the same review, trusted the same counsel, and followed its recommendations. This created a pulse of interest that propagated throughout the town. Lastly, there is "speed". If customers had waited a few months before trying the bistro, the owners would have had time to adjust and slowly learn to deal with crowds. But there just wasn't enough time.
To appreciate the raw power of "connectivity", "uniformity" and "speed", it's best to view them in isolation.
Consider "connectivity", and how it's responsible for America's fear of crime. Every night, the local news leads off with a story about a violent crime in America. With perfect communications, one terrible crime efficiently serves the prurient needs of an entire nation! Worried about a killer on the loose in California, doors three thousand miles away slam shut at night, locking out a perceived, but ethereal threat. Children grow to fear their neighbors, interpreting differences in color or culture as a sign of danger. Despite strong evidence of stable or decreasing crime rates, newscasters always manage to find one frightening story to anchor the evening news. And thus to mislead a country's intuition.
This same disease infects the Internet. With perfect communications, a few pornographic sites appear to dominate the content of the web. Is it any wonder the cries for regulation are as strong as the demands for more police on the street?
"Uniformity" is the second contributor. In biology, a genetically uniform crop is called a "monoculture", which farmers often encourage by dedicating their entire acreage to a single plant. Specializing in one variety is often a good idea- harvesting a field is simpler when all the corn ripens at once, farmers can spread the cost of developing new seed across a wider base, and yields are often higher. Besides, this is the same seed your friends and county extension agent recommended- why bet the farm on anything else?
But the dangers of a monoculture are severe. In 1845, the potato crop in Ireland failed. Potatoes nestled in the ground and potatoes carefully stacked in barrels rotted and grew malodorous. Fully a third of Ireland depended on this one food for their staple diet, but each year the potato crop failed. The "Great Hunger" dragged on through 1846, 1847, 1848 and 1849. Two million people died, and millions more emigrated to America. All because Ireland depended on just two species of potatoes- both of which fell victim to the potato blight Phytophton infestans.
Combined, the first two elements of c|u|sp form a deadly pair, no more so than we are experiencing in the recent AIDS epidemic. Without the network of roads and planes linking all human immune systems together, AIDS might have confined its damage to a few villages in Africa. Excellent communication, generally a positive tool, contributed to this disaster. And, if people were not made of the same genetic material, the virus could not move from victim to victim: uniformity is a fertile breeding ground for disease. So why is AIDS still spreading, despite all the dire warnings broadcast to the global village? Well, most people will voluntarily avoid exposure to a deadly virus, but sex is such a good idea...
Fortunately, AIDS is a slowly moving virus- granting us time to develop new drugs, launch public health programs, and build up natural resistance. But what if AIDS was virulent, what if it could be caught as fast as the common cold, causing death in weeks? This is the danger of "speed".
AT&T's telephone network is a superbly engineered organism, with high guaranteed reliability against all possible contingencies. Switches are designed to be continuously available, with less than one minute/year of downtime. One minute a year! Compared to the Internet, where "Server not available" is displayed more frequently than commercials on T.V., the telephone network approaches a state of nirvana. But in January 1990, disaster struck.
For sound reasons of operational efficiency, all 114 switches on the network run the same code (uniformity) and are connected to their neighbors for redundancy (communications). When any of these 114 switches detects a potential fault, it takes itself off-line for a full reset, and sends out a message to its neighbors to reroute all calls in the interim. Six seconds later, it comes back on-line, and starts accepting new calls. So far, so good. A model of operational efficiency based on years of sound, incremental improvements.
But in December 1989 a few lines of code were changed to add another layer of redundancy and to shave yet another second or two off potential downtime. When a New York switch detected a fault on January 15, 1990 and came back up six seconds later, other switches in the network were listening, waiting to restore network traffic. Unfortunately, on rising from the dead it sent out two "Hi. I'm O.K." messages, just 1/100th of a second apart. This confused the other switches, which were expecting only a single message, and they took themselves out of service for a refreshing on/off cycle. Of course, when they came back up, each switch repeated the same script, and sent out its own doublet of messages 1/100th of a second apart. Which disabled yet more switches. In a few minutes, the entire network was bobbing up and down in a wild orgy of message passing. And no longer accepting anyone's phone calls...
Given the freakishness of a c|u|sp event, AT&T's quick and knowledgeable response was amazing. They learned from this disaster and modified their network to trap future chain reactions.
But why should this experience trouble the Internet? It's much less uniform and more fluid than AT&T's network. Each day, millions of watchful eyes are looking for problems and improving the system. This high "genetic variability" will provide enough immunity. Or will it?
Surprisingly, the personal computer industry hawks one of the least innovative products of the 20th century. Underneath a computer's hood rumbles an engine powered by the semiconductor industry- improving the price/performance ratio of computation by over a factor of one hundred each decade. Yet a p.c. still costs around $2000, and runs the same four or five applications (spreadsheets, word processors, games and communications) demonstrated by industry pioneers in the late 1960s. Worse yet, with perfect communications, no new idea goes unnoticed, and all good ideas are adopted industry wide. A popular feature in one spread sheet is immediately adopted by its competitors, leading to the common observation that all such programs look and feel the same. If a small company makes a splash with a disk compression program one year, next year it is built in to the operating system. Or becomes a standard like TCP/IP. The Internet is more uniform than you think.
Imagine a world where all communications are secure and private, thanks to RSA encryption. Security and privacy- that seems like a good idea. RSA is thought to be essentially unbreakable, and more and more companies are incorporating it within their software's genetic code. It's so reliable, an encrypted document can be left in plain view without the slightest worry. Now image half of all economic commerce takes place on the net, with your entire bank account and pension plan drifting with the rising financial tide.
But then one day a c|u|sp occurs, and RSA is compromised. In minutes worldwide financial activity is halted. Your bank account (when it is reconstructed from incomplete backup records two months later) is half as large as it should be. And your company folds because competitors can easily read its proprietary documents.
If not RSA, what about Word? If we all used Word 6.0, and someone discovered a way to drop a Trojan horse in amongst the tables of a document, the damage may propagate out of control before a remedy is found. Say, a table of Olympic scores picked up off the Internet in 1996. Its not much fun waking up one morning and discovering all your files were erased overnight.
Some popular sites on the Internet are visited a million times a day. Today, most people take back souvenirs of images and text. But tomorrow, with Java and its offspring mesmerizing visitors with cool animation, Internet tourists will download small programs. Java is modestly secure, but imperfect. The worst scenario is for Java to be widespread before it breaks- by then millions of machines will be infected with small programs that could wreak havoc. Communications, uniformity and speed- made attractive by copying good ideas.
Can the Internet be saved? Well, one possibility is to stop the world from contributing any new good ideas. Just let a single organization run the whole web. Microsoft is a particularly safe choice- when was the last time it had an original idea? And speed to market won't be an issue either. But of course, it might be 1998 before the bugs from "WInternet96" are exterminated...
Another possibility it to think global, and act local. All your friends use Netscape 2.0; why don't you experiment with another browser? The world is full of interesting, quirky shareware waiting to be adopted and given a warm home. Encourage variety to live on your hard disk. And isn't it a bit disquieting to realize 80% of all computers use Intel processors? You might want to try a Motorola or MIPs machine- and celebrate the smaller discounts you'll receive as a form of disaster insurance.
We'd better learn to cultivate variety and live in islands of diversity. After all, the house special won't always be crab cake.