From Wired 3.03 March 1995

N 2088, our branch on the tree of life will come crashing down,  ending a very modest (if critically acclaimed) run on planet Earth. The culprit? Not global warming. Not atomic war. Not FLESH EATING BACTERIA. Not even too much T.V. The culprit is the integrated circuit, aided by the surprising power of exponential growth. We will be driven to extinction by a smarter and more adaptable species, the computer. And our only alternative is to tinker with the very stuff that makes us human, our genes.

Behind this revolution lies a simple story of exponential change. You hear about exponential curves all the time. Exponential inflation is out of control -- running 15%, 25%, 100% a year! Exponential population growth is overwhelming the earth! Yet exponentials don't seem real -- if population growth was out of control, why can I still get a seat on the bus? In fact, humans endure a more or less confined life, far removed from the hurried pace of exponentials. Forty-five Fahrenheit is cold, eighty-five Fahrenheit is warm. Five hundred calories a day, you starve; three thousand a day, you're as fat as a pig. Our lives advance along a very narrow road, and our minds can't grasp even the vaguest concept of rapid, but predictable, change. So how do we know the computers are coming?

If we can't trust human intuition to provide an early warning, consider this simple example of continuous change. Imagine the price of automobiles drops exponentially. At $200,000, a Rolls Royce is large, expensive and unaffordable. You'd never even consider a Rolls as the family sedan. But imagine the Rolls drops in price by a factor of two each year.

After one year it costs $100,000 -- still out of price, out of mind. In the second year at $50,000, the car stays parked in England. In the third year, at $25,000, you start comparison shopping the Rolls against the Taurus. In the fourth year, the kids take one with them to college. After eleven years, the Rolls costs under $100. Now, instead of renting a car on vacation, you buy a Rolls at the airport and leave it with the redcap in lieu of a tip on your return. In twenty years, a Rolls costs under a quarter and they are soon redesigned as ocean breakwaters and highway barriers.

Exponentials start slowly and remain disarmingly out of sight. Yet they build strength relentlessly until they've grown too large to ignore. By then, whole industries have changed and whole cultures have fallen.

The cost and intelligence of computers follows an exponential curve, having improved by a factor of two every 30 months over the last century. That's a factor of one trillion every hundred years, and there is no sign the pace will slow appreciably for another century. Processors that once filled entire floors directing the manufacture of automobile engines, now sit inside V-8s, adjusting valve timing. And by 2088, that box of silicon, wires and plastic will place humans on the endangered species list.

Why 2088? Well, by 2088, the next factor of a trillion enables computers to match human beings in skills and intelligence. In some respects, they already have.

The "operating system" for a person is their genetic code. DNA instructs each cell in your body how to grow, how to deal with infection, and how to wire neurons in the brain to think. The program is about three billion bits long. Sure, three billion bits sounds impressive, but the genetic code is small enough to fit comfortably on a CD-ROM. Like DOS, each new version contains pieces of code with no obvious function, as well as modules incorporating the baggage of previous generations -- the history of evolution is truly written in our genes. But unlike DOS, even some minor code upgrades are worth the price. If you unzipped all three billion odd bits of human DNA and laid it next to DNA from a chimpanzee, all but a fraction of a percent would match exactly. So would the DNA from any two people. It may be hard to believe, but you and Rush Limbaugh are just a few snips of the genetic shears apart.

Still, DNA without a brain is useless. Even common intestinal bacteria take their cue from DNA, but that doesn't make bacteria smart. In fact, an intestine full of bacteria is a bit like an office full of P.C.s; each one contains essentially the same software, has the same amount of memory, and runs the same program needed to eat and reproduce. But a million bacteria are no smarter than a single cell. Chairman Bill, despite his wealth and avaricious charm, has fathered nothing more intelligent than E-coli, mindlessly digesting spread sheet after spread sheet.

But take Bill's dumb computers, wire them together so they can collaborate, and you have the beginning of a new species.

No one is exactly sure how many neurons fill our cranium, but they're thought to number around 100 billion. And each neuron turns on and off about one thousand times a second, and is wired to thousands of other similar cells. If our brain were a computer, it would be rated at 100 trillion operations/second. For a processor today to even come close to matching a human's skills, you'd have to plunk down a hundred million dollars and link 100,000 Pentiums together. Or, you could wait til 2088 when the same amount of processing power will cost peanuts and come in a box that takes a size 8 hat.

By 2088, enough code will exist to fill the silicon brain. Unlike today's software, where each line of code is hand crafted in a desperate attempt to anticipate all possible future uses, tomorrow's software will be inherited from simpler computer ancestors. Some of the code may start out as modules intended to help factory cameras track parts. These modules will be recycled for the computer's eyes. Weather forecasting models will join the brain's genetic program, helping the computer anticipate changes in its environment. Trading software from Wall Street will sharpen its negotiation skills. Some parts will literally write themselves, as the computer varies its genetic code and keeps only those changes it judges beneficial. Like our own DNA, the computer's genetic code will betray its heritage from millions of experiments. Unlike our genetic code, the computer won't have to die if the experiment goes bad -- it will just issue an "undo" command and start afresh.

The scary thing isn't that computers will match our intelligence by 2088, the scary thing is that this exponential curve keeps on going, and going, and going. By 2090 the computer will be twice as smart and twice as insightful as any human being. It will never lose a game of chess, never forget a face, never underestimate the lessons of history. A factor of two in nature conveys an enormous advantage -- after all, a lion only twice as fast as its prey can eat very, very well.

Slowly, computers will evolve their own unique culture -- reminiscent of ours, but strangely distant. They will be like tourists in a foreign land, speaking a curious language and marveling at the quaint local customs. By the year 2100, the gap will grow to the point where homo sapiens, relatively speaking, might make a good pet. Then again, the computers of 2088 might not give us a second thought.

What's a poor human to do? We might fight back, smashing integrated circuit fab lines, but society couldn't function without our silicon co-dependents. No integrated circuits? Then no Swatch watches, no low polluting cars, no credit cards. We need integrated circuits as much as they need us. Anyway, the threat will be obvious only after it's too late, so the fabs will keep on spitting out 486s and DRAMs by the billions. If we had a little time, say another couple of hundred million years, evolution might have time to kick in. After all, evolution propelled mammals out from under the feet of dinosaurs into the canyons of Manhattan; perhaps Darwin's great insight could get us out of this mess as well.

Unfortunately, the benign pruning of human genes by evolution's hand is a bit too slow and undirected to serve our purpose. A hundred years isn't much time for evolution to work its magic. So mankind is fortunate gene splicing arose at the same moment in history as the computer.

Today, gene splicing coaxes vats of bacteria into producing human insulin by the gallon. The first tentative steps repairing human genetic disease have succeeded, "upgrading" a child's genetic code beyond their parents' initial, but flawed, design. Eventually, DNA engineering will be commonplace. People will snip out genes to regrow damaged nerves from a car accident or modify others to control their cholesterol. You might even snip in the genes from a firefly, so your hands can glow gently at night while reading the latest thriller.

Tampering with nature seems foolhardy, but there really isn't any choice. Humans are programming computers today that will someday take our place in nature -- it would be foolhardy not to program our own genes in response.

So next time you power down a computer at night, do so with pleasure. In 2088, the computers are shipped without an "off" switch.

Greg Blonder is a scientist in the communications industry. He works with computers all day long and enjoys their company. Really.

Comments can be sent to author here.

 

---

Author's Note: February 1995

"Faded Genes" first appeared in WIRED 3.03. Originally entitled "The End of Homo Sapiens -- Or Don't Answer The Phone If Mr. Coffee Rings", the article was planned as a 3000 word essay, weaving together projections of future computer capabilities alongside a parallel story of life in 2088. However, Idee Forte runs under 1000 words, so the article had to be completely rewritten for publication in WIRED. That version was edited for style and length but lost some of the connective tissue in the process. The enclosed version restores the cut material and fixes a few of the typos. But it was fun to be published among the colorful clutter of WIRED, even if the message was slighted.

Actually, my extrapolations indicated the cross-over would be 2044, but internal reviewers were skeptical, so we went with 2088. Since the point of the essay was the difficulty of grasping exponential improvement, the exact date was of secondary interest. Well, it matters for the robot apocalypse, but not for the article.

Two subjects more fully discussed in the planned essay are the role of networking and the possibility the silicon curve will peak before human intelligence is reached. For example, the hundreds of millions of DOS machines in the world are little more than the silicon equivalent of yeast; each cell is loaded with the same genetic code, and perform the same dull functions. Not a bad way to brew beer, but no way to write the Great American Novel. But, with the rise in networking represented by the Internet, the DOS machines could be linked into one large "brain," much as the nerves in our brain link together many "dumb" identical cells into a much more capable whole. The consequences are worth considering.

Some people maintain the silicon curve will peak well before human capabilities are reached. But this is more a matter of wishful thinking than a reasonable extrapolation to the future. Conventional silicon still has about 20 years of headroom left before it runs out of steam. By then, our mastery of silicon technology will extend transistors into the third dimension with richly connected wiring -- dramatically increasing its capabilities. After silicon matures, organic-based transistors will follow. While the brain is a wonderful machine, it must play out a hand dealt by evolution. This includes carrying electrical currents while immersed in a vat of conducting liquids, and limiting switching signals to a few thousand operations per second. On both counts silicon is superior. That next factor of a trillion is a dead cert...