Chapter 91 15.9 The end of engineering supremacy

Danny Hillis came to the same conclusion.He said very seriously that he wanted his "connection machine" to evolve into business software. "We want these systems to solve a problem that we only know how to state but not how to solve." One example is how to program millions of lines to fly an airplane.Hillis proposes a swarm system in which tiny parasitic programs try to crash planes to evolve better software for "piloting skills."As his experiments showed, the parasites cause the system to converge more quickly to a navigation program that is error-free and robust to interference."We would rather spend more time programming a better parasite than spend countless hours designing code and troubleshooting things," Hillis said.

Even if a technician succeeds in designing a large program, such as navigation software, it is impossible to test it thoroughly.But what evolved was different. "This software has grown up with thousands of full-time critics," Hillis said, recalling his own parasites, "and anyone who survived them has withstood the rigors." In addition to being able to make things we can't make, evolution has one more thing to boast about: it can make things with fewer defects. "I'd rather fly in an airplane piloted by software that evolved than piloted by software that I wrote myself," said Hillis, a prodigious programmer.

A long-distance telephone company's call-routing program has a total of 2 million lines of code.And three lines of wrong code out of those two million lines of code caused a cascading collapse of the national telephone system in the summer of 1990.Now, 2 million rows is not too much.The combat computer aboard the Navy's Seawolf submarine contains 3.6 million lines of code. The new operating system "NT" released by Microsoft in 1993 contained a total of 4 million lines of code.A program of 100 million lines is not far away from us. When computer programs balloon to billions of lines of code, just maintaining the programs, keeping them up and running can become a major burden in itself.Too many economic activities and human lives depend on these billion-line programs to let them fail even for a moment.David Ackley believes that reliability and uptime will become the number one priority for software. "I'd venture to say that for really complex programs, it's going to cost more resources just to survive." Currently, only a small portion of a large program is dedicated to maintenance, bugfixing, and cleanup. "In the future," Akeley predicted, "ninety-nine percent of raw computer cycles will be spent keeping this monster self-monitoring to keep it going. Only the remaining one percent will be spent executing user Quests — phone exchanges or whatever. You know, the monster can only complete user missions if it survives."

As software gets bigger and bigger, survival becomes more important and more difficult at the same time.Surviving day-to-day use means having to be able to adapt and evolve, and that takes a lot more work.Only by constantly analyzing its own situation, modifying its own code to meet new needs, purifying itself, constantly eliminating abnormal situations, and keeping adapting and evolving, can the program survive.Computing must have life and vitality.Ackley calls this "software biology" or "vigorous computing."Even if a programmer keeps a pager on 24 hours a day, there is no guarantee that billions of lines of code will not fail.Artificial evolution may be the only way to keep software alive and alive.

Artificial evolution is the end of engineering supremacy.Evolution can enable us to plan beyond our own ability; evolution can carve what we can't; evolution can achieve more perfection; evolution can care for the world we cannot care for. But, as the title of this book suggests, the price of evolution is loss of control."One of the problems with evolutionary systems is that we give up some control," Ray said. The pilot program on Danny Hillis' flight was evolved, and no one could understand the software.It's a mess of thousands of threads, maybe only a small part of it is really needed, but it's guaranteed to run without a hitch.

Akeley's telephone system is managed by evolved software that is "alive".When it goes wrong, no one can troubleshoot because the program is buried in an incomprehensible way in an unknown network of small machines.When it goes wrong, though, it fixes itself. No one can grasp the final fate of Tom Ray's broth.They are good at designing all kinds of little tricks, but no one tells them what tricks they need next.Only evolution can handle the complexity we have created, but evolution is beyond our control. At Xerox Palo Alto Research Center, Ralph Merkler is making tiny molecules that can replicate themselves.Because these molecules are on the nanometer scale (smaller than bacteria), this technology is called nanotechnology.In the not-too-distant future, the engineering skills of nanotechnology and biotechnology will converge: they both see molecules as machines.For pure life, nanotechnology can be regarded as bioengineering; for artificial evolution, nanotechnology is equivalent to biomolecules.Merkel told me, "I don't want nanotechnology to evolve. I want it to be within certain frameworks and subject to international conventions. There is nothing more dangerous for nanotechnology than mating. Yes, I think there should be an international convention to limit the use of mating in nanotechnology. Once you mate, you have evolution; as long as you evolve, you're in trouble."

Evolution is not completely out of our control; relinquishing some control is simply a matter of putting it to better use.The things we pride ourselves on in engineering—precision, predictability, accuracy, and correctness—will all be watered down by evolution. And these things must be downplayed, because the real world is a world full of unpredictable events, a world of ever-changing changes; to survive in this world, a little fuzziness, relaxation, more adaptability and less precision are needed. manner.Life is uncontrollable; living systems are unpredictable; living creations are not either-or.When talking about complex procedures, Akeley said: "'Correctness' is a moon in water, and it is a characteristic of small systems. In the face of huge changes, 'correctness' will be replaced by 'survivability'."

There is no such thing as a right way when the telephone system is run by highly adaptable evolutionary software.Akeley continued: "To say that a system is 'correct' sounds like slang. A system is judged by its responsiveness to unexpected situations and the creativity of its responses." Rather than being right, it is better to be flexible , not as durable; the so-called "a good death is worse than a hard life".Ackley said: "Small and correct programs are like ants, ignorant of the world around them; and responsive programs are often runaway behemoths, spending only one percent of their energy on the problem you are trying to solve. On. Which one is better and which one is worse is self-evident."

Once, in Stuart Kaufman's class, a student asked him, "How does your evolution deal with what you don't want? I know you can make a system evolve what you want stuff, but how can you be sure it won't make something you don't want?" Good question, kid.We can define what we want precisely enough that we can breed it.However, we often don't know what we don't want.Even knowing it, the list of undesirables is unrealistically long.How can we weed out those unwanted side effects? "You can't," Kaufman replied bluntly. That's the evolutionary deal.We trade control for power.For those of us obsessed with control, this is the devil's deal.

Relinquish control, and we will artificially evolve a whole new world and undreamed-of opulence.Just let it go and it will blossom. Have we ever resisted the temptation of the devil?