Chapter 84 15.2 Evolution can do what you cannot do

Ray discovered "creatures" that human programmers could not program. "I started by writing something that was 80 bytes," Ray recalls, "because that was the best design I could come up with. I figured maybe evolution could get it down to about 75 bytes, so I let the program run. All night. Turns out the next morning something new appeared - not a parasite, but something completely self-replicating - and it was only 22 bytes! To my bafflement, in the absence of How can a computer virus replicate itself with only 22 instructions, while stealing other people's instructions like a parasite? To share this new discovery with others, I am posting its basic algorithm online. MIT A computer student at the college saw my explanation but somehow didn't get the code for virus 22. He tried to recreate it by hand, but his best grade also required 31 instructions. When he learned that I was in When he got 22 instructions while sleeping, he was devastated."

What humans cannot do, evolution can do.Ray showed the trace of 22 doubling in culture on a monitor to best illustrate his statement: "Think about how random program changes can outperform elaborate hand-programming. It sounds absurd, but here's a case in point." It suddenly became clear to the bystander that the creativity of these "brainless" hackers was endless. Because viruses consume computer cycles, smaller (shorter instruction set) viruses have certain advantages.Lei rewrote the code of "Earth" so that the system allocated computer resources to the virus in proportion to its size, and large viruses got more cycles.In this mode, Ray's viruses inhabit an unbiased world.Because the world treats big and small viruses alike, it might make more sense to run long-term.At one point Ray ran it for 15 billion computer cycles.Around the 11 billionth cycle or so, a virus with a length of 36 bytes was born, which was so clever that it was almost cunning.It calculates its real size, and then shifts the length value one bit to the left at the "tail" (let's call it that), which is equivalent to doubling in binary.By lying about its size, virus 36 stealthily steals resources from virus 72, which means it gets twice as much CPU time as it actually needs.This variant naturally swept the entire system.

Perhaps the most amazing thing about Tom Ray's electro-evolution machine is that it creates sex.No one told it what sex was, yet it discovered it.In one experiment, Ray let the “culture” run with no added errors to see what would happen if mutation was turned off.To his astonishment, the evolution occurred even without the programmed mutation. In real natural life, sex is a far more important source of variation than variation.Sex, conceptually, is genetic recombination—the combination of some genes from the father and some from the mother to create an entirely new genome for the offspring.In "Earth", parasites sometimes "borrow" the replication function of other viruses in asexual reproduction, and the "harvester" may happen to kill the host in the process, and the original space of the host is occupied by the new virus.When this happens, the parasite uses parts of the new virus's code, as well as parts of the "dead" virus's interrupted replication.The resulting offspring is a natural new combination that has not been deliberately mutated. (Ray also says that this eccentric breeding is "the equivalent of having sex with a dead man!") This interrupted mating actually happens all the time in Ray's "culture", but only when he turns off the mutation function, does he Take note of this.It turns out that inadvertent reorganization alone is enough to drive evolution.There is enough irregularity in the memory spaces that organisms inhabit at the time of death, and this complexity provides the variety needed for evolution.In a sense, the system has evolved variation.

For scientists, the most exciting thing about Ray's artificial evolution machine is that his small world appears to exhibit punctuated equilibrium.In a relatively long period of time, the population ratio has maintained a relatively stable situation, with only occasional species extinction or new species being born.Then, almost in the blink of an eye, this balance was immediately interrupted by an overwhelming alternation of new and old species.For a short period of time, change is rampant and unfettered.Then things worked out, and stillness and balance reigned supreme again.Fossil studies show that this form predominates in nature on Earth.Stillness is the norm; change is always sudden.The same pattern of punctuated equilibrium can be seen in other computer models of evolution, such as Christian Lindgren's Prisoner's Dilemma-style co-evolving world.If artificial evolution mirrors biological evolution, you must be wondering what would happen if Ray let his world go on forever?Will his viral monsters create multicellularity?

Sadly, Ray never marathoned his world to see what would happen months or years later.He's still tinkering with his program, improving it so that it can collect the massive amounts of data (50 megabytes per day) that the long run produces.He admits, "Sometimes, we're like a bunch of boys with a car. We're always in the garage with the hood open and the engine parts out, but we almost never drive because we're so obsessed with putting on horsepower. .” In fact, Ray is focusing on developing a new piece of hardware that should be a new technology.Ray thought he could "burn" a virtual computer and the basic language written for it into a computer chip—a silicon chip for evolution.This off-the-shelf Darwinian evolution chip becomes a module that can be plugged into any computer, and it will rapidly multiply things for you.You can evolve code, or subroutines, or perhaps even entire software programs. "I found it rather odd," Ray confided, "that, as a tropical plant ecologist, I should be doing computer design."

The promise of what a Darwinian evolution chip might bring is fantastic.Imagine having one in your personal computer, and the word processor you use on your computer is Microsoft Word.Word evolves as you work, thanks to Darwinian evolution resident operating system.It uses the processor's idle cycles to improve and learn in a slow evolutionary way, adapting itself to your work habits.Only those changes that improve speed and accuracy are preserved.Still, Ray is convinced that messy evolution should be kept separate from work. "You should separate the evolution from the end user," he said.He envisions "digital farming" taking place offline in the background, so that the errors and failures integral to evolution are invisible to the user; evolution is also "dormant" while the user is using it.

Evolution is no longer a fantasy in the market.Today you can buy spreadsheet modules with similar functionality.Its name is called "evolutionary". "Evolutionary" was a spreadsheet template on a Mac -- very complex, packed with hundreds of variables and "what-if" functions.It is used by engineers and database specialists alike. Say, for example, you have medical records for 30,000 patients.You may be interested to know the symptoms of a typical patient.The bigger the database, the harder it is to see where your data is stored.Most software can calculate an average, but this does not extract a "typical" patient.What you want to know is, among the thousands of categories of data collected, which set of measurements has similar meaning for the largest number of people.This is a problem of optimizing a large number of interactive variables.This is an all-too-familiar problem for any living being: how to maximize the output of thousands of variables?A raccoon must ensure its own survival, but there are a thousand variables (foot size, night vision, heart rate, skin color, etc.) that can change over time, and a change in one parameter will cause a change in another.The only way to move through this vast space of possible outcomes, with any hope of reaching the summit, is to evolve.

The "Evolver" software optimizes the broadest medical records for the largest number of patients.It tries to give a basic description of a typical patient, then checks how many patients fit this description, then makes multi-dimensional improvements to the medical record to see if more patients fit it, then revises, selects, revises, until The largest number of patients fit this description.This work is especially well suited to evolution. Computer scientists call this process "hill climbing."Evolutionary programs attempt to climb to the top in a library of forms containing optimal solutions.By continually pushing toward better solutions, the program keeps climbing until it can't climb any higher.At that point, they reach a peak—a maximum.However, there is always a question: is this summit the tallest around?Or is the program stuck at a local high point, separated by a canyon from a much higher nearby peak, with no way back?

Finding a way up to a high point is not difficult.Evolution in nature and evolutionary programs in computers are good at climbing to the highest point in the overall sense - the main peak, in a terrain with undulating mountains and higher mountains.