Chapter 102 17.4 Survive by breaking the rules

Koza pointed out that the reason why human beings pursue simple formulas like Newton's f=ma is because we firmly believe that the universe is built on the basis of simple order.More importantly, simplicity is convenient for humans. The formula f=ma is much easier to use than the monster that Koza determined the helix, which makes us appreciate the beauty contained in the formula even more.Before computers and calculators, simple equations were more practical because they were less error-prone.Complicated formulas are tiring and unreliable.However, within a certain range, neither nature nor parallel computers will worry about complicated logic.Those extra steps that we find ugly and dizzying, they work with tedious precision.

Although the brain works like a parallel machine, the human consciousness cannot think in parallel.This ironic fact has cognitive scientists baffled.Human intelligence has an almost mystical blind spot.We do not intuitively understand concepts in probability, lateral causality, and simultaneous logic.They don't fit our way of thinking at all.Our minds fall back on serial narratives—linear descriptions.That's why the earliest computers used the von Neumann serial design: because that's how humans think. And that's why parallel computers had to be evolved rather than designed: because we're all idiots when it comes to thinking in parallel.Computers and evolution think in parallel; consciousness thinks serially.In a highly controversial article in the Winter 1992 issue of Daedalus, James Bailey, marketing director of Thinking Machines, described the flyback effect of parallel computers on the human mind.In the article titled "First We Reinvent Computers, Then Computers Reinvent Us," Bailey pointed out that parallel computers are opening up new frontiers of knowledge.The new logic of the computer in turn forces us to ask new questions and perspectives.“Maybe, there are some radically different ways of computing out there, ways that can only be understood by thinking in parallel,” Bailey suggests. Thinking like evolution might open new doors in the universe.

According to John Koza, evolution's ability to deal with loosely defined parallel problems is another unique advantage.The difficulty with teaching a computer to solve problems is that, to this day, we end up reprogramming it verbatim to solve every new problem we encounter.How can a computer do things on its own without having to be told step by step what to do and how to do it? Koza's answer is: evolution.In the real world, a question may have one or more answers, and the scope, nature, or value domain of the answers may be completely ambiguous.Evolution has allowed computer software to solve such problems.For example: Bananas are hanging on the tree, please give the picking procedure.Most of the computer learning to date cannot solve such problems.Unless we explicitly provide the program with some explicit parameters as clues, such as: How many ladders are nearby?Is there a long pole?

And once the boundaries of the answer are defined, it is equivalent to answering half of the question.If we don't tell it what kind of rocks are nearby, we know we won't get the "throw a rock at it" answer.In evolution, this is entirely possible.More likely, evolution will come up with completely unexpected answers, such as: use stilts; learn to jump high; ask birds to help; wait out the storm;Evolution didn't necessarily require insects to fly or swim, only to be able to move quickly to escape predators or catch prey.Open questions yielded such varied but unambiguous answers as waterflies tiptoed on water or grasshoppers hopped.

Everyone who has dabbled in artificial evolution has been amazed at the ease with which evolution can produce fantastic results.Tom Ray said, "Evolution doesn't care if it makes sense; it cares whether it works or not." It is the nature of life to take pleasure in exploiting the loopholes of routine.It breaks all its own rules.Just look at these jaw-dropping wonders of biology: female fish fertilized by resident males, life forms that shrink as they grow, plants that never die.Life is a quirk shop, the shelves are never out of stock.There are almost as many oddities in nature as there are all life; every creature is in some sense reinterpreting the rules to find its own way.

Human inventions are not so rich.Most machines are built to perform a specific task.They obey our old-fashioned definitions and obey our rules.However, if we were to imagine an ideal, dream-like machine, it should be able to change itself to adapt to its environment, and ideally, it should be able to evolve itself. Adaptation is the distortion of one's own structure so that it can slip through a new hole.Whereas evolution is a deeper change that changes the very architecture of the building structure itself—that is, how change is produced—a process that often provides new vulnerabilities for others.If we predetermine the organizational structure of a machine, we also predetermine what kind of problems it can solve.The ideal machine should be a general-purpose problem-solving machine, a machine that can only do what you can't think of.This means that it must have an open structure."The size, form, and structural complexity [of the solution] should all be part of the answer, not part of the problem," Koza wrote. When we realize that it is the structure of a system itself that determines what answer, then what we ultimately want is how to make a machine that has no pre-defined structure.What we want is a machine that constantly renews itself.

Those working to advance AI research will no doubt sing its praises.Being able to come up with a solution without any prompts and without limiting the direction of the answer—what people call lateral thinking—is almost equivalent to human intelligence. The only machine we know of that can reshape its own internal connections is the gray living tissue we call the brain (gray matter).The only productionally reconfigurable machine we can currently envisage would be a software program capable of adapting itself.Sims and Koza's evolutionary equation is the first step toward self-programming.An equation that can reproduce other equations is the ground for this kind of life.The equation that reproduces other equations is the open universe.Any equation can be generated there, including self-replicating equations and ouroboros-style infinite loop formulas.This recursive program, which acts on itself and rewrites its own laws, contains the most magnificent power in the world: creating constant innovation.

"Continuously new" is a phrase used by John Holland.He has been working on artificial evolution methods for years.In his words, what he was really engaged in was a kind of constant new mathematics.That's the tool that can create endless new things. Carl Sims told me: “Evolution is a very practical tool. It’s a way of exploring new things you never thought of. A way to explore programs. If a computer is fast enough, it can do all these things.” Exploring beyond our comprehension and distilling what we gain is the gift of directed, supervised and optimized evolution.Tom Ray said: "However, evolution is not just optimization. We know that evolution can go beyond optimization and create new things to optimize." When a system can create new things to optimize, we have a constantly new tool and Open to evolution.

Sims' image selection and Kozana's program selection by logical reproduction are both examples of what biologists call breeding or artificial selection. The criteria for "eligibility"—the criteria for being selected—are determined by the breeder and are therefore artifacts or artificial.To achieve permanence—to find something we didn't expect—we must let the system define the criteria for its selection.This is what Darwin meant by "natural selection."The selection criteria are determined by the nature of the system; it emerges naturally.Open artificial evolution also requires natural selection, artificial natural selection if you like.The selected features should arise naturally from inside the artificial world.

Tom Ray has loaded up the tools of artificial natural selection by letting his world choose the fittest for itself.Thus, his world theoretically has the ability to evolve entirely new things.But Ray did "flip a little" to get the system to work.He couldn't wait for his world to evolve self-replicating abilities on its own.So from the beginning he introduced a self-replicating mechanism, once introduced, the replication never ends.Using Ray's metaphor, he kick-started life in the state of a single-celled organism, then watched a "Cambrian explosion" of new organisms.But he is not sorry. "I'm just trying to get evolution, and don't really care how I get it. If I need to push the physical and chemical components of my world up to a level that can support a variety of unlimited evolution, I'm happy to do so. I have to I don't feel guilty about manipulating them to get to this level. If I could manipulate a world to the tipping point of the Cambrian Explosion and then let it boil itself over the edge, that would be something I'd never forget. And what the system produces The fact that I had to manipulate it to a breaking point was nothing compared to the results."

Ray believes that starting open artificial evolution is already extremely challenging, and he doesn't necessarily have to evolve the system itself to that extent.He will control his system until it can evolve on its own.As Carl Sims said, evolution is a tool.It can be combined with control.Ray switched to artificial natural selection after several months of control.The opposite process is also possible—perhaps someone will evolve it for a few months and then control it to get the desired result.