atomic_fungus (atomic_fungus) wrote,
atomic_fungus
atomic_fungus

#4636: There's no algorithm for machine evolution.

Og, who is a robotics engineer, talks today about the whole "rise of the machines" concept, and how it can't happen with current technology.

The spontaneous evolution of machine intelligence lacks a mechanism by which it can occur. Simply connecting a huge number of machines together (the critical mass conjecture) isn't going to do it, any more than you can get a functioning computer by tossing a bunch of transistors and passive components into a box and applying a current.

Present state of the art for evolutionary algorithms is pretty primitive, particularly since they are built around solving specific problems. There is no generic evolutionary algorithm into which you can input the parameters of whatever problem you happen to have, and this would be a crude first step towards an auto-evolving machine. In fact, there's a thermodynamic problem with a self-programming computer, and most of the time even self-modifying code is supremely limited in what it can do to itself. Certainly there is no way to design a computer which can program itself--or even other computers!--which would be a vital component of machine intelligence.

Neural networks are also too primitive. A neural network simulates the activity of neurons in a brain, but at present we can't even simulate an insect brain let alone self-awareness or sentience--and even if we could simulate the number of neurons in an insect brain, there's still the underlying behavior of an insect; we have no idea how a biological brain is pre-programmed with certain behaviors, and programming a computer of any kind to--for example--build a web like a spider does would be a herculean task.

Think about that example. A tiny spider, which has a brain the size of a pinpoint, is capable of building a web that spans a certain space, using whatever anchor points are convenient. It does this under completely uncontrolled conditions, and it does it so well that the spiderweb has a characteristic shape which is consistent across species.

We can't program a robot to build a spiderweb in a random bush, carefully selecting anchor points and spinning silk and-and-and. We can program a robot to build a web on a carefully placed framework with known anchor locations, and it'll build a million webs with micrometric precision, but if you drop it in the woods and command it to start making a web, the arm will smash branches and wreck things, and--most likely--not even end up building a proper web.

Continuing with this example: when we have a robot which can analyze a random bush and decide where to place anchor lines, and then reliably spin a web that looks like any other, then we have made the first step towards machine intelligence. The first step, because the robot no longer has to be told how to make the web in each specific case.

...and this hypothetical web-building robot will be unable to do anything else, because being able to build a proper web in a random bush is trivial compared to being able to decide when and where to build a web. The web-bot will make webs all day long until your entire forest is festooned with them, but it won't make them under its own initiative.

You can give your robot a convincing simulation of initiative, but what you'll actually be doing is writing a random number generator.

Our machines are complex, but they're inert matter. They cannot do anything by themselves. Even programs which execute automatically are simply watching a timer or a count and then executing when certain values line up; it's not sitting there and then thinking, "You know, I haven't run a virus scan today; I'd better get after that." It's merely waiting for the right pattern of bits to show up and then running a program based on preset conditions. Absent those preset conditions the machine will never, never, ever run that program.

If we want killer robots, we're going to have to go to a great deal of effort to build them. I'm not going to hold my breath waiting for them to auto-evolve.

* * *

The Anchoress comments on the whole "SF arch diocese hoses down the homeless" thing, and says--in kinder, gentler fashion--approximately the same thing I said about it.

* * *

I would really like to be able to write "about it" without first writing "abou tit". I don't know what abou tit is, and don't want to know.

My spacebar timing is too far advanced.

* * *

High school is an incandescently stupid waste of time.

I was acquainted with someone in high school--I never spoke to him, but knew his name and career aspirations--who was considered one of the top students of my class. He got excellent grades, he was an athlete, and he was popular. He was intent on attending the Air Force academy, and showed every sign of being headed for great things.

Reportedly, he flunked out of academy in his first semester.

The problem is, public high school in America is easy. The toughest courses you can take are watered down, predigested pap. The class I shared with this guy was college prep physics (it would be "AP physics" these days) and it was taught at such a glacial pace that I learned more from reading the book than I did from the class itself. This was thirty years ago and I know that the situation has not improved since then.

...but when you present a real intellectual challenge to a guy like that, a guy who was in contention for valedictorian in high school, who never had any trouble scoring highly on tests or pulling down a 4.0 GPA, suddenly he is faced with something he never had to face before: difficulty.

When I learned the guy had flunked out of academy, I felt sorry for him. But the fact that he did so was not remarkable, because a certain number of guys flunk out every year, usually for the same reason. It's not that they're stupid; it's that they've never faced a real challenge, and when they get to a place where they do, some of them can't rise to it. Sometimes it's not the intellectual challenge, but the social one--hazing, and so on--as these guys are typically not on the bottom of the social ladder in high school, and quite unexpectedly find themselves way down the food chain.

Certainly I do not (and did not) think I would have--could have--fared any better than he.

It used to be that high school was necessary. That was a long time ago, though, back when they still taught useful things in high school. It used to be that a person with a high school diploma would be able to read, write, figure, and know history, to one extent or another. These days we allow people to graduate who cannot read much of anything, can barely write, who are utterly ignorant of history, and cannot comprehend any mathematics more complex than "2+2=4". (Example.)

But they sure know how to use condoms, apply for public aid, and vote Democrat!

* * *

Democrat Senators want Obama to delay the implementation of the law they voted for. Obamacare is going to cause a lot of problems, over and above the problems it has already caused, and the Senate Democrats want Obama to (unconstitutionally) delay the implementation of parts of the law so that small business owners won't get shafted, long and hard and dry, by the law.

Problem: they are going to get shafted, long and hard and dry, just like the rest of us. It's a question of "when", not "if".

Obamacare has turned out to be a royal mess. I'm not surprised; I predicted it would be--a lot of us on the right did--and it didn't take any special training or prescience to make that prediction, because all we had to do was to look at how similar programs have fared, both here and abroad.

Which is to say, "not at all well".

Obamacare is foundering for all the predicted reasons, and the Democrat response is to pretend that the law doesn't say what it actually says and--where necessary--get the Supreme Court to rubber-stamp their prevarications.

But the entirely predictable negative consequences of Obamacare continue to arise, because once the thing was signed into law those consequences were as inevitable as the tides.

* * *

Obama was crowing about what a success Yemen is, but Yemen is not a success. Out here in flyover country, Obama's Yemen policy has all the appearance of being something we call "an abject failure". That's because we lack the sophistication and nuance of the D.C. elites, you see, and we insist on calling things by their proper names instead of using terms our betters deem more appropriate. We're hicks! We don't know any better.

Anyway, "abject failure" is a term we also apply to our sitting President, because, well, we're all racists. You know how it is.

* * *

I got to bed after 4 AM, because I was up watching YouTube videos by a couple of guys who took a wrecked Harley and rebuilt it. Go to YouTube and search for "auction bike build" and you'll find it. It was pretty interesting stuff, at least if you're of the "take it apart and put it back together again for fun" persuasion.

But since Mrs. Fungus had to get up early, I only got a couple hours of sleep; and since there is nothing I absolutely must do today, guess what I'm going to do now?
Subscribe

  • Post a new comment

    Error

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.
  • 2 comments