Re: Best Programming language for AI in Nanobots


In article <11tocto9vo7avfa@xxxxxxxxxxxxxxxxxx>, rhooker123@xxxxxxxxxxx
says...

> " Consider that Adrian Thompson's first piece of evolved
> hardware
> was a signal discriminator that had to distinguish between two signals
> of different frequencies. The resultant design was an order of
> magnitude
> smaller than anything a human designer would have produced - nobody
> ever
> did figure out what the little cluster of isolated gates was doing but
> it wouldn't work without them. "
>
> Well that is the problem, is a company going to be able to market a HIV
> nanobot which no one is entirely sure how it works.

That's a subtler question than most people realize, I think.
I keep meaning to read up on patent law in that area, as well-- can you
patent something that a genetic algorithm "designed" for you?

> Given the risk of mass GOO from a nanobot failure I strongly doubt
> underwritters will have any interest in releasing large amounts of self
> replicating nonbots in to the environment which have not been fully
> designed out.

What about non replicating bots?

> With chemcials or organisms its different because these have been
> around for billions of years, but nanobots will be a bit more unknown.
>
> I am also not so sure that evolution is that optimal, if you look at
> the current exponential growth in memory and computing power of
> designed computers you see that humans, with far less waste, have been
> able to repeat in my 50 years what it took evolution to produce in
> maybe a billion years, maybe more. The Singularity is probably only a
> matter of decades away, so in 100 of computing with digital computers
> design will have accomplished 6 billion years of evolution.

Well, first, there is no optimal optimization technique over the class
of all optimization problems. The "No free lunch" theorem proves that
rather conclusively. Over the class of all problems, all optimization
techniques perform equivalently. Which is depressing. However, over
subsets, some techniques may dominate.

Second, vast vast amounts of real engineering design is done these days
by optimzation techniques of one sort or another. I've ragged
publically on the idea of simulation over all, but there are some cases
where the techniques work fairly well. It's possible for relatively
inexperienced engineers to design filters today that would have taken
narrow filter specialists to design thirty years ago, just on the
strength of optimizing simulators. Incredible feats of routing
optimization (for digital circuit boards, anyway) are performed not by
engineers, but by CAD technicians who set up the constraints and then
let the workstation chug over night. There are many *many* more
examples.

In this sense, I see absolutely no difference between a filter (say)
optimized by evolutionary techniques, and one optimized by gradient
minimax with a few random stips to get out of local minima, from time to
time.

Third, I'm not sure you're making a proper analogy between evolutionary
design and cognitive design-- there are multiple and deep flaws in the
analogy you've presented. For instance, "true" evolution doesn't just
optimize for memory speed, it optimizes for survival; evolutionary
optimization techniques, however, can optimize for just about anything
you like, as long as you can set up the fitness functions appropriately
and code a useful genetic representation.

> I am a former programmer who now works on contracts for massive systems
> and frankly I don't see any big agency, required for such a project,
> accepting a program that just works, they want to see specifications,
> and frankly I would oppose the release of any nanobot in to the
> environment which had properties not understood fully.

I am a hardware designer who designs massive systems. It would not
surprise me to learn that parts of your or my most recent project
contained small subsections which were designed with evolutionary
techniques.

As long as the subsystems can be specified narrowly enough and tested
rigourously enough, they can be used in place of rationally designed
subsystems or components.

I'm not a fan of the idea that you can design something like a
television or a car by nothing but one big genetic algorithm. I know of
no technique, yet, that will allow that. But just as you might design
an antenna, a display subcircuit to be replicated thousands of times, a
power supply, the same may prove true for nanobots.
>
>

--
John S. Novak, III
The Humblest Man On The Net

.

Relevant Pages

  • Re:Urgent requirement for Java developers.
    ... Identifies key business and technology drivers that impact application ... and application design, coding and design standards, best practices, ... application of techniques and standards such as service-oriented ... Expert level on Java J2EEdevelopment knowledge (5+ years on J2EE ...
    (comp.lang.java.beans)
  • Sr. Java Architect Needed
    ... Performs design, development and maintenance of highly complex ... Researches problems, discovered by QA or product support, and develops ... Determines what tools and techniques we will utilize to meet the ... -Expert analytical and design skills at ...
    (comp.lang.java)
  • Re: Graphic designers creating with Word
    ... A Word file does not contain the concept of a "page". ... And stop listening to people who say Word can't do page design. ... it's not helpful :-) Neither can inDesign. ... programs, so if you use the techniques from inDesign or Quark, everything ...
    (microsoft.public.mac.office.word)
  • Re: OOA?
    ... I don't think they are design choices. ... "One of the most crucial steps in software engineering is to model the ... The most frequently used diagrammatic techniques are ... diagrams, Entity-Relationship Diagrams, Object Diagrams, and ...
    (comp.object)
  • Re: OOA, OOD, design all first, how to go about it all?
    ... direction, and then the iteration. ... Other techniques are probably more ... > always been to do some level of analysis and a modicum of design at ... > development to test that you understand the rules of the game. ...
    (comp.object)
Quantcast