From: Aaron McBride (amcbride@jps.net)
Date: Sat May 12 2001 - 14:13:50 MDT
(Is there a convention in this mailing list about posting replies on top,
or on bottom?)
I'm not a hardware guy, so I'm sure I'll be corrected when I get something
wrong. :)
You're assuming that I want to read the e out in digital form (hence the
need for the very accurate volt-meter). I wouldn't need to ever convert it
to digital for though to get 2e for example though. Let's say you wanted
to get a certain voltage at a certain place in your system. In digital,
you'd have to turn it all into an estimate (read a voltage), do your math,
then convert back into a voltage. But if we were doing it with analog (for
a neural network let's say), then we would start with a voltage, run it
through a system, and have another voltage come out the other side. As far
as the noise goes, from what I understand, the "noise" (the fuzziness) in
the qbits in the quantum computers is what we are measuring/operating on.
So, in a sense that is another thing that analog can do that digital can't:
use noise as part of the computation.
(Ok, noise can be simulated in a digital system, but that takes even more
processing power.)
Once again, I'm not saying that the whole AI needs to be analog, just that
it should use at least some analog (where appropriate). We could get into
a whole discussion on free will possibly arising out of the noise, but then
we need to ask if the AI will need free will, and what free will is, and a
thousand other questions.
-Aaron
At 12:36 PM 5/12/2001 -0700, you wrote:
> > 1) What can analog do that digital can't.
> > Analog can store infinitely complex values in a single unit. The value e
> > could be stored in full precision (from what I've read) in a quantum
> > computer. Digital systems are always limited by their resources (# of
> > flops, how many bits of storage does it have, etc...). Sure, a digital
> > computer can get very close to e, but it can never touch the
> > value
>
>i think you are missing the major reason why digital is being used as
>opposed to analogue. you are right, with a digital system our storage (and
>computation) is limited by the number of bits we have to store. so arbitrary
>precision requires arbitrary number of "components" (bits of ram for
>example). also true, that a single analogue component (say a capacitor, i
>know it's not a good analogue component, but it can store a signal in the
>electric field) can store an arbitrary precision number(it would have to be
>scaled by some value, so we would need two per number). the problem is
>twofold. first when you read you 'e' back from the storage, you must have a
>very accurate voltmeter. so the precision of the computer now depends
>completely on your ability to accurately measure voltage to a very large
>number of significant digits. so if you cared about 100 digits of e, you
>would need a voltmeter with 100 digit precision. which is a lot harder to
>find than 100 flip-flops to represent e in a digital computer. the second
>major factor is (always) noise. in a digital computer, an analogue signal
>can take on only two values, so it can be distinguished from noise very
>easily. if you use analogue components, somewhere along the line, the
>precision of the signal you are storing/computing will be drowned out by the
>noise of the system. so again, you cannot store a number more precise than
>the noise.
>
>both of these problems have a very easy solution. use more "components". if
>you need to store 100 digits of e, but your voltmeter is only good at
>reading 4, store the number in 25 "components". and if your signal to noise
>ratio is to high for you to read those digits, use 50, or 100. as you can
>see, its very easy to transform an analogue system into a digital one. in
>fact, because you are so limited by noise and instrument precision, you will
>have to do it anyway.
>
>so in reality, digital and analogue are equivalent, one requires more bits,
>the other, more precise instruments. bits are cheaper, easier to
>manufacture, and scale.
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:36 MDT