X-Message-Number: 14935
From: "Pat Clancy" <>
Date: Thu, 16 Nov 2000 12:40:14 -0800
Subject: Re: Simulating People and Animals

Henri Kluytmans wrote:

> 
> >We will never have HAL running on a computer, not even by 20,001.
> 
> So you claim it will take more than 18,000 years to develop human 
> (or more) equivalent AIs...   Hmm,...  (sigh)

My claim is only that it will never happen based on a digital computer, i.e. 
Turing machine. And in this case it's a matter of possible vs. impossible, not 
development time.

> 

> >I hope that some other form of substrate can be developed that can support an
> >artificial mind. But this will depend on our achieving some degree of

> >understanding of how the mind works. Currently we have _no idea_ how the mind
> >works. 
> 
> We do not have to know how the mind works to be able to develop 
> an artificial substrate. We only have to know how it's building 
> blocks function (i.e. the neurons). See my other postings.
> (Of course, this is a simplified statement...)

Yes I agree that starting with something like neurons as building blocks 
would seem like the most promising approach. However, identifying building 
blocks before you know how the thing works that you're building is somewhat 
bass-ackwards IMHO. So what if you have artificial neurons - how do you 
hook them up?

> 
> >To go back to computers, but only as an analogy: it's as if you 
> >were given a computer, and you could observe it and run programs, 
> <snip>
> >certainly couldn't infer anything about even the lowest level 
> >virtual machine with your voltmeter. 
> 
> If you could use a microscope and a voltmeter, and if you 
> would have multiple computers that you could take apart and 
> analyse piece by piece, then somebody could certainly infer 
> everything, especially at the lowest levels. (Assuming that 
> that somebody would have a certain level of understanding of 
> physics.) You could start for example by analysing the 
> circuits on the CPU, and find out that the transistors 
> form logic circuits ... etc.

If you look at, say, the code for an operating system, you would have to 
conclude that the possibility of deriving this code, by reverse-engineering 
using only a voltmeter and a microscope, is extremely close to zero. 
Knowledge of physics wouldn't help, it's more a problem in logic and 
combinatorics. Remember you're starting with zero knowledge, no 
programming language you'd have to invent it, no knowledge of the software 
architecture or that there even was one (and the architecture of the OS alone 
can be incredibly complex). I've worked on operating systems during my 
career and for any real-world case it's just barely possible to understand how 

they work even if you have all the code there in front of you written in a nice
high-level programming language.

So with the brain/mind, it seems to me that we have a similar problem - we 
don't even know what language to use to describe how it works, much less 
have a "software" architecture. Even if the exact connections between some 
group of neurons can be described, we have no idea at all how that can give 
rise to a thought, because we have no scientific language to describe exactly 
what a "thought" is (of course there's plenty of subjective language to 
describe it), and therefore we have no way to describe in engineering terms 
how the thought arises from neurons. With a computer, we have multiple 
symbolic languages that are used at each virtual machine level to describe 

how, say, the click of a user interface button at the top level gives rise to a
sequence of function calls down the levels until you reach the individual 
gates ("neurons"). With the brain we only know how some of the gates are 
connected together; everything "above" that is still unknown to us.

Pat Clancy

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=14935