X-Message-Number: 8532
From:  (Thomas Donaldson)
Subject: Re: CryoNet #8521 - #8526
Date: Wed, 3 Sep 1997 22:18:43 -0700 (PDT)

Hi everyone!

I should start with a quite material matter.

In my next to last message I said that I would come back with at least one
place where you could go to get disability insurance ie. insurance that would
replace a proportion of your income if you became disabled. 

Naturally I first called up my own insuror. It turned out that they were not
writing that business any more. For various reasons, they'd gotten out of it.
They were, of course, still paying for those policies written BEFORE they got
out the business (mine being one).

This put the matter up in the air. I am still looking. Apparently a number of
companies still write income replacement policies, no doubt with differing
conditions and prices. I have found 4 different companies, but so far not
got names and addresses of agents for them (except for one).

		Principal Mutual Life Insurance (call 1-800-654-1485)
			Des Moines, IA 50392-0001
		UNUM (formerly Union Mutual)
		Provident Life and Accident
		Paul Revere Insurance

The kind of policies available, as I said, may vary a lot. I am not done 
with looking into this issue and will report more later. So take this just
as an interim report.

And now ...

As for the Turing Test, I would certainly agree that it was an ATTEMPT to 
provide a litmus to tell if we had devised a mind equal to a human mind in
a computer. Those of us who don't like it as a test would agree that we still
need such a test. However I notice in both John Clark and Peter Merel an
interesting looseness about just what the Turing Test requires the computer
(or person) to do. (And Peter Merel decides to attach various metaphorical
meanings to "uploading" too).

Look at John Clark's statement: "Einstein was not intelligent, he just talked,
wrote, and behaved like he was. Do you see anything crazy about the preceding
sentence?" I would agree that the sentence is odd, but it does not obviously
bear on the original Turing Test for intelligence --- whatever intelligence
may be. It's the "..and behaved like he was." that gives it away.

If we are to understand these issues it would really help for us to either
define what we mean or accept some third fixed definition. Everything I've
read about the Turing Test says that it is set up in one special way: 
an interrogator asks questions (spoken or even just written) of an object
in a different room, and receives replies. If the interrogator cannot tell
the difference between the replies of the object and the replies of a 
human being, then the object has human intelligence. If there are much more
general tests for human intelligence that are also called "the" Turing Test,
I'd like to know their source. 

As such, this limits things a LOT. I believe it limits them too much.

Again, more precision is needed when we talk about a computer, even the
simple kind Turing set up. As an intellectual exercise, if John or others
don't like the idea of an "infinite" tape that the machine can mark up
however its program says, then we can imagine a daemon which provides it
with more tape whenever it runs out.

Moreover, the abstract study of computing has used a notion of an analog
machine, too. I suspect this is what John Pietrzak was referring to, but
here goes. Suppose that our machine, instead of simply being forced to put
down a 0 or 1, could put down any floating point number of arbitrary length.
(Remember that we are considering an abstract machine). Such a machine 
could be taken as a model for an analog computer: instead of parts that
eventually become indivisible, we have something which can be divided into
arbitrarily small pieces. Such a device can do much more than the original
Turing machine. We might, of course, introduce other abstract versions of
such machines, too. (One reason to do so is to work out in detail, not just
by waving hands, just what a Turing machine can and cannot do).

Yes, all of these machines, including the original Turing machine, will
never be realized in the real world. There are no daemons willing to provide
arbitrarily large amounts of tape to our Turing machine. Anyone who criticises
these abstract models on the ground that they are not possible misses their
point completely. At the same time, I believe it will definitely help to 
consider much less abstract models, since we must deal with the world 
and POSSIBLE computers, too. That's how the issue of time taken to do a 
computation comes in, and the issue of finite memory sizes, and so on and
on. The abstract models are far from useless, but they're just a start.

And finally, one last point re. Anders Sandberg's comment. Perhaps (even
probably) he was just speaking loosely. But our memories are necessarily
chemicals of some kind and conformation. When he says that permanent 
memories are coded in the structure of our brain (ie. its connectivity)
he is talking about chemicals. In case anyone hasn't noticed, EVERYTHING
is a mixture of chemicals of one kind or another. The real issue is that of
just how long these conformations of chemicals will last. 

I have stated in PERIASTRON that we aren't done when we work out that 
memories are coded in the connectivity of our neurons. We need to also know
how that connectivity is maintained in an ongoing way. Anders is quite correct,
though, if we forget this central point. The immediate chemicals which cause
short term memory, and even the chemical changes in synapses causing LTP,
are unlikely to code long term memory. They are too labile. What phenomena
such as LTP (Long Term Potentiation, for those who haven't been following 
this issue in detail: a special reaction of synapses by which on a sufficiently
strong stimulus they will continue to respond more strongly than before for
some time) ... what phenomena such as LTP do is to keep the memory while
other reactions build a firmer, more long lasting substrate for it.

And for those who are interested in computing, I would suggest that an
understanding of how brains work actually provides some interesting comments
on current computing. Ignore all issues of which is better at what. The 
real point is that as we work out how our brains work, we are finding a 
very different way of organizing a "computing machine"... even down to the
detailed operation of memories. If it does no more than tweak the imagination
of designers it's provided something quite valuable --- and I'd suggest that
the whole idea of neural nets (and evolutionary computing, too, as I understand
it, came from thinking about how brains might work) is just a start to the 
ideas such an understanding will evoke.

Of course, we're not going to upload, or restore anyone now suspended to life,
without understanding just how brains work, too. And that's a brutal 
necessity.

			Long long life,

				Thomas Donaldson

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=8532