From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!hsdndev!morrow.stanford.edu!nova1.stanford.edu!dow Mon Dec  9 10:47:19 EST 1991
Article 1789 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!hsdndev!morrow.stanford.edu!nova1.stanford.edu!dow
>From: dow@nova1.stanford.edu (Keith Dow)
Newsgroups: comp.ai.philosophy
Subject: Re: Physical limits when programming neurons and minds
Message-ID: <1991Dec2.005246.2168@morrow.stanford.edu>
Date: 2 Dec 91 00:52:46 GMT
References: <43772@mimsy.umd.edu> <288@tdatirv.UUCP> <57751@netnews.upenn.edu> <1991Nov29.164139.1588@morrow.stanford.edu> <57850@netnews.upenn.edu>
Sender: news@morrow.stanford.edu (News Service)
Organization: Stanford University
Lines: 198

>>From a physicists perspective, the brain is just a solution of
>>Schroedinger's equation.  Whether you know the boundary conditions and
>>have the computing power necesary to solve the equation is a problem 
>>for engineers.  
>
>This is ignorant.  Not all PDEs can be solved numerically.  As I said:





Schroedingers equation can be solved numerically.  It is the basic
equation of non-relativistic quantum mechanics.  All other solutions,
if they are CORRECT, are just subsets of the solutions to Schroedingers
equation. 



>>>The work of Pour-El and Richards has shown that certain physically
>>>plausible PDEs do not have computable solutions, so until we have an
>>>incrediblely greater amount of knowledge of the physics of neurons,
>>>the belief that every property of neurons is programmable is at best
>>>tentative.
>
>>This physics of every phenomena larger than a proton is well understood.
>
>This is also ignorant.  Incredibly so.  Perhaps you can point to
>references that explain high temperature superconductivity?

I can point to hundreds of papers that claim to have solved the question
of high temperature superconductivity. :-)  The problem now is having 
enough data to weed out the bad ones.  

No physicist doubts that superconductivity is a solution to the standard
equations used in quantum mechanics.  The problem is selecting among the
many different solutions.



>>Therefore every propety of a neuron is programmable.  People may or may 
>>not have the talent necesary to do it.  That is a completely seperate
>>issue.
>
>I repeat: there may be metamathematical obstructions to solving the
>relevant physics, even if it's just Schroedinger's equation.  No amount
>of talent can solve the unsolvable--except by changing the rules.

I don't know what metamathematics is.  Also your sentence which I quote
below, is logicaly inconsistent.  

"No amount of talent can solve the unsolvable--except by changing the
rules."

I still don't see the need for anything beyond Schroedinger's equation
to solve the problem.


>>>But for the sake of argument, let's assume that neurons can be digitally
>>>simulated.  Even with that much conceded, you still have a leap of faith:
>>>
>>>>If every property of a neuron is programmable, then so is intelligence.
>
>>>This does not follow in the least.  If our brains exploit quantum
>>>cryptography to get an internal sense of privacy, then short of a
>>>revolution in physics, there is no way you can simulate our minds
>>>on a digital computer.  There may indeed then be a different way
>>>to get digital intelligence, but it will not be found via the above
>>>implication.
>
>>The above is not true at all.  Schroedinger's equation is a second order
>>partial differential equation.  The solutions are completely
>>deterministic.  You can stick it on a computer and solve it.
>
>I repeat: this is 100% totally ignorant.  It has no connection with
>known reality.  You can stick Schroedinger's equation--a nice catch
>phrase for an infinite family--into a computer and get numbers.  You
>do not know, without further mathematical analysis, if your numbers
>have anything to do with the original problem.  This analysis cannot
>be automated.

Of course the analysis can be automated.  When you solve this type
of problem you have to put the answer in a form that is useful.  In
this case it will be the firing pattern of the neurons.  People
have already looked the firing pattern of neurons and learned a lot.
I don't see why the computer would have any problem doing it.

Also, a lot of work has been done on molecular modeling.  I sure this
field is doing well, since drug companies like to do so much of it.


>>Just because we may not be able to solve it for a particular case at the
>>present time does not mean it can't be solved in the future.
>
>No one made such a claim.
>
>>>In other words, Gell-Mann and Hartle are suggesting that a particular
>>>quantum mechanical configuration is an essential part of our minds.
>>>And so Bell's inequality may prevent any digital computation from
>>>ever being a simulation of our minds.
>
>>Ton's of quantum mechanics problems have been solved on computers.  The
>>results have been verified many times with experiments.  So what is the
>>problem?  
>
>Tons is not all.  That is the problem.  There are known lions and tigers
>and bears in the PDE forest.  And because of this, anyone who says that
>all PDEs are numerical pussycats is a fool.

We don't have to solve all PDE's.  All we have to do is solve
Schroedingers equation.  People have been doing this for over a hundred
years.  Drug companies make millions doing it accurately.  



>>Long range correlations can be handled on a computer.  So Bell's 
>>inequality is no big deal.
>
>Indeed, I must retract the Bell's inequality assertion.  To spell it
>out, Bell's inequality only forces us to reject local hidden variables.
>Programs are free to adopt nonlocal hidden variables or whatever.
>
>But I retract no more.  Our consciousness gives us a very strong sense
>of "self".  How so?  Is it an illusion?  Or is it something more fun-
>damental?  I don't know.  But it's conceivable that it may be more than
>an illusion, and it may depend on some quantum cryptographic privacy
>notions.  I cited GM&H as evidence that QM considerations of the mind
>are more than figments of Penrose's imagination.  Without evidence one
>way or the other, the bald assertion that knowing neurons means that we
>eventually know how to program the mind is a leap of faith.  For now,
>the only way to prove that we can program minds is to write the program.

There is plenty of evidence that neurons are what should be analyzed 
when one is trying to understand the human brain.  What else do you
propose should be studied?  Quantum mechical considerations only mean
the analysis of neurons is more complicated than people thought.  Again
though, we have the tools, it is only a question of how to apply them.





>> Physicists agree on at least two things.
>>They don't understand quantum mechanics. They don't need to
>>understand quantum mechanics because all they have to do is calculate
>>what happens next.  And they know how to calculate what happens next
>>using quantum mechanics.
>
>Here it is again--total ignorance.  Physicists sometimes know how to
>calculate what happens next.  This is what gets published.  They normally
>do not include a theoretical analysis that shows their finite difference
>scheme or whatever approximation is actually converging to the true
>solution.

We agree that garbage in gets garbage out.  But as I stated earlier, the
tools already exist, it is only a question of how to apply them.  If you
model of reality is wrong, and you plug it into Schroedinger's equation,
you will get garbage out.  

So the tools exist to solve your problem.  Whether people are capable of
using the tools correctly is another question.  

Also most of the physics papers I have seen go into gory detail when they 
talk about calculations.   


>You know what this means?  They sometimes publish garbage solutions.

I agree, they often publish garbage solutions.  Also people often can't
balance checkbooks.  This doesn't mean someone else can't balance their 
checkbook.  And it doesn't mean they should give up arithmetic.

>cheers

>Whenever I taught numerical analysis, I always tried to emphasize that it's
>not plug-in-and-crank-out.  Why?  Because of this sickeningly dangerous
>it's-a-bright-beautiful-world-of-computation-that-always-works attitude.
>The idea of people building bridges and airplanes without ever making a
>reality check is frightening.  It is certainly not cause for "cheers".
>
>Get a copy of Forman Acton NUMERICAL METHODS THAT (usually) WORK.  The
>numerical solutions of PDEs is simply not a crank, and because of the
>work of Pour-El and Richards, it is known that it never will be.  See
>their book COMPUTABILITY IN ANALYSIS AND PHYSICS.

If you think that details can burn you in theoretical work, you will
just love experimental work!   This explains why it takes
experimentalists about a year longer to get their degrees in physics.

The problems you talk about are certainly hard.  They don't involve any
new physics though.  They just involve finding new techniques to solve
the problems.

Also millions have been spent on accurately modeling molecules.  I don't
see any problem with progress continuing in this area.  Since computers
increase in power about a factor of two every years modeling will
improve a lot in the next ten years.

cheers


