Re: Problem for physicalist evolutionists

On 31 Mar, 07:39, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article
<ccbb122d-d3bd-4025-88d9-36bf4a27a...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

On 30 Mar, 21:47, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article
<86cb5ac5-ccb1-4d35-aaf6-98df76fd6...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

On 28 Mar, 19:26, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article
<50737035-42b4-4820-a7d1-01aa7b013...@xxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

On 28 Mar, 18:37, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article
<78c8db0a-a508-4e6c-a8fe-bf33be7cf...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

On 28 Mar, 15:02, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article
<e3d8af50-5edb-4edb-a7a3-68380de23...@xxxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

On 28 Mar, 09:02, Tony Raymonds <to...@xxxxxxxxxxxxx> wrote:
In article

<ea346f72-61e5-478a-b08a-83da5803f...@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
someone2 <glenn.spig...@xxxxxxxxxxxxxx> writes

Regarding an ANN system where the nodes communicate by

part of it, 100 nodes, which were fed information (from
camera)
about a 10 x 10 picture (a pixel each), where the 100 nodes *do*
communicate the information they contained to the rest of

I think you were saying that in this scenario, the system

first person perspective containing the picture.

100 nodes does not a neural network make.

In your above example, are the 100 nodes feeding in to a
larger network
made of billions of neurones or are the 100 nodes all there is?

If 100 nodes is the entire network then no, it could not
first
person perspective.  You cannot build a neural network out
100 nodes
all of which are connected to the sensory input.  At best
you have a
"retina" capable of some pre-processing of the input

passing information on.

If it's connected to a neural network then yes, that network
may have a
first person perspective, or it may not - depending on its
construction.

Well let us assume a construction that you would consider to have a
first person perspective.

So your 10X10 grid communicates to a neural network made of
billions of
neurons which is constructed in such a way that it is
capable of a FPP.

Fair enough.

Regarding an ANN system where the nodes communicate by laser had a
part of it, 100 nodes, which were fed information from 2 cameras,
about two 10 x 10 pictures (a pixel from each to each
node), where the
100 nodes *do* communicate the information about the second picture
they contained to the rest of the system. I am asking that in this
scenario, will the system only have a first person
perspective of the
communicated second picture, or will it have a first person
perspective of both the pictures.

You do not state whether or not the 10X10 grid passes on any
information
about the first picture to the rest of the network so
technically the
answer is that the question cannot be answered.  Not enough
information.

Assuming you meant that no information about the first picture
is passed
on to the rest of the neural network then no, the parts of
network
capable of a FPP will have no knowledge nor experience of the first
picture.

So?

If I cut the optic nerve behind one of your eyes then show a
picture to
that eye then you would be in the same position.

Ah ok. Then supposing you were to consider the example in
http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html

regarding 3.2 the firing rules, and 3.3 the example of pattern
recognition. Now let us substitute the second 10 x 10 picture, for a 3
x 3 picture as used in the example. Let us only consider also only
three nodes that the information is passed to, as in the example on
the website. Now supposing the only information each neuron passed on
to the system was the outputs after its firing rules had operated on
the three pixels it was given. Would you still be happy that the
system could have a first person perspective of the picture?

Yes

[If you can see the point I am getting at, you have agreed that the
system wouldn't have information about the first picture in the
example in the last post, as the information wasn't communicated.
Similarly the system wouldn't have information about the firing rules,
nor would it for the system to function so as to pass the behavioural
test. Without the information about the firing rules, how does it know
what the picture was, to have a first person perspective of it.]

Wow, you made a bit of a jump there don't you think?

You went from no information being passed, to no information about how
the data was processed being passed, and for some reason you think they
are equivalent.  They aren't.

There is no need for the rest of the neural network to know how
the data
was processed, just as you don't need to know exactly what
pre-processing your retina is doing.  Look it up, there is a lot of low
level processing going on there e.g. edge detection.

http://en.wikipedia.org/wiki/Retina

So the retina is doing exactly the kind of pre-processing that
you claim
would lead to the inability to have a FPP.  Congratulations, you have
just argued that you are incapable of having a first person
perspective!

I'll say one thing though, you are showing signs of learning.  Most
dangerous don't you think?

I'm not talking about a human. I'm not suggesting the human is like a
biological robot, or hadn't you realised?

Of course I realised.  What I'm pointing out is that the retina does
exactly the kind of processing that you claim would rule out artificial
neural net!

So why do you think it rules out artificial systems from having a FPP
but not humans?  You don't even apply your own reasoning consistently.

So let's consider the 3x3 picture to be of an H, and each node
communicated a 1 to the rest of the system. As in the 3.3 recognition
example in
http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html

You said that from this information the system can have a first person
perspective of the H.

You have also said earlier that the system would not have information
that wasn't communicated to it (it wouldn't have had information about
picture 1 in the earlier example where there was one picture
communicated, and one not).

Now consider the nodes firing rules being replaced such that they will
fire 1,1,1 when the picture they receive is a black square, with white
centre. The change in the node firing rules isn't communicated to the
system. Now are you suggesting that the system has a first person
perspective of the square or the 'H'?

So.you didn't understand the web page on neural networks.  What a shame,
I though you were learning something as well.

Your getting boring again...

I did understand the website. What bit don't you think I understood?
Why don't you answer the question.

OK, I'll answer even though it's pretty obvious.

The bit you misunderstood is that 111 would represent a "T" at the
moment before your change and  "H" would be represented by 000.  If you
changed the firing rules to give 111 for a black square with a white
centre (and showed that picture) then no "H" would be seen, since that
was never the representation for "H" in the first place.  A "T" would be
seen.

In other words, if the firing rules were changed  then the output would
be corrupted and the neural network would see the wrong thing.  It would
have to re-learn what useful information could be gleaned from the new
information being passed to it.

If the firing rules were changed randomly then you would get garbage as
output as the rest of the network would be unable to adapt to
continually changing input.

The same would happen if the low level processing in the neurons in your
retina were disrupted.

So?

Actually we were going from the paper
http://www.doc.ic.ac.uk/~nd/surprise_96/journal/vol4/cs11/report.html
and in there the H was three 1's.

Look at 3.3  The H produces an output of 000 if you either read the
firing rules or simply look at the little diagram with the -> from the H
pointing to the white vertical line.  I case you hadn't realised, a
black square represents 1 and white represents 0.

The point was that the system would be communicated the same for if
the picture was an 'H' and the firing rules were for an 'H' as it
would if the picture was a square, and the firing rules were for the
square. The system isn't aware of what the firing rules are, and all
the rest of the system knows about the picture is what is what is
communicated, which was three 1's. The issue is whether the FPP of the
system (assuming it had one) would be of an H, a square, or something
else.

So?

If I changed the firing rules for the edge detecting neurons in your
retina then you would no longer be able to see edges.   Would that prove
that you cannot have a FPP?

There are lots of nice optical illusions which demonstrate that what you
think you see is not what is actually there.  Your eyes are not simply a
camera to a TV set in your brain that some kind of spiritual pixie is
watching.

Here is a very good example:

http://www.popularscience.co.uk/features/feat16.htm
--

If you look at the neural network article, 3.3 as you say, you will
see just under the diagram, above the truth tables where it says: "If
we represent black squares with 0 and white squares with 1 then the
truth tables for the 3 neurones after generalisation are;"

So care to answer the question and suggest why the H instead of the
square, or the square instead of the H?

The importance of it, is two fold, firstly it cuts through the stories
people have made up during this very discussion, where they declared
the "I" of the system would know the information by virtue of it being
in the system.

Secondly there is the representation issue, which you will be familiar
with if you actually read the article www.answernot42.com

.

Relevant Pages

  • Re: Problem for physicalist evolutionists
    ... does not a neural network make. ... they do no need to communicate anything ... along any of those network connections to "assemble" the picture. ...
    (talk.origins)
  • Re: Problem for physicalist evolutionists
    ...  You cannot build a neural network out ... 100 nodes *do* communicate the information about the second picture ... Similarly the system wouldn't have information about the firing rules, ...
    (talk.origins)
  • Re: Problem for physicalist evolutionists
    ... does not a neural network make. ... they do no need to communicate anything ... along any of those network connections to "assemble" the picture. ...
    (talk.origins)
  • Re: Problem for physicalist evolutionists
    ... sense that it could have a first person perspective of the picture. ... lot of neurons need to interact if a first person ... communicate information about picture 1, ... your idea that all of us physicalists are running away from some truth ...
    (talk.origins)
  • Re: Problem for physicalist evolutionists
    ... communicate the information they contained to the rest of the system. ... first person perspective containing the picture. ... on to the rest of the neural network then no, ... Similarly the system wouldn't have information about the firing rules, ...
    (talk.origins)