Artificial Sight

Matt Jones jonesmat at physiology.wisc.edu
Mon Jul 30 12:14:17 EST 2001


Hi Glen,

In response to your last post, first let me clarify something: When I
used the word "semantic" I was absolutley -NOT- implying triviality.
What I was trying to do is show that the disagreement occurred due to
our different usage of the same word (which it clearly does). In my
opinion, such semantic differences are not trivial. But generally a
discussion can't proceed very far if the participants are speaking in
different languages, so I think it's important to at least identify
places where different definitions can be an impediment to further
discussion.

I do not think your point of view is trivial. Nor do I think the
subject of sight is trivial, whether approached biophysically,
psychophysically or psyhologically.

Next:

> GS: *I* am not defining "message." I realize that I
> don't have that luxury. Or rather, *I* realize that to
> take such license in these circumstances is nonsense
> (see my comment on operationism). Even at this
> point, I knew what you were about to say, right
> down to citing Shannon. The problem is - and
> speaking somewhat colloquially - that you can not
> escape the meanings of colloquial terms by simply
> redefining them to meet your philosophical needs.
> Specifically, the ordinary usage of "information" and
> "message" implies a PERSON "receiving," if not
> both "receiving and sending." Ironically, even
> though this aspect of the colloquial meaning is what
> Shannon sought to eliminate by defining
> "information" in physical terms, it is just that aspect
> that allows positions like yours. 


I don't think Shannon was interested in eliminating colloquial
meanings. Neither am I. However, there simply is no such thing as "the
colloquial meaning" for any particular word, because every speaker has
a slightly different meaning in mind. Colloquial meanings are
generally ambiguous.

In science, however, meanings need to be more reproducible than that.
So people need to be clear about specific definitions. That is one
reason why different scientific definitions of information exist
(i.e., Mutual Information, Conditional Information, Fisher
Information, etc) even within the same discipline. It's not that any
one of them is "better" or more "correct". It's simply that they mean
different things, but do so unambiguously.

Nobody wants to get rid of colloquial meanings. But just try building
a communications network (or a brain) with nothing but colloquial
meanings to go on. It won't work. On the other hand, given one halfway
decent equation, where the symbols are -unambiguously- defined, one
can understand the workings of a system very precisely. -Within- the
context of the definition that was applied during the analysis.
Obviously, applying Shannon information measures to vision probably
won't satisfy many philosophers or psychologists. But that is not the
point of using that definition. The point is to arrive at -some sort
of framework- where the working of the system 'makes sense' (i.e., is
internally consistent). I have given one exmple of such an analysis,
in which Shannon Information was used to make quantitatively testable
predictions about visual processing, which were accurate within a
reasonable margin of error that would be acceptable to most
biologists. Can you give a similar example in which colloquial
definitions yield similar levels of quantitative accuracy?
 

> After all, what is on
> the receiving end of a telephone? No, not the
> "receiver".....it is the person. What comes out of a
> telephone "receiver" is not "interpreted" by the
> telephone receiver - it is interpreted by the person.

No. It -may- be interpreted by a person, but it doesn't have to be.
Maybe a better (but almost identical) example is the packet-switching
that goes on during internet communications. What is on the receiving
end of the request that your newsposter program sends to the
newsserver? Is it a person? No. It is a newsserver daemon running on
the server. Does it know that your intended message to me is that I am
"transparently vacuous"? No. What it knows is that your client sent a
specific -message- in TCP/IP, that requested a certain set of actions
(behaviors), with certain instructions for how the piggybacked usenet
post should be catalogued, displayed and stored. There are no people
involved here at all. My opinion is that something similar goes on at
many stages in the brain.

Now ultimately, in order to satisfy psychologists, one must come up
with an explanation of how -we- (i.e., whatever it is that we mean by
our consciousness) finally interpret the result of all the lower level
processing. However, that highest level interpretation is quite
obviously -optional-. The lower level messages and their lower level
interpretation goes on constantly in our brains without ever making
its way to the top where a "person" re-interprets it. In fact, it goes
on all the time in organisms in which we would probably deny that
there is any "person" home at all. Or do you not think that clams can
see?


> Even if we refer to the activity of sensory organs
> and "the things they are connected to" as
> transducing a message (which is itself nonsense),
> there is nothing to "interpret it as such" unless one
> wants to posit an homunculus, albeit implicitly. See?
> You want to say your are considering only the
> analog of a telephone (both transmitter and receiver)
> but the physics of the telephone are sterile with
> respect to the issues that should matter to
> psychology and behavioral neurobiology. 

Like I said, I'm rather less concerned with the "issues that should
matter psychology and behavioral neurobiology" than I am with
understanding how nervous systems process information. Last time I
checked, most of the real progress in that area had been made in
blowflies, crickets, squids, salamanders and crabs. Some has also been
done in cats, rats and monkeys, but less progress has been made in
those animals. What is the state of the art in blowfly psychology
these days?


>You want to eliminate
> all that is psychological but you want your telephone
> analog to address psychological issues.

God, that's the last thing I want. I want my telephone to shut up and
do its job, which is to act as a reliable information channel, plus
implement whatever source coding and channel coding algorithms that
are most likely to get the message from point A to point B with
minimal distortion. That's pretty much exactly what i want from my
visual system too. If you want to call understanding how that all
happens "psychology", that's your prerogative.



> GS: This is all beside the point. First,
> "communication channel" begs the same issue. So
> does "information," despite the fact that
> "'information' may be given a particular
> [mathematical no less!] definition" as in Information
> Theory. Since you have defined "information," how
> could it not be true that "The signals passing over
> the optic nerve to the thalamus and cortex are certainly
> 'messages' in the information theoretic sense." It is
> ironic that you raise the Stanley et al paper since it is
> Stanley et al that do the interpreting! Or are you
> saying that something in the brain acts as Stanley et
> al!
> 

Well, we can define things so that internally consistent theories can
be formulated, and then test them against experimental observations.
When I see an example of success with that tactic, I'd call it
"predicyion". Apparently, you'd call it "circular reasoning".

Most people wouldn't claim that the brain acts as the algorithm
implemented in Stanley et all (although I and many other people would
probably admit that, yes, it should be considered as a possibility
requiring further testing). Stanley et all are -not- really doing any
interpreting. They're simply crunching some spiketrains through a
particular algorithm, and then showing us the results, which to me
look quite a bit like the visual stimulus they showed the cat. So I
guess -I'm- doing the interpreting, or at least the evaluation.


 
> GS: But this is transparently vacuous. All it does is
> assert that somehow neurobiology is relevant to the
> functioning of the organism. Oh yeah.....it also
> asserts that part of what neurobiology does is to
> make (and no doubt store) copies of the world. So,
> the world is taken into the brain on the analogy of
> the telephone. But once there, you cannot say how
> this copied world "brings about actions." This is the
> weakness of mainstream cognitive psychology; the
> making of copies cannot contain anything of import
> to behavior without something to turn copies into
> behavior. But wherever and whatever the analog of
> the telephone receiver is "in the brain" it can only be
> connected up with more stuff which can also be
> given the same physical description. So how is it that
> a copy "comes out as behavior?"


<sigh>

Ok. Here's a simple example. You have a nerve cell in your leg that
connects on one side to your kneecap and on the other side to another
nerve cell. Cell #2 connects to a muscle attached to your kneecap. You
thwak your kneecap with a little rubber mallet. This activates
mechanoreceptors in cell #1, causing it to fire spikes. The spikes
travel up the axon, releasing glutamate onto cell #2. Cell #2 has
glutamate receptors that respond by causing spikes in cell#2, that
travel down its axon, releasing acetylcholine onto the muscle. The
muscle twitches, moving your leg. Familiar knee-jerk response.

This is how the copy "comes out as behaviour". By the way, it's not
just a copy. It's a series of  multiple trasnformations, each serving
a different role.


> GS: Again, the telephone receiver does not interpret.
> Information Theory is of little or no use to
> behavioral neurobiology because it has nothing to do
> with behavior. 

Information theory was not formulated to be of use to behavioral
neurobiology or behaviour. It was formulated to understand the
movement and transformation of information in the presence of noise.
But if behavioral neurobiology can't see any use for something like
that, it has bigger problems that worrying about its definitions.

 
> GS: I have already pointed out that this notion is
> silly and distracting.


Yeah. Don't forget "pernicious" and  "vacuous".


> MJ: What's the alternative? The only alternative is
> that seeing takes
> place -without- the transmission of information
> between the eyes and
> the brain.
> 
> GS: Yes, it takes place without "information" except
> when "information" is defined in such a way as to be
> devoid of import to the analysis of behavior or its
> potential reduction to nuerobiological events. In
> such a case, it is no explanation at all.


Well, I think the link between information theory and neurobiological
events is crystal clear in the examples we've been discussing. The
subject of the analysis is literally the spiketrains of neurons (which
are, um, neurobiological events).  As for it's import for the analysis
of behavior, well, it -is- behavior. In exactly the same way that
locomotion is behavior (in fact, many of the same molecules are
involved). Unless you mean something else (something 'special') when
you say behavior...


Have a nice day,

Matt



> Warmly,
> 
> Glen




More information about the Neur-sci mailing list