The Flynculus

M. F. Land's doodle "The Flynculus" depicts a miniature fly poised before
two relatively enormous compound lens view screens (displaying a window and
curtains; yes, I would have preferred a swatter), its tiny wings harnessed
via levers and pulleys to a much larger pair of wings. The caption: "The
little fly in the fly's brain trying to fly the fly".

The term "representation" seems to tax the patience, if not the imagination,
of some of those of the Behaviorist persuasion. Any talk of "representation"
elicits charges of ontological mysticism, variously characterised as faith
in either little copies of the world in the brain, chiaoscuro miniatures

flickering like shadows on the walls of some Platonic cavern, debunked by
appeals to rather gruesome thought experiments on cerebral vivesection
exposing no copies, no blue cubes, no maps, no representations, only blood
and dopamine, tattered neuroglia, and outraged avatars of once peaceful
Hebbians no longer exercising a lost right to assemble; or little men,
homunculi, scurrying from neuron to neuron, representation to
representation, reading the maps, taking notes, and pulling the levers of
motor volition. This is what you assert, claim the posts, if you speak of
representation: little worlds in the brain, little men in the brain, or
both.

Those of the software developing persuasion, schooled in Dijkstra's dictum:
data structures plus algorithms equals programs, must find such claims a
little strange - if choosing the right problem to solve is, as Aho,
Hopcroft, and Ullman claim, half the battle, then some 90% of the remaining
half is developing the best data structures for the task at hand, the best
available representations of the data, and efficient transformations between
them. And when the task comes from the realm of digital signal processing it
is at most a minor exageration to claim that finding alternative
representations, felicitous decompositions of the signal into well chosen
time-frequency atoms is the name of the game: to locate and label transient
coherent structures in the passing show, be it a matter of including
multi-scale information in the characterization of hydrolic conductivity
distributions, wavelet analysis of diurnal and nocturnal turbulence above a
maize crop, or simultaneous noise suppression and signal compression using a
library of orthonormal bases and the minimum description length criterion -
all are matters of transforming a signal to representations enhancing the
detectability of features of interest.

Few or no engineers fret, I dare say,
that such transformations, or talk of effected representations, lead to an
infinite regress of computunculi - a need for little computers netested in
each other to read the maps and pull the levers.

Think of a thermostat as an intentional agent, a goal-directed sensory-motor
loop. That its innards represent, somehow, a desired state of affairs, ideal
room temperature, and that beneath the skin it maintains also a represention
of the current state of affairs, ambient temperature, and that such
reprentations can be effected via a variety of physical substrates, does
not, evidently, doom such a device to inaction without the aid of nested
thermunculi. Why, then, should we believe what is possible for thermostats,
representation without regress, is unavailable to spiking neurons? I suspect
that at least some of the disagreement stems from divergent notions of
"representation".

Here is a simple experiment you can try at home: watch a fly buzzing about a
room. You might notice that the fly's flight path consists mainly of
relatively straight segments punctuated by sharp turns. Quantitative
tracking confirms this. Now turn out the lights. The fly lands (you may need
to flick the lights back on for confirmation, timing is critical, practice
may be required). Luke Skywalker can navigate by trusting the force, but
flies seem to rely heavily on the video signal (even though, like all
eukaryotes, they are loaded with microtubules):

"One can demonstrate the visual input to flight control by tethering the fly
so that it hangs, wings flapping, from a torsion balance. If the visual
environment of the fly rotates (on a drum surrounding the fly, or on a video
monitor), the fly generates a torque. The sign of the torque is such that it
tends to compensate the rotational motion. One can close the sensory-motor
feedback loop artificially by giving the visual environment an added
velocity proportional to the negative of the measured torque, as would
happen if the fly were free to turn. Under these closed loop conditions, the
fly will spontaneously fixate an object, creating (as best as possible under
the circumstances) the image of flying straight toward that object" (Rieke
et al.).

Now, most observers agree that it is not the angular velocity of the visual
field that generates torque, rather it is the flapping of wings. Further, I
will go way out on a limb here and predict that removing the fly's retinal
manifest would result in torques not correlated to the visual environment.
Between fly retinas and wings there are some levers and pulleys, and a few
spiking neurons to boot, among them the H1 neurons, which "respond most
strongly to 'wide field' motion - coherent motion across the entire visual
field, as would be induced by rigid rotation of the fly itself" (again,
Rieke and crew). "Under favorable conditions it is possible to record
continuously from H1 for periods of many days, using an imobilized fly,
almost completely intact save for a small hole in the back of the head that
allows access to the lobula plate. In these long experiments one must pause
occasionally to feed the fly, but it should be clear that this very stable
preparation makes it possible to address questions that require a very large
statistical sample of neural responses.

From such studies it is possible to characterize the "response conditional
ensemble" P[s(tau)|t{i}], where s(tau) is a signal, in this case the value
of the angular velocity of wide field horizontal motion at time tau, and
t{i} is a train of spikes, a spike occuring at each time i, and
P[s(tau)|t{i}] is the conditional probability of having observed the angular
velocity waveform s(tau), given the spike train response t{i}. Given the
spike train it is possible to reconstruct the stimulus. The spike train
ecnodes the stimulus. The spike train represents the stimulus. The neural
code, in this case, has been deciphered.

"We can 'read' the spike train and translate back all the way to the
stimulus itself.... Stimulus reconstruction is not necessarily a problem the
animal must solve. It is, however, of the same character as the problems the
animal must solve. For example, the fly can initiate a turn based on visual
motion signals alone, which means that it translates the spike output of the
motion sensitive visual neurons into a torque, and this torque has a
component roughly proportional to the time dependdent angular velocity. The
torque signal is a continuous analog waveform that the fly synthesizes out
of discrete spike sequences in its sensory neurons. The problem of
recovering analog signals from the spike train is then a fundamental step in
the neural processing of sensory data."

Rieke, Warland, Ruyter von Steveninck, Bialek, Spikes: Exploring the Neural
Code".

Hello Michael,

Thanks for the extraordinarily high quality post. Something I have
been missing on c.a.p. :)

Of course the behaviorist posts by Doktor DynaSoar or Wolf are quite
misdirected.

1. There *is* neuroscientific evidence for representation in the
brain.
2. Representation doesn't mean homunculi or silly ontological
commitments.

Our behaviorists have no idea what representation means.


And indeed, some of the representations we have discovered in
theoretical subjects such as signal processing may constitute some
cognitive universals.

Few or no engineers fret, I dare say,
that such transformations, or talk of effected representations, lead to an
infinite regress of computunculi - a need for little computers netested in
each other to read the maps and pull the levers.

Good analogy, and one that effectively demolishes the silly homunculi
arguments of the behaviorist infestation. Those arguments belong to
about 1950, IIRC. They have no place in 21st century, because we know
a good deal more about complex systems, be it artificial or natural.

Hello Michael,

Thanks for the extraordinarily high quality post.

Michael Olea <oleaj@sbcglobal.net> wrote in message
news:<BC39E9EF.4C79%oleaj@sbcglobal.net>...
The Flynculus

"One can demonstrate the visual input to flight control by tethering the fly
so that it hangs, wings flapping, from a torsion balance. If the visual
environment of the fly rotates (on a drum surrounding the fly, or on a video
monitor), the fly generates a torque. The sign of the torque is such that it
tends to compensate the rotational motion. One can close the sensory-motor
feedback loop artificially by giving the visual environment an added
velocity proportional to the negative of the measured torque, as would
happen if the fly were free to turn. Under these closed loop conditions, the
fly will spontaneously fixate an object, creating (as best as possible under
the circumstances) the image of flying straight toward that object" (Rieke
et al.).

Now, most observers agree that it is not the angular velocity of the visual
field that generates torque, rather it is the flapping of wings. Further, I
will go way out on a limb here and predict that removing the fly's retinal
manifest would result in torques not correlated to the visual environment.
Between fly retinas and wings there are some levers and pulleys, and a few
spiking neurons to boot, among them the H1 neurons, which "respond most
strongly to 'wide field' motion - coherent motion across the entire visual
field, as would be induced by rigid rotation of the fly itself" (again,
Rieke and crew). "Under favorable conditions it is possible to record
continuously from H1 for periods of many days, using an imobilized fly,
almost completely intact save for a small hole in the back of the head that
allows access to the lobula plate. In these long experiments one must pause
occasionally to feed the fly, but it should be clear that this very stable
preparation makes it possible to address questions that require a very large
statistical sample of neural responses."

From such studies it is possible to characterize the "response conditional
ensemble" P[s(tau)|t{i}], where s(tau) is a signal, in this case the value
of the angular velocity of wide field horizontal motion at time tau, and
t{i} is a train of spikes, a spike occuring at each time i, and
P[s(tau)|t{i}] is the conditional probability of having observed the angular
velocity waveform s(tau), given the spike train response t{i}. Given the
spike train it is possible to reconstruct the stimulus. The spike train
ecnodes the stimulus. The spike train represents the stimulus. The neural
code, in this case, has been deciphered.

It should be. If the encoding at the output of a sensory neuron were
too complex to decipher, it would be meaningless. BTW, is the
reconstruction perfect?

"We can 'read' the spike train and translate back all the way to the
stimulus itself.... Stimulus reconstruction is not necessarily a problem the
animal must solve. It is, however, of the same character as the problems the
animal must solve. For example, the fly can initiate a turn based on visual
motion signals alone, which means that it translates the spike output of the
motion sensitive visual neurons into a torque, and this torque has a
component roughly proportional to the time dependdent angular velocity. The
torque signal is a continuous analog waveform that the fly synthesizes out
of discrete spike sequences in its sensory neurons. The problem of
recovering analog signals from the spike train is then a fundamental step in
the neural processing of sensory data."

Rieke, Warland, Ruyter von Steveninck, Bialek, Spikes: Exploring the Neural
Code".

Absolutely fantastic. Here are the bits in CNS.

However, this seems to me a long way from determining the functions
over bits.

Regards,

--
Eray Ozkural

The Flynculus

M. F. Land's doodle "The Flynculus" depicts a miniature fly poised before
two relatively enormous compound lens view screens (displaying a window
and
curtains; yes, I would have preferred a swatter), its tiny wings harnessed
via levers and pulleys to a much larger pair of wings. The caption: "The
little fly in the fly's brain trying to fly the fly".

The term "representation" seems to tax the patience, if not the
imagination,
of some of those of the Behaviorist persuasion. Any talk of
"representation"
elicits charges of ontological mysticism, variously characterised as faith
in either little copies of the world in the brain, chiaoscuro miniatures
flickering like shadows on the walls of some Platonic cavern, debunked by
appeals to rather gruesome thought experiments on cerebral vivesection
exposing no copies, no blue cubes, no maps, no representations, only blood
and dopamine, tattered neuroglia, and outraged avatars of once peaceful
Hebbians no longer exercising a lost right to assemble; or little men,
homunculi, scurrying from neuron to neuron, representation to
representation, reading the maps, taking notes, and pulling the levers of
motor volition. This is what you assert, claim the posts, if you speak of
representation: little worlds in the brain, little men in the brain, or
both.

Those of the software developing persuasion, schooled in Dijkstra's
dictum:
data structures plus algorithms equals programs, must find such claims a
little strange - if choosing the right problem to solve is, as Aho,
Hopcroft, and Ullman claim, half the battle, then some 90% of the
remaining
half is developing the best data structures for the task at hand, the best
available representations of the data, and efficient transformations
between
them. And when the task comes from the realm of digital signal processing
it
is at most a minor exageration to claim that finding alternative
representations, felicitous decompositions of the signal into well chosen
time-frequency atoms is the name of the game: to locate and label
transient
coherent structures in the passing show, be it a matter of including
multi-scale information in the characterization of hydrolic conductivity
distributions, wavelet analysis of diurnal and nocturnal turbulence above
a
maize crop, or simultaneous noise suppression and signal compression using
a
library of orthonormal bases and the minimum description length
criterion -
all are matters of transforming a signal to representations enhancing the
detectability of features of interest. Few or no engineers fret, I dare
say,
that such transformations, or talk of effected representations, lead to an
infinite regress of computunculi - a need for little computers netested in
each other to read the maps and pull the levers.

Think of a thermostat as an intentional agent, a goal-directed
sensory-motor
loop. That its innards represent, somehow, a desired state of affairs,
ideal
room temperature, and that beneath the skin it maintains also a
represention
of the current state of affairs, ambient temperature, and that such
reprentations can be effected via a variety of physical substrates, does
not, evidently, doom such a device to inaction without the aid of nested
thermunculi. Why, then, should we believe what is possible for
thermostats,
representation without regress, is unavailable to spiking neurons? I
suspect
that at least some of the disagreement stems from divergent notions of
"representation".

Here is a simple experiment you can try at home: watch a fly buzzing about
a
room. You might notice that the fly's flight path consists mainly of
relatively straight segments punctuated by sharp turns. Quantitative
tracking confirms this. Now turn out the lights. The fly lands (you may
need
to flick the lights back on for confirmation, timing is critical, practice
may be required). Luke Skywalker can navigate by trusting the force, but
flies seem to rely heavily on the video signal (even though, like all
eukaryotes, they are loaded with microtubules):

"One can demonstrate the visual input to flight control by tethering the
fly
so that it hangs, wings flapping, from a torsion balance. If the visual
environment of the fly rotates (on a drum surrounding the fly, or on a
video
monitor), the fly generates a torque. The sign of the torque is such that
it
tends to compensate the rotational motion. One can close the sensory-motor
feedback loop artificially by giving the visual environment an added
velocity proportional to the negative of the measured torque, as would
happen if the fly were free to turn. Under these closed loop conditions,
the
fly will spontaneously fixate an object, creating (as best as possible
under
the circumstances) the image of flying straight toward that object" (Rieke
et al.).

Now, most observers agree that it is not the angular velocity of the
visual
field that generates torque, rather it is the flapping of wings. Further,
I
will go way out on a limb here and predict that removing the fly's retinal
manifest would result in torques not correlated to the visual environment.
Between fly retinas and wings there are some levers and pulleys, and a few
spiking neurons to boot, among them the H1 neurons, which "respond most
strongly to 'wide field' motion - coherent motion across the entire visual
field, as would be induced by rigid rotation of the fly itself" (again,
Rieke and crew). "Under favorable conditions it is possible to record
continuously from H1 for periods of many days, using an imobilized fly,
almost completely intact save for a small hole in the back of the head
that
allows access to the lobula plate. In these long experiments one must
pause
occasionally to feed the fly, but it should be clear that this very stable
preparation makes it possible to address questions that require a very
large
statistical sample of neural responses.

From such studies it is possible to characterize the "response
conditional
ensemble" P[s(tau)|t{i}], where s(tau) is a signal, in this case the value
of the angular velocity of wide field horizontal motion at time tau, and
t{i} is a train of spikes, a spike occuring at each time i, and
P[s(tau)|t{i}] is the conditional probability of having observed the
angular
velocity waveform s(tau), given the spike train response t{i}. Given the
spike train it is possible to reconstruct the stimulus. The spike train
ecnodes the stimulus. The spike train represents the stimulus. The neural
code, in this case, has been deciphered.

"We can 'read' the spike train and translate back all the way to the
stimulus itself.... Stimulus reconstruction is not necessarily a problem
the
animal must solve. It is, however, of the same character as the problems
the
animal must solve. For example, the fly can initiate a turn based on
visual
motion signals alone, which means that it translates the spike output of
the
motion sensitive visual neurons into a torque, and this torque has a
component roughly proportional to the time dependdent angular velocity.
The
torque signal is a continuous analog waveform that the fly synthesizes out
of discrete spike sequences in its sensory neurons. The problem of
recovering analog signals from the spike train is then a fundamental step
in
the neural processing of sensory data."

Rieke, Warland, Ruyter von Steveninck, Bialek, Spikes: Exploring the
Neural
Code".

MO: This is what you assert, claim the posts, if you speak of
representation: little worlds in the brain, little men in the brain, or
both.

GS: Indeed, this is the assertion. And the reason is simple;
"representation," like "computation" is not a physical property or set of
properties of anything (though they are physical) - they are defined by
their effect on a person. We respond to representations in some of the same
ways we respond to the things they represent. We respond to a picture of a
tree in some of the same ways we do a tree. THAT is what makes the picture a
"representation."

in article 98c865a9d3261092c2b78e6a411db704@news.teranews.com, Glen M.
Sizemore at gmsizemore2@yahoo.com wrote on 1/27/04 6:39 AM:


MO: This is what you assert, claim the posts, if you speak of
representation: little worlds in the brain, little men in the brain, or
both.

GS: Indeed, this is the assertion. And the reason is simple;
"representation," like "computation" is not a physical property or set
of
properties of anything (though they are physical) - they are defined by
their effect on a person. We respond to representations in some of the
same
ways we respond to the things they represent. We respond to a picture of
a
tree in some of the same ways we do a tree. THAT is what makes the
picture a
"representation."

"There are 10 kinds of people, those who understand binary numbers, and
those who don't" - anonymous

Oh....one last thing. The fact that certain metaphors don't cause problems
when used in certain areas (even when they can be shown to be silly) is no
measure that they won't cause problems elsewhere.

MO: "There are 10 kinds of people, those who understand binary numbers, and
those who don't" -anonymous


GS: No, there are two kinds of people in the world: those that are
interested in finding out how Nature works, and those that are interested in
finding out whether or not their theories were correct.

On Tue, 27 Jan 2004 14:39:04 GMT, "Glen M. Sizemore"
gmsizemore2@yahoo.com> wrote:

Oh....one last thing. The fact that certain metaphors don't cause problems
when used in certain areas (even when they can be shown to be silly) is no
measure that they won't cause problems elsewhere.

and the fact that certain metaphors do cause problems somewhere is no
measure that they will do elsewhere

What you say is useless


Pat

Oh....one last thing. The fact that certain metaphors don't cause problems
when used in certain areas (even when they can be shown to be silly) is no
measure that they won't cause problems elsewhere.

On Tue, 27 Jan 2004 14:39:04 GMT, "Glen M. Sizemore"
gmsizemore2@yahoo.com> wrote:

Oh....one last thing. The fact that certain metaphors don't cause
problems
when used in certain areas (even when they can be shown to be silly) is
no
measure that they won't cause problems elsewhere.

and the fact that certain metaphors do cause problems somewhere is no
measure that they will do elsewhere

What you say is useless


Pat

MO: "There are 10 kinds of people, those who understand binary numbers,
and
those who don't" -anonymous


GS: No, there are two kinds of people in the world: those that are
interested in finding out how Nature works, and those that are interested
in
finding out whether or not their theories were correct.

"Michael Olea" <oleaj@sbcglobal.net> wrote in message
news:BC3C895A.4D8E%oleaj@sbcglobal.net...

in article 98c865a9d3261092c2b78e6a411db704@news.teranews.com, Glen M.
Sizemore at gmsizemore2@yahoo.com wrote on 1/27/04 6:39 AM:


MO: This is what you assert, claim the posts, if you speak of
representation: little worlds in the brain, little men in the brain,
or
both.

GS: Indeed, this is the assertion. And the reason is simple;
"representation," like "computation" is not a physical property or set
of
properties of anything (though they are physical) - they are defined
by
their effect on a person. We respond to representations in some of the
same
ways we respond to the things they represent. We respond to a picture
of
a
tree in some of the same ways we do a tree. THAT is what makes the
picture a
"representation."

"There are 10 kinds of people, those who understand binary numbers, and
those who don't" - anonymous

In article <nn5f101h776jthdj7313ec25basblrd7gh@4ax.com>,
Bouh@?.?.invalid writes
On Tue, 27 Jan 2004 14:39:04 GMT, "Glen M. Sizemore"
gmsizemore2@yahoo.com> wrote:

Oh....one last thing. The fact that certain metaphors don't cause
problems
when used in certain areas (even when they can be shown to be silly) is
no
measure that they won't cause problems elsewhere.

and the fact that certain metaphors do cause problems somewhere is no
measure that they will do elsewhere

What you say is useless


Pat

It may just be useless to *you* because *you* don't know how to use it
appositely.

"David Longley" <David@longley.demon.co.uk> wrote in message
news:3ryyrHFbI6FAFw3$@longley.demon.co.uk...
In article <nn5f101h776jthdj7313ec25basblrd7gh@4ax.com>,
Bouh@?.?.invalid writes
On Tue, 27 Jan 2004 14:39:04 GMT, "Glen M. Sizemore"
gmsizemore2@yahoo.com> wrote:

Oh....one last thing. The fact that certain metaphors don't cause
problems
when used in certain areas (even when they can be shown to be silly) is
no
measure that they won't cause problems elsewhere.

and the fact that certain metaphors do cause problems somewhere is no
measure that they will do elsewhere

What you say is useless


Pat

It may just be useless to *you* because *you* don't know how to use it
appositely.


Since you, Longley, (Sizemore a bit less - although his actual knowledge of
certain sciences is woefully lacking- at least he can venture outside the
realm of radical behaviorism once in a while if goaded), post the same
drivel against every issue, context or argument, we can only assume that
your scientific breadth is severely constrained. Unlike myself or others
here that can reference multiple sources , conceptualize and discuss many
many aspects of reality as we know it so far, and provide fodder for thought
in a plethora of contexts.

In article <1faf3acb435ddc7ef53b604f0343d571@news.teranews.com>,
OmegaAlpha2004 <OmegaZero2003@yahoo.com> writes

"David Longley" <David@longley.demon.co.uk> wrote in message
news:3ryyrHFbI6FAFw3$@longley.demon.co.uk...
In article <nn5f101h776jthdj7313ec25basblrd7gh@4ax.com>,
Bouh@?.?.invalid writes
On Tue, 27 Jan 2004 14:39:04 GMT, "Glen M. Sizemore"
gmsizemore2@yahoo.com> wrote:

Oh....one last thing. The fact that certain metaphors don't cause
problems
when used in certain areas (even when they can be shown to be silly)
is
no
measure that they won't cause problems elsewhere.

and the fact that certain metaphors do cause problems somewhere is no
measure that they will do elsewhere

What you say is useless


Pat

It may just be useless to *you* because *you* don't know how to use it
appositely.


Since you, Longley, (Sizemore a bit less - although his actual knowledge
of
certain sciences is woefully lacking- at least he can venture outside the
realm of radical behaviorism once in a while if goaded), post the same
drivel against every issue, context or argument, we can only assume that
your scientific breadth is severely constrained. Unlike myself or others
here that can reference multiple sources , conceptualize and discuss many
many aspects of reality as we know it so far, and provide fodder for
thought
in a plethora of contexts.



You're an idiot, and what you keep posting here is idiotic because you
don't realise that I (and Sizemore) have been presenting a systematic
*philosophical* approach to some specific issues which bear on the
philosophy of AI. That's why, to some, what we post has value.

You, on the other hand, like Michaels, browse, what for you, amounts to
little more than a large candy store of science fiction 'fodder', and
you do so much the same way that others browse and recount newspaper
stories.

You haven't a clue what philosophy or science is. The reason? Because at
best, you're an ill educated, ignorant and irreverent technician - and
you're incorrigible with it.

You'd do yourself more good by listening and learning.
--
David Longley