ambiguous figures

[From Bill Powers (2004.05.09.1342 MST)]

Martin Taylor 2004.05.09.10.15 --

Since I don't seem to get the "attractor" notion yet, I thought I'd do
something with ambiguous figures. Attached is an interesting experiment in
a preliminary form that others might want to expand on and improve. It's a
dynamic Necker cube and the zipped attachment is called DynaNeck.zip.

The source code for Delphi is included as well as the executable code (the
executable file is called Project2.exe, since I didn't bother to change any
files from their default names. You can rename it something more sensible).

A cube is shown on the screen which, with the mouse cursor centered on the
screen, wanders in two angular directions around the face-centered
orientation. The mouse can also rotate the cube left-right and up-down. The
object of the task is to use the mouse to keep the view exactly
face-centered.(that is, the rear square exactly behind the front square, so
the figure appears to be a single square).

An experimental run lasts for one minute. The angular errors are recorded,
and after the run are plotted, x in red and y in green.

As you will see, this task is devilishly hard. Since no perspective is used
(the z dimension is suppressed), there is absolutely no way to tell which
is the front square and which is the back one. As a result, the effect of
the mouse on whichever square you think you're watching changes sign each
time you inadvertently switch perceptions. This leads to a momentary
runaway condition followed by a recovery, as in the paper by Rick and me in
_Mind Readings_. These are fairly easy to see as spikes in the plot, and
will probably become more obvious with a well-practiced subject.

There are other orientations of the cube that the participant can try to
maintain -- I've done almost no experimenting with this yet. When the
Necker effect occurs, the sign of feedback would presumably change, causing
the runaway and recovery when the controlling person switches the sign of
the output to get back to negative feedback.

The screen is sampled 60 times per second; there are 3600 data points. The
data are not yet recorded in a file -- if anyone's interested I'll add that
so you can import the numbers in ASCII to other programs like spreadsheets.
Other suggestions are welcome, though I hope someone will volunteer to take
over the improvements to this idea.

Note that while there are switches in perception, the switch is not just
between one static configuration and another one. The whole figure is
affected even while its orientation is changing. And if preliminary
appearances hold up, the switch is strictly perceptual, since it seems to
cause a positive feedback situation whenever it occurs, with the output
function then switching to compensate some time later -- 400 milliseconds
later, if we extrapolate from the paper in _Mind Readings_.

A similar thing can be done with bubbles and dents by adjusting the shading
in a circular image according to Lambert's Law. I hope someone else beats
me to it. I'm really working on other things.

Best,

Bill P.

DynaNeck.ZIP (197 KB)

[From Rick Marken (2004.05.09.1510)]

Bill Powers (2004.05.09.1342 MST)--

Since I don't seem to get the "attractor" notion yet, I thought I'd do
something with ambiguous figures. Attached is an interesting
experiment in
a preliminary form that others might want to expand on and improve.
It's a
dynamic Necker cube and the zipped attachment is called DynaNeck.zip.

This is _really_ great, Bill. I don't suppose you'd believe me if I
said I was thinking of building just this kind of experiment when the
ambiguous figure issue came up. I was going to build a java tracking
program where you control the necker cube's orientation using a mouse.
It turns out to be a very nifty way of doing the polarity reversal
expriment (a la the _Mind Readings_ paper) by letting the perceptual
system create the reversal. I'll covert your code to java as soon as I
have the time.

The data are not yet recorded in a file -- if anyone's interested I'll
add that
so you can import the numbers in ASCII to other programs like
spreadsheets.
Other suggestions are welcome, though I hope someone will volunteer to
take
over the improvements to this idea.

The only suggestion I have would be to plot the disturbance and mouse
movements along with the error curve. I can see the spikes in the error
curve where my perceptual reversals (producing he positive feedback
effet) occur but I think it would be even clearer (given the raggednes
of the error curve) and we could get a better idea of the dynamics of
the changes if we could see the relationship between disturbance and
output. Of course, unlike in the _Mind Readings_ paper we can't tell
exactly when the feedback polarity reversal (perceptual reversal in
this case) occurs. But maybe we can infer it from the
disturbance/response relationship.

Best

Rick

ยทยทยท

---
Richard S. Marken
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

[From Bill Powers (2004.05.09.1622 MST)]

Rick Marken (2004.05.09.1510)--

The only suggestion I have would be to plot the disturbance and mouse
movements along with the error curve. I can see the spikes in the error
curve where my perceptual reversals (producing he positive feedback
effet) occur but I think it would be even clearer (given the raggednes
of the error curve) and we could get a better idea of the dynamics of
the changes if we could see the relationship between disturbance and
output.

Right. Actually the numbers _are_ the error, but it would be nice to see
output and disturbance too. Be my guest.

I rather suspect that Bruce Abbott will be along with a much nicer version
and more whistles and bells, in Delphi so it's easy to run on a PC.
Assuming that you guys will get to work, I'll look into the bubbles and
dents demo. Actually it will be easier to do as a trench and ridge -- not
so many complex calculations.

Best,

Bill P.

[Martin Taylor 2004.05.08.2040]

[From Bill Powers (2004.05.09.1342 MST)]

Martin Taylor 2004.05.09.10.15 --

Since I don't seem to get the "attractor" notion yet,

Should I continue my efforts? I think it could be worthwhile.

I thought I'd do
something with ambiguous figures. Attached is an interesting experiment in
a preliminary form that others might want to expand on and improve. It's a
dynamic Necker cube and the zipped attachment is called DynaNeck.zip.

That's lovely. I tried running it in my simulated PC, but I guess
it's too slow, because one or more lines of the cube keeps flickering
out as the cube rotates. I don't suppose there's anything you can do
to the code to allow the display to keep showing. It's only 12 lines,
isn't it?

I tried keeping the configuration at the hexagon with the diagonals
drawn. I don't know whether that was easier or more difficult, but
the final display presents a problem. The upper graph shows perfectly
well, but only the highest peaks of the lower graph peek out from the
bottom edge of the window frame. Could you add a scroll bar without
too much trouble, or subtract the average error before graphing?

I can accept that the spikes are due to perceptual reversals, but
there's no perception OF the reversals. That's something of an
oddity, since ordinarily when you are looking at an ambiguous figure,
the moments of change cause quite a perceptual wrench (I don't know
what else to call it).

Martin

[From Bill Powers (2004.05.09.2012 MST)]

Martin Taylor 2004.05.08.2040 --

That's lovely. I tried running it in my simulated PC, but I guess
it's too slow, because one or more lines of the cube keeps flickering
out as the cube rotates. I don't suppose there's anything you can do
to the code to allow the display to keep showing. It's only 12 lines,
isn't it?

The actual code isn't much, though the rotation routine requires eight sine
and cosine computations per iteration. I used to know how to synchronize
displays to the vertical retrace rate, but with all the improvements to
Windows I can't figure out how to do it now. If anyone knows I'd appreciate
a tip. I hope you can find a PC to test it on.

I tried the "trench and ridge" version of bubbles and dents, and it's
totally unconvincing. No impression of depth at all, just two rectangles
getting brighter and darker. Do you happen to have a picture of a bubble or
dent that you could scan for me? With a bitmap image I could just vary the
proportional brightness of each pixel and make the smooth transitions that
way. Come to think of it, I ought to be able to take a picture of something
myself with my little digital camera, so never mind.

I tried keeping the configuration at the hexagon with the diagonals
drawn. I don't know whether that was easier or more difficult, but
the final display presents a problem. The upper graph shows perfectly
well, but only the highest peaks of the lower graph peek out from the
bottom edge of the window frame. Could you add a scroll bar without
too much trouble, or subtract the average error before graphing?

Good idea, I'll do that tomorrow (subtract the average). I ran into the
same problem. Maybe, just for now, I'll put a slider control on the screen
to adjust the frame rate -- then you can at least stop the blanking bars
from rolling through the display.

I can accept that the spikes are due to perceptual reversals, but
there's no perception OF the reversals. That's something of an
oddity, since ordinarily when you are looking at an ambiguous figure,
the moments of change cause quite a perceptual wrench (I don't know
what else to call it).

The case of keeping the square face-on, or in your case keeping the
diagonals intersecting in the center, is not quite like the usual effect.
The ambiguity is total -- there is simply no way to know whether the
feature you've selected is in the foreground or background. When the
squares merge for a moment, they might separate again after the rotation
has reversed, or with the same direction of rotation, and the displays
would look exactly the same. So the relationship of the mouse movement to
the apparent direction of cube rotation is completely indeterminate -- it
can reverse very rapidly sometimes.

For me, the reversals do not look like reversals, either; it feels as if
the mouse just stops working right for a moment, until I can switch myself
to the opposite polarity of control. Making that switch is occasionally
very hard -- it feels as if I want to switch and know I should switch, but
can't do it. I feel that my internal polarity switching has a large
cognitive component, though I don't know how effective it really is.
Sometimes I know that I should switch before the switch occurs. It doesn't
feel very automatic.

As to "attractors," I'm not sure how interested I am -- it seems to me that
there are ways to approach the same phenomena without all the metaphors.
Why not just go on working on your exposition and try it out on me when you
think you have a point to make? I don't want to be a wet blanket, but I
have too much going on to be very receptive. I think it's partly just a
matter of different tastes in explanatory methods.

Best,

Bill P.