Date:         Thu, 23 Sep 1999 23:26:24 +1000
Reply-To:     "PSYCHE Discussion Forum (Biological/Psychological emphasis)"
              <[log in to unmask]>
Sender:       "PSYCHE Discussion Forum (Biological/Psychological emphasis)"
              <[log in to unmask]>
From:         Bruce Mangan <[log in to unmask]>
Subject:      Re: Chicken sexing (real scientific inquiry!)
Content-type: text/plain; charset=us-ascii

I've just finished reading the Biederman & Shiffrar (1987) paper on
chicken sexing. It contains some indirect evidence for implicit
cognition that I'll get to at the end of this note. But the most
striking thing about the Biederman paper is that it presents very solid
experimental evidence for a consciously mediated process of EXPLICIT
cognition. I gather this is just what Minsky asked us for more than a
week ago:

Who can supply a provable example of explicit cognition. Aha, you claim
that you recognize the chicken because of its feet, or its, beak, or its
cute little eyes. Well, that's your conscious theory. But most
commonsense theories turn out to be wrong-and I suspect that this will
often apply to "articulated criteria".

I can't let the last sentence above of Minsky's pass without a brief
challenge. Maybe he just made a slip of the keyboard. Does Minsky really
believe that MOST commonsense theories are wrong? And would he actually
have us use this proposition as an evaluative principle in science? Is a
theory really more likely to be true if, all else being equal, it
*violates* commonsense rather than resonates with it? At least in
relation to the humble act of chicken sexing, it looks like we do have,
contrary to Minsky's intuition, solid evidence for recognition via
articulated feature identification.

(1) I find the background of Biederman's experiment fascinating in its
own right. The current method for sexing baby chicks (99.4% success
rate) was introduced into North America by a group of Japanese chicken
sexers invited to the University of British Columbia in 1933. The
Japanese technique was to invert the chick's vent (cloaca) and focus on
a region of tissue about the size of a pin head. It is still difficult
to reliably discriminate male from female this way, in part because
there is a good deal of variation among baby chick's genitalia (now
called the "bead" in the profession). But by showing their audience
numerous physical examples of male and female genitalia augmented by
verbal descriptions of the many different kinds of sub-type features
(passed laboriously through an interpreter) the Japanese were finally
able to transfer their knowledge. Schools of chicken sexing sprang up
immediately, and the Japanese technique soon became the standard method
for chicken sexing, at least on this continent.

In 1984, a human interest story in the San Francisco Chronicle told
about the retirement of one Heimer Carlson. He had sexed over 55 million
chicks in his career, and was one of the first people to graduate from a
chicken sexing school set up immediately after the Japanese excursion to
Canada. The experimenters visited Carlson, showed him photos of male and
female inverted cloaca taken of difficult genital features to sex, and
asked Carlson to circle what he saw as the bead in each picture. I'd
estimate that the beads occupy something like 10%-15% of the total area
of each photo.

To quote Biederman & Shiffar: "Our examination of the [circled
photographs of the] beads revealed a simple difference in the contours
between males and females. In males, the eminence was convex; in
females, flat or concave. This differentiation corresponded to the
descriptions offered by some of the sexers who described males as
'round' and females as 'pointy.' When a sketch of shapes of the
eminences was drawn ... and showed to Mr. Carlson, he agreed to its
diagnosticity." It does not seem from this account that Carlson himself
had consciously formulated a round vs. pointy description of baby chick
genitalia or some other simple schematic; but Carlson did agree with the
remarkably simple diagrams the experimenters devised when they showed
them to him.

(2) These diagrams then served as the basis of Biederman & Shiffrar's
experiment. Their overall procedure and findings were as follows
(quoting from the abstract):

"Naive subjects were shown 18 pictures of cloacal regions of male and
female chicks ... and asked to judge the sex of each chick. The pictures
indicated a number of rare and difficult configurations. The subjects
were then instructed as to the location of a critical cloacal structure
for which a simple contrast in shape (convex vs. concave or flat) could
serve as a indicant of sex. When the subjects judged the pictures again
... accuracy increased from slightly above chance to a level comparable
to that achieved by a sample of experts. The correlation (over items)
between the naive subjects and the experts before instruction was .21;
after instruction, .82 .... The rate of learning in these cases could be
greatly increased through the use of simple instructions that specified
the location of diagnostic contour contrasts."

(3) I do not see how anyone could deny that in this case the subject's
increased ability to recognize a male or female baby chick was mediated
by the antecedent conscious identification of minute features in
chicken's sex organs. All it took to do this was a few simple, well
chosen diagrams and descriptive phrases. This was enough to shift the
behavioral capacity of naive subjects to that of seasoned professionals
in the photo identification task. I gather that the instruction phase of
the experiment took less than five minutes.

Nothing in this means that consciously mediated feature analysis is in
general the ONLY cognitive process behind discriminant recognition. It
does mean that this "commonsense" theory of recognition is certainly
true sometimes, and that recognition via explicit feature identification
IS A VERY NATURAL AND POWERFUL COGNITIVE STRATEGY, WELL ADAPTED TO THE
OPERATION OF OUR ORGANISM IN GENERAL AND TO THE NEEDS OF CONSCIOUSNESS
IN PARTICULAR. Biederman gives a collateral example of American solders
learning to distinguish Warsaw Pact from NATO tanks that is also based
on a simple feature identification: The back section of all Warsaw tank
turrets were completely rounded; NATO tank turrets had linear backs. And
of course we all know of a huge number of analogous examples, e.g., the
criteria naturalists use in the field to quickly distinguish similar
species from one another as opposed to DNA analysis in the lab.

In their professional training, chicken sexers are expected to remember
many different complex types of sex features. In contrast, the full set
of features used in the Biederman and Shiffrar experiment was simplicity
itself. Professional training aims to sex-type a host of different rare
configurations. It may be that this degree of complex feature analysis
is not strictly necessary. (But it would still make sense in the larger
social context: The students feel they are getting their money's worth
from the chicken sexing schools, and the skill seems more difficult and
so perhaps more valuable to the larger world of the poultry industry).
These findings also illustrate the plasticity of the feature analysis
technique for human cognition. Complex biological systems often
accomplish the same end using a variety of means. So, without intending
to, the Biederman and Shiffar experiment may also show us it may be
possible for either an implicit or an explicit mechanism to accomplish,
roughly speaking, the same cognitive end.

(4) The indirect evidence for implicit cognition in Biederman's paper
derives from the fact that his subjects were able to correctly sex the
photos significantly above chance (62% percent) BEFORE they had any
instruction in explicit feature identification. Biederman and Shiffar
try to explain this by speculating that "the most likely reason for this
level of accuracy was that the subjects' not-so-naive hypothesis about
what might be diagnostic was partially accurate. For the most part, the
presence of a prominent bead -- which is the central singular structure
-- was interpreted as being male." There is no report of subjects
actually saying anything like this, but by itself this speculation is
certainly plausible.

However, the paper begins by emphasizing a very different point: "Like
many perceptual psychologists, the first author (Irving Biederman) was
introduced to this problem by the first figure in E.J Gibson's
*Principles of Perceptual Learning and Development* (1969). Professor
Gibson had reported a figure from Canfield (1941) that showed the
genetic eminences of pullets and cockerel's (Figure 1). Despite a
considerable initial effort and occasional attempts distributed over
almost two decades at studying this figure [which later served as his
experimental stimuli], Biederman was unable to determine what
information could be used to distinguish males from females. His
experience was not unique: a number of other perceptual and cognitive
psychologists that we questioned had also studied that figure and all
but one (not verified before the results of this research were discussed
with him) admitted to being unable to derive a perceptual basis for the
discrimination."

Is it likely that undergraduates in a psych pool would be able to
*explicitly* recognize a diagnostic feature in these stimuli after a
single brief exposure to them, but that many interested graduate
students and professional psychologists, after years of careful
scrutiny, could NOT explicitly identify the same feature in the stimuli?
If the answer is no, then we have good reason to assert that Biederman's
study offers experimental evidence for the operation of implicit
cognitive mechanisms in chicken sexing for completely naive subjects. In
effect, then, this study would seem to straddle the transition from
implicit to explicit cognition.

It is more difficult to use the Biederman and Shiffrar article to
unravel the possible contributions of implicit and/or explicit
mechanisms in professional chicken sexers. The professional method of
training certainly works, and it also rests on a feature identification
assumption. But, again, the features it concerns itself with are far
more complex and variable than those identified by Biederman and
Shiffrar.

Unfortunately there is virtually nothing in their study addressing the
phenomenology of chicken sexing beyond the remark that all the
professionals interviewed "reported that they could listen to music
while working, [and] four of the sexers reported that they could not
'think of anything else' while sexing a chick." On general theoretical
grounds, I would expect that an activity like sexing just the most
frequent and straightforward types of genitalia would be more likely to
implicit cognitive activity, and that this would be represented in
consciousness primarily as fringe experience (Mangan, 1991, 1993a,
1993b, 2000). The less frequent and difficult cases would be more likely
handled by explicit cognitive mechanisms, represented in consciousness
by focal experience of relevant diagnostic feature(s).

But professional chicken sexers must work extremely fast (Carlson only
spent about half a second looking at each eminence) and must be
extremely accurate. There is no way to know when, in the constant flow
of chicks, a difficult case will occur. Even if a demanding judgement is
requiored only 10% of the time, it would still have to be made every few
seconds. So chicken sexing seems to demand a state of constant conscious
vigilance. This may help explain why the typical chicken sexer "can't
think of anything else."

In contrast, workers doing other well learned and highly repetitive
tasks such as rolling cigars can at the same time enjoy listening to
newspaper articles and novels read to them as they work. This suggests
an experiment. Reduce the need for vigilance and other task demands by
using a set of chicks from which all the difficult cases were already
removed by other sexers. I would expect that a chicken sexer with this
less demanding task would begin to report conscious flights of reverie,
and otherwise show signs that his or her conscious experience was now
able to represent significantly more information about non-chick matters
even when their output per minute remained at its previous level.

Finally, to come full circle: consider the question Nigel Thomas raised
about implicit cognition and chicken sexing that started our discussion
early this month:

I know I have read somewhere that people who are employed to sort new
hatched chicks into males and females typically cannot articulate the
criteria on which they do it. That is to say, they do not appear to be
conscious of how they are performing the task or of whatever the cues
are that they use to distinguish the sexes. They can just tell, as it
were. This seems to be an interesting and perhaps historically important
example of implicit cognition (I think I heard about this long before
implicit cognition was a fashionable research topic).

I, too, remember hearing this many years ago. In my memory this is
connected with two further points that I doubt I could have manufactured
out of the air. This makes it even more unlikely that what Nigel and I
heard long ago reflected North American chicken sexing practices.

As I recall, the chicken sexers who "just knew" a male chick from a
female had been trained using a mass of photographs of male and female
chick genitalia. No explicit feature analysis whatsoever. This kind of
exposure was by itself supposedly able to produce a reliable chicken
sexer. In remarkable contrast to this, Biederman notes that none of the
professional chicken sexers he interviewed had even laid eyes on a
picture of a chick's genetalia before the experiment. Apparently photos
played no part in the Japanese technique as adopted by the Americans.

The second thing I recall is that the example of "just knowing" a male
from a female baby chick was said at the time to be a favorite example
of Australian philosophers.

Could it be, then, that there was (or still is) a school of chicken
sexing (perhaps in Australia) not based at all on explicit feature
analysis, but on purely implicit training and implicit detection? In
effect, Biederman's study gives us reasonable grounds to think that
implicit mechanisms can do this job to some degree. The question is, how
far can this implicit capacity be pushed with training?

Bruce Mangan
Institute of Cognitive Studies
University of California, Berkeley
[log in to unmask]