Mute inglorious leaky analogy

Our friend Google leads us to Thomas Gray, ‘Elegy in a country churchyard’:

Some village-Hampden, that with dauntless breast

The little tyrant of his fields withstood;

Some mute inglorious Milton here may rest,

Some Cromwell guiltless of his country’s blood.

And the leaky analogy derives from Edward Sapir, “all grammars leak”.Â


On Thu, Dec 7, 2017 at 11:24 AM, Martin Taylor wrote:

[Martin Taylor 2017.]

[From Bruce Abbott (2017.12.07.1025 EST)]


      Continuing my series of posts on Kenneth J.

W. Craik’s unfinished book, The Mechanism of Human Action ,
Chapter 4 was to be entitled “Brain Mechanisms: Levels.â€? The
chapter begins as follows:


      One of the ways of

approaching the mechanism of the brain is to think how we
should design a machine to do what it does, and to see how far
we are driven to the same devices as those used in the brain.Â
One typical example is the segregation into levels, both of
patterns of activity and of cerebral structure. It is
necessary in the administrative sphere that information should
be selected and classified for consideration by the highest
officers, and similarly that the commands of those officers
should relate to units of activity which are worked out by
subordinates. For instance, the C.-in-C., Fighter Command,
presumably says: ‘We want a sweep carried out over such and
such an area’; he does not have to add, ‘This means that
Spitfire No. so-and-so on such and such a station must have so
many gallons of petrol in its tanks and care must be taken
that its plugs are clean and its guns loaded.’ These latter
details are delegated to subordinates. In just the same way,
for rapidity and certainty in action, it is essential that
certain units of activity, such as looking at an object,
walking, grasping, using words, or balancing one’s body,
should be delegated to lower levels, in which they are
organized as units which can be rapidly and certainly turned
on or off at the command of these higher centres. Conversely,
on the sensory side, it is clear that it is desirable that we
should react sometimes to objects in a simple and unanalytic
way—for instance by dodging an approaching motor-carâ—but that
at other times a more closely analytical attack is requisite;
the lower levels can react in a simple way to outer objects
while leaving the higher levels free for the more analytical


      This is not a description of the levels of

perception/control found in Powers’ PCT, but it does show that
Craik was thinking along similar lines:Â The brain is
organized into levels of perception and control, with goals at
higher levels being pursued by setting the goals (references)
to be pursued by the levels below them, and with the lower
levels able to deal with simple disturbances without awaiting
orders from the higher levels.


      It would be interesting to know whether

Bill Powers was aware of Craik’s work along these lines. A
pair of Craik’s papers was published posthumously (Craik,
1947, 1948) entitled “Theory of the Human Operator in Control
Systems� (Parts 1 and 2) that describes some of the ideas that
appeared in Craik’s unfinished book, and the book itself, in
so far as it was written, appeared in 1966 as part of a book
of selected writings. During WW II, Craik was part of a group
of individuals whose work contributed to the founding of
cybernetics (including Warren McCulloch of “Pitts-McCulloch
neuron� fame), and it is possible (although perhaps unlikely)
that Bill may have learned of Craik’s work during his contact
with the cybernetics movement in the early 1950s.




This is all fascinating stuff, and it makes me think about how much

the history of science resembles high-level perceptual control.

Powers may well have encountered Craik's ideas, but not assimilated

them into his thinking directly so much as have them pave the way to
his later insights. I imagine this possibility from my own
experience. I read Bill’s 1972 Science paper when it was published
and thought well enough of it to tear the pages out and keep them in
my personal archive. But I did not then follow up his ideas at the
time. About a dozen years later, when we were looking for a
practical underpinning for multimodal interaction between a computer
and its human user (voice, vision, and keyboard), I and a colleague
(David Waugh) came up with the Layered Protocol Theory of Human
Computer interaction (soon extended to human-human dialogue).

It was not until about ten years still later that I re-encountered

PCT and Powers, having been directed to the predecessor of CSGnet by
a contributor to the Systems Analysis mailing list. As soon as I
discovered PCT, it was obvious that LPT was just a special case of
what Powers had been talking about 20 years previously (and as I
soon discovered, much longer than that). When I discovered my then
ancient pages of Bill’s Science paper, I was completely astonished
to learn that I had learned of the existence of PCT so much earlier.
How much of it had been somewhere in my deep memory and had
eventually smoothed the way for LPT?

Why does this suggest perceptual control to me? There are two ways.

One is the overall progress toward a reference level by reducing the
difference between what is and what is wanted. That’s a high-level
slow process that takes centuries and like any control process may
never result in perfection.

The other is in the reorganization process, in which components come

together to produce new perceptual functions and perceptions that
are simpler to control than is the complex set of components treated
individually, together with new tools and techniques to control them
(output processes). For example, the idea of perceptual levels goes
back at least to Donders in the mid-1800’s, and was sufficiently
current in the 1940s that it was the topic of an article in my
Childrens’ Encyclopaedia that stuck with me ever since.

Multi-level perception wasn't new to Craik. Likewise, for the output

effector processes Craik uses an analogy to a social control process
that was well known probably even at the dawn of organized warfare,
He applies the two old concepts together to describe what might be
happening inside a person – reorganization that creates a new
control unit from conscious control operations. That was probably
new, though the idea of control by humans as analogous to
servomechanisms was not.

I can't place the quote (Blake?), but a line sticks in my head,

about some stone-age person gazing at the night sky: “What mute,
inglorious Milton…”. The history of science must be full of many
such, but some of their ideas may live on in the inventions of
others. The history of science is sometimes seen as plodding
progress toward the truth (moving what is toward what is wanted),
and sometimes as a trek along forking paths with many backtracks
when the wrong path has been taken, but “All analogies leak” [From
Bruce Nevin (2017.12.03.23:00 ET)].

Fascinating. Keep these extracts coming.


(PS. My own introduction to neural networks was at a NATO seminar

meeting in 1963 that got me interested enough to build some tiny
ones on our Burroughs LGP-30 drum-based computer. Warren McCullough
was one of the stars of that meeting. I liked his description of
them as being an “anastomatic net” like the braided streams of a
river delta, but what hooked me was a weather-forecasting
application for California based on scattered temperature and
pressure readings in the Pacific a day or two earlier. I don’t
remember who did that one, but they claimed the results to be as
good as the forecasts produced by human forecasters.)