(ref.doc)minsky 040495

Newsgroups: comp.lang.misc,comp.programming,comp.os.misc,comp.ai,comp.object
Subject: Re: Agent technologies (was: Sun's Hot Java product)
Date: Tue, 4 Apr 1995 16:53:23 GMT Note:
 Minsky

In _The Society of Mind_ I tried carefully to make an important
distinction when using words like "agent": the distinction between
using something 

  (1) without knowing how it works (as a "black box") and
  (2) knowing how it works.  

and I used "agent" for (1) and "agency" for (2).  A big trouble with
the field is that no one makes this distinction.  I would never use a
term like "emergent" because, it seems to me, that this word is used
mostly by people who don't realize when they're making this confusion.
That is, they say that a certain phenomenon is "emergent" when they
don't know why it appears, and don't realize that they might know if
they understood the situation better.

In any case, no one picked up on the word "agency" -- which means a
not-black box whose parts you know about. Incidentally, the
translators could not find word-pairs like "agent <--> agency" in most
other languages...

Here's the page (Section 1.6) in which these terms are introduced:

===============   ================

We want to explain intelligence as a combination of simpler things.
This means that we must be sure to check, at every step, that none of
our agents is, itself, intelligent. Otherwise, our theory would end up
resembling the nineteenth-century "chessplaying machine" that was
exposed by Edgar Allan Poe to actually conceal a human dwarf inside.
Accordingly, whenever we find that an agent has to do anything
complicated, we'll replace it with a subsociety of agents that do
simpler things. Because of this, the reader must be prepared to feel a
certain sense of loss. When we break things down to their smallest
parts, they'll each seem dry as dust at first, as though some essence
has been lost.

For example, we've seen how to construct a tower-building skill by
making Builder from little parts like Find and Get. Now, where does
its "knowing-how-to-build" reside when, clearly, it is not in any
partQand yet those parts are all that Builder is? The answer: It is
not enough to explain only what each separate agent does. We must also
understand how those parts are interrelatedQthat is, how groups of
agents can accomplish things.

Accordingly, each step in this book uses two different ways to think
about agents. If you were to watch Builder work, from the outside,
with no idea of how it works inside, you'd have the impression that it
knows how to build towers. But if you could see Builder from the
inside, you'd surely find no knowledge there. You would see nothing
more than a few switches, arranged in various ways to turn each other
on and off. Does Builder "really know" how to build towers? The answer
depends on how you look at it. Let's use two different words, "agent"
and "agency," to say why Builder seems to lead a double life. As
agency, it seems to know its job. As agent, it cannot know anything at
all.  (see 01.06)
 
When you drive a car, you regard the steering wheel as an agency that
you can use to change the car's direction. You don't care how it
works. But when something goes wrong with the steering, and you want
to understand what's happening, it's better to regard the steering
wheel as just one agent in a larger agency. it turns a shaft that
turns a gear to pull a rod that shifts the axle of a wheel. Of course,
one doesn't always want to take this microscopic view; if you kept all
those details in mind while driving, you might crash because it took
too long to figure out which way to turn the wheel. Knowing how is not
the same as knowing why. In this book, we'll always be switching
between agents and agencies because, depending on our purposes, we'll
have to use different viewpoints and kinds of descriptions.

==============   =================

The entire next chapter discusses this in more detail.  Section 2.2
might interest you:

==============   =================

NOVELISTS AND REDUCTIONISTS|

It's always best when mysteries can be explained in terms of things we
know. But when we find this hard to do, we must decide whether to keep
trying to make old theories work or to discard them and try new ones.
I think this is partly a matter of personality. Let's call
"Reductionists" those people who prefer to build on old ideas, and
"Novelists" the ones who like to champion new hypotheses.
Reductionists are usually rightQat least at science's cautious core
where novelties rarely survive for long. Outside that realm, though,
novelists reign, since older ideas have had more time to show their
flaws.

It really is amazing how certain sciences depend upon so few kinds of
explanations. The science of physics can now explain virtually
everything we see, at least in principle, in terms of how a very few
kinds of particles and force-fields interact. Over the past few
centuries reductionism has been remarkably successful. What makes it
possible to describe so much of the world in terms of so few basic
rules? No one knows.

Many scientists look on chemistry and physics as ideal models of what
psychology should be like. After all, the atoms in the brain are
subject to the same all-inclusive physical laws that govern every
other form of matter. Then can we also explain what our brains
actually do entirely in terms of those same basic principles? The
answer is no, simply because even if we understood how each of our
billions of brain cells work separately, this would not tell us how
the brain works as an agency. The "laws of thought" depend not only
upon the properties of those brain cells, but also on how they are
connected. And these connections are established not by the basic,
"general" laws of physics, but by the particular arrangements of the
millions of bits of information in our inherited genes. To be sure,
"general" laws apply to everything. But, for that very reason, they
can rarely explain anything in particular.

Does this mean that psychology must reject the laws of physics and
find its own? Of course not. It is not a matter of different laws, but
of additional kinds of theories and principles that operate at higher
levels of organization. Our ideas of how Builder works as an agency
need not, and must not, conflict with our knowledge of how Builder's
lower-level agents work. Each higher level of description must add to
our knowledge about lower levels, rather than replace it. We'll return
to the idea of "level" at many places in this book.

Will psychology ever resemble any of the sciences that have
successfully reduced their subjects to only a very few principles?
That depends on what you mean by "few." In physics, we're used to
explanations in terms of perhaps a dozen basic principles. For
psychology, our explanations will have to combine hundreds of smaller
theories. To physicists, that number may seem too large. To humanists,
it may seem too small.

============================