Blocks world

[From Bill Powers (920813.1400)]

Oded Maler (920810) --

A couple of points concerning your "unfocussed essay."

In defining the formal or abstract approach to AI and AL you give two
examples:

1) An algorithm, that given a "goal" (in this setting, a representation
of some configuration of the "environment") produces a sequence of
"actions" that leads from any state to this state.

2) A set of simple local dynamical rules, that when employed by many
similar abstract entities leads to an interesting emergent global
"behavior" (=trajectories in the state-space).

Although you say that there is "some environment which is also an abstract
object with its own formal dynamics," I don't see in either example any
acknowledgement that this environment may include unpredictable
disturbances. I know that this isn't supposed to be the real environment,
but even in a formal environment, all the essential properties of a real
one have to be included if the model is ever to be applied to a real
situation.

In (1), the implied model is one that computes the output required to
achieve a result in the environment. If the environment contains any
unpredictable disturbances that also contribute to the result, there is no
"sequence of actions" that will produce a preselected result.

In (2), the same problem exists: there can be no regular state-space
trajectories that result from application of a fixed (action) rule unless
there are no unpredictable disturbances capable of altering those
trajectories directly.

When I speak of disturbances, many people automatically think in terms of
adding noise to the input or the output. But the disturbances of which I
speak can be, for example, large sustained forces that suddenly appear and
persist, that act directly on the outcome independently of action or
perception -- for example, a 2-kilogram load suddenly attached to an arm
and left there.

Also, there is a tendency to assume that all disturbances are low-level:

In other words, the real world is not the Blocks-world. On the other
hand, we feel from introspection that *some* of our mental activity is
done in this (graph-searching, if-then-else) mode, and *if* the lower-
level systems take responsibility for all the messy details,
uncertainties, and disturbance, it can transduce the real-world into a
Blocks-world, and then a higher-level system can operate at this >abstract

level...

Even if the mechanics of behavior are taken care of by lower-level control
systems, higher-level perceptions are NOT immune to disturbances. If a
cognitive program concludes that the best strategy is to buy 100 shares of
IBM at 132-7/8, an open-loop model will simply assume they have been bought
and go on from there. In real life, such an order may easily result in
buying 0 shares of IBM -- perhaps nobody will sell at that price, perhaps
the telephone is out of order, or perhaps your broker informs you that your
credit is no good. These aren't mechanical disturbances, but they cause
errors in a high-level control system just the same.

It would be possible to program AI or AL programs that let the real world
into the loop. These programs would have to check constantly to see whether
any commanded act has in fact been carried out. They would have to sense
the ACTUAL outcome, compare it with the intended outcome, and treat the
difference as the basis for a change in the commands to act. If AI and AL
programs were organized like that I would have no more complaints.

Even if it were not true that "the planning problem in such a world is
computationally intractable" the presence of unpredictable disturbances
would render any plan obsolete before it could be carried out. I don't see
the primary difficulty with AI and AL models as being the complexity of the
real world. It's the failure to find an organization that can deal with
disturbances without having to know about them in advance.

I've been attracted to the bottom-up approach (e.g., Powers, Brooks)
hoping that some principles discovered while going up in the levels >will

explain how, after all, it is possible to plan and solve problems >in the
real-world in the higher level.

I think there's an unwarranted assumption here, which is that planning and
solving problems in the abstract actually works. I think that plans and
abstract problem-solving very seldom work -- they can't deal with changes
in the world that are unexpected or unobserved, and they always leave out
critical details. The best that a plan can do is sketch in roughly what
might be done assuming nothing surprising happens, and assuming that all
necessary information is available. Most plans require emergency
modification as soon as they are put into effect. Abstract problem-solving
(as in the Prisoner's Dilemma game) usually yields a result only when the
game remains abstract and everyone agrees to play by exactly the assumed
rules. Try to play it in real life and somebody will cheat (I would confess
to the crime in a way that involves a falsehood certain to be discovered).

There are exceptions to the general failure of plans and problem-solving to
work. Those are the cases where the plan or solution entails setting up a
closed-loop system that takes no outcome for granted, but relies on
constant monitoring of results. This is not the approach generally taken.

ยทยทยท

------------------------------------------------------------------
Rick Marken (9208120 --

Just saying "CONTROL WON'T WORK (in the long run) precisely because people
ARE controllers" is enough to convince anyone who already understands the
principles of control. To others it just sounds like a slogan. When we talk
with people who don't already understand the principles, we have to drop
down a level and talk about the REASONS for which control means that
controlling people won't work. And so on. That's why I keep harping on
disturbances -- they're the main reason that control is necessary, and the
main reason why other non-feedback models can't work in the real world. I
don't think it's much use to keep saying that control is the answer, when
the question to which it's the answer isn't perfectly clear.
-----------------------------------------------------------------------
Tomorrow I go and get my new 486-33 with supervga screen. What the hell, I
won't be eating much when I'm 85 anyway.
-----------------------------------------------------------------------
Best to all,

Bill P.