COMMENT ⊗   VALID 00004 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.skip 5
C00009 00003	Epistemological Problems
C00014 00004	#. %3Concurrent events and actions.%1
C00016 ENDMK
C⊗;
.skip 5
Outline of epistemology

Time, events, states, persistence

Objects, creation, persistence, destruction
Material

Space, location, motion

Causality and ability

Approximations, models, and correspondences

Persons - ability, goals

Information - knowledge, belief, desires and goals.

Refinable knowledge.

Vagueness - To a child, an imaginary horse and a distant horse are very similar.
Experimental problems

	The traders buy and sell wheat by telephone.  There are at
least farmers and bakers, but perhaps also warehouse owners, wholesale
dealers and speculators.  We must represent such facts as, "If
I don't accept his offer now, he may buy what he wants from somebody
else before I can call him back".

	Missionaries and cannibals dialog with the heckler who
suggests that the boat may leak or there may be a bridge.

	More blocks.

	I need to book my flights with the travel agent, but I don't
need to know now the gate from which my plane will leave, but I
need to be able to find out in Bahrein even without speaking Arabic.

General notes

	Include a summary of concepts paper.

	The problem of refinable knowledge.

	The role of predicate calculus.

	Once one has decided to attack epistemological problems
separately from heuristic problems, the obvious candidate formalism
for representing facts is some system of mathematical logic, and
the most obvious candidate there is first order logic.  However,
merely choosing first order logic still leaves most of the choices
to be made.  The first major choice is the domain of individuals.
Next comes choosing the predicates and functions.

21 June
	1. Many of the concepts we use are imposed on the world, i.e.
ascribed to objects etc.  These include mental concepts like beliefs
purposes and abilities.  The concepts are defined only in terms
of approximate theories.  When the world is looked at in more detail,
some of the concepts become undefined and even undefinable.  Thus the
concept of a man's weight becomes undefined as soon as we start to
haggle about whether the air in his lungs is part of him and if so
how far does it  have to be exhaled before it is no longer part of
him.  Kant said that humans impose a framework (he thought Euclidean
geometry was part of it.) on the world because of human nature.
Chomsky thinks that certain aspects of language are innate and stem
from man's biological nature.  Neither Kant nor Chomsky examined the
extent to which the imposition of a framework of approximate concepts
is required for intelligent behavior and would therefore be necessary
for a Martian or a robot.

	2. Digression into philosophy
It's not a digression.

	3. We can tell whether we have solved the epistemological
problems without also solving the heuristic problems by determining
whether we can make a program that can be led to the solution.

22 June

Problems of epistemology.

	1. The relation between appearance and reality.  What does
a person know or believe about a 3-dimensional object such as another
person on the basis of what he now sees, what he previously saw,
his other senses (e.g. touch), and his general knowledge and beliefs
about humans as 3-dimensional objects?

	2. Concurrent events and actions.  The simple sequential
model recognizes situations and events. s' = result(e,s) is the
new situation that results when event e occurs in situation s.
Many laws of causality and ability can be described in terms of
this formalism, but what do we do when there are several sequences
of events occurring concurrently but at unknown relative times
even if they don't interact.

	3. The frame problem.  What are the general rules that
describe what doesn't change when an event takes place.

	4. Knowledge, belief, wanting, intending, possibility and
necessity.

	5. Objects - making out of parts, decomposition into
parts, material (quantity thereof),

	6. Causality and ability.

Epistemological Problems

	We contend that %2at least%1 the following epistemological
problems must be solved before artificial intelligence of human
level can be achieved.  That all the necessary problems have been
discovered seems unlikely, but solving these may well permit very
powerful programs.
In every case, expressing the facts is much easier than expressing
what an observer can know about the facts.
.item←0

	#. %3States of the world, events, and time%1.
When complete
information is available, it may be adequate to regard the state
of the world as a function of a real variable %2t%1.
However, in many common sense circumstances, the numerical values of the
relevant times cannot be determined accurately enough to be useful.
Fuzzy concepts like %2later%1 and %2much sooner%1 are sometimes
useful, but I don't think these are the key to an adequate formalization
of common sense time.  One key distinction is between circumstances that
can be treated as sequences of situations and those in which the
treatment of simultaneous action is essential.
In (McCarthy and Hayes 1969), a situation is treated as a complete
state of the world about which partial information exists.  When it
is necessary to treat parallel events, it may be better to use
%2partial situations%1, in which case we need a relation of one
partial situation being an extension of another.

	The notion that a situation is attached to a definite time
lacks invariance according to the theory of relativity.  Relativity
would suggest that the appropriate concept is a space-like surface,
but it isn't clear that this generalized point of view is necessary
or convenient for treating common sense problems.  Nevertheless it
would be esthetically preferable to use a formalism that wouldn't have
to be changed when physics has to be taken into account.

	From a common sense point of view, the general formalism
for situations, events and time should not presuppose any particular
theory of space.  Again physics suggests differently.

	#. %3Causality and ability%1.
In (McCarthy and Hayes 1969), we use an automaton model to
treat ability as a function of an actor's position in the world
rather than as a function of his internal structure.  This still seems
correct.

	#. %3Objects: their creation, persistence and destruction%1.
The main formal property of objects is that the same object exists
in a variety of situations both when one situation follows another
in time and when it is considered as an alternative to the other
in terms of how the world may develop or what it may be like.  Their
creation and destruction are events.

	#. %3Space, location and motion%1.  Present AI formalisms
are rather simple-minded about space.  Objects are considered to have
locations and are not considered as extended in space.  No connection
has been made between the vision formalisms that recognize the
surfaces of objects and the formalisms that consider blocks on top
of one another.
#. %3Concurrent events and actions.%1

	For most common sense purposes, it is metaphysically
adequate to regard the situation as a function of a time variable
%2t%1 taking real numbers as values.  That is, the laws of motion
of the world can be expressed in these terms as long as we don't
have to worry about quantum mechanics.

	However, such a representation may not be
epistemologically adequate.  This is especially likely in
the case of concurrent events and actions.  A person or robot
will not usually know what times to assign to different events,
and he will have complicated states of knowledge about which
events precede which others.

	Without a better understanding the problem, the best
that can be done to define it is to present the following English
sentences and ask how the information ordinarily conveyed by them
is to be represented: