COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00011 00003	.skip 1
C00012 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source;
.CB TOWARDS A LOGICAL THEORY OF HUMAN IRRATIONALITY


	It is a truism that humans often behave irrationally.  The
object of this essay is to make a start on a logical theory of
irrational behavior.   The theory is concerned with the logical
possibilities and is little concerned with psychology.  I will
try to justify this approach after presenting the somewhat meagre
results.

	We will regard behavior as rational if it can be explained
as an attempt to optimize something under constraints and with
certain possibilities of acquiring information.  A commonly
postulated form of rationality optimizes some function of
the history of the sensory input of the organism over the life
of the organism.  Often the function chosen is a weighted average
of some function of the instantaneous input with the future
discounted.  Thus we may have

!!a1:	%2lifevalue = integral(0,∞,pleasure(input(t))e%5-αt%2 dt)%1.

We may call such behavior sensory rationality, and many writers
take it for granted, not considering other possibilities.  John
Rawls in his %2Theory of Justice%1 assumes something like it.

	As a first variant we may have a realist rationality
where what is optimized is a function of the history of the
world - not a function of the individuals sensory experience.
In my view, this is a better approximation to many people's
behavior.  For example, a person may devote his efforts to
advancing a goal that won't be realized until after his death.

	The idea that people do this seems to arouse a certain
amount of indignation, and this indignation gives rise to
complicated theories to the effect that the philanthropist is
not optimizing events that will occur after his death but is
really optimizing the sensations (e.g. of self-esteem) that he
will have while he is alive.  However, the philanthropist
may reject an opportunity to be hypnotized into believing that
his goals will be completely realized in favor of a mere
chance of genuinely realizing them in part.  Likewise the man
who hates another will take an opportunity to do his enemy an
injury that will be realized only after his death.

	It may be possible to reinterpret goals expressed in
terms of affecting the world as goals of affecting situations,
but even if this turns out to be possible, describing the goals
in real-world terms likely to be vastly simpler.  It would be
worthwhile to try to formulate a technical mathematical problem
formalizing the question of whether real world goals can always be
reinterpreted as sensory goals.

	What goals and motivations to humans have and what causes
them?

	One hypothesis is that human goals are biologically determined,
e.g. are combinations of the goals of eating, resting, having sex, gaining
parental approval, taking care of children, self-esteem, etc.  More
complex goals are regarded as subgoals of these biological goals, i.e.
means of realizing them.  It seems to me that this hypothesis does not
agree with observation.

	In fact people have  a great variety of goals, and these
goals arise in quite arbitrary ways.  One may make an analogy with
the phenomenon of imprinting in birds.  A newly hatched chick "imprints"
on the first moving object it sees and behaves as if that object
were its mother.  Similarly, goals are often quite accidentally
caused but acquire a life of their own independent of the circumstances
of their origin.  For example, a goal whose choice was caused
by a desire to please a person's parents may be pursued in spite
of the displeasure of the parents.

	Moreover, there are higher order goals - goals of having
goals.  Many people actively seek a "purpose in life".  This
process may be described as seeking a goal that meets certain
criteria, because candidate goals are often examined and discarded -
sometimes after trial.  It often happens that a person never finds
a "purpose in life" satisfying his criteria.

	Most likely, the behavior of most people cannot be explained
by regarding them as pursuing a "purpose in life".  On the other hand,
their short term behavior is often describable in terms of purpose.
Pat wishes to get a certain job.  His behavior is often substantially
explainable in terms of this goal, the information at his disposal,
and the obstacles he encounters in pursuing the goal and his ideas
of how to overcome them.

	Thus it seems that humans are episodically rational
but are rarely rational with respect to their whole lives as a whole.
Indeed death makes long term rationality difficult for most
people.

	Rationality itself may be regarded as a higher level goal.
Namely, many people try to establish long term purposes that they
can pursue rationally.  The theories of human rational behavior
are expressions of this idea that humans ought to be rational.

	Suppose that an individual comes to the conclusion that
he himself is not rational on the time scale of his life.
This recognized irrationality has no relation to other phenomena
called irrationality.  He need not be superstitious, and he can be
entirely objective in the sense of wanting to believe only the
truth.
.skip 1
.begin verbatim
John McCarthy
Artificial Intelligence Laboratory
Computer Science Department
Stanford University
Stanford, California 94305

ARPANET: MCCARTHY@SU-AI
.end

.turn on "{"
%7This draft of
IRRATI.ESS[S78,JMC]
PUBbed at {time} on {date}.%1