Psychological Review

Vol. 63, No. :2, 1956

RATIONAL CHOICE AND THE STRUCTURE

OF THE ENVIRONMENT*
HERBERT A. SIMON
Carnegie Institute of Technology

A growing interest in decision making in economic theory. Evidently, organ-

in psychology is evidenced by the recent isms adapt well enough to "satisflce";
publication of Edwards' review article they do not, in general, "optimize."
in the Psychological Bulletin (1) and If this is the case, a great deal can
the Santa Monica Conference volume, be learned about rational decision mak-
Decision Processes (7). In this work, ing by taking into account, at the out-
much attention has been focused on the set, the limitations upon the capacities
characterization of rational choice, and and complexity of the organism, and by
because the latter topic has been a cen- taking account of the fact that the en-
tral concern in economics, the theory of vironments to which it must adapt pos-
decision making has become a natural sess properties that permit further sim-
meeting ground for psychological and plication of its choice mechanisms. It
economic theory. may be useful, therefore, to ask: How
A comparative examination of the simple a set of choice mechanisms can
models of adaptive behavior employed we postulate and still obtain the gross
in psychology (e.g., learning theories), features of observed adaptive choice be-
and of the models of rational behavior havior?
employed in economics, shows that in In a previous paper (6) I have put
almost all respects the latter postulate forth some suggestions as to the kinds
a much greater complexity in the choice of "approximate" rationality that might
mechanisms, and a much larger capacity be employed by an organism possessing
in the organism for obtaining informa- limited information and limited compu-
tion and performing computations, than tational facilities. The suggestions were
do the former. Moreover, in the lim- "hypothetical" in that, lacking definitive
ited range of situations where the pre- knowledge of the human decisional proc-
dictions of the two theories have been esses, we can only conjecture on the ba-
compared (see [7, Ch. 9, 10, 18]), the sis of our everyday experiences, our in-
learning theories appear to account for trospection, and a very limited body of
the observed behavior rather better than psychological literature what these proc-
do the theories of rational behavior. esses are. The suggestions were in-
Both from these scanty data and from tended, however, as empirical state-
an examination of the postulates of the ments, however tentative, about some
economic models it appears probable of the actual mechanisms involved in
that, however adaptive the behavior of human and other organismic choice.2
organisms in learning and choice situa-
Now if an organism is confronted
tions, this adaptiveness falls far short
2
of the ideal of "maximizing" postulated Since writing the paper referred to I have
1
found confirmation for a number of its hy-
1 am indebted to Allen Newell for numer- potheses in the interesting and significant
ous enlightening conversations on the subject study, by A. de Groot, of the thought proc-
of this paper, and to the Ford Foundation for esses of chess players (3). I intend to discuss
a grant that permitted me leisure to com- the implications of these empirical findings for
plete it. my model in another place.
129
130 HERBERT A. SIMON

with the problem of behaving approxi- age rate and is able to store a certain
mately rationally, or adaptively, in a amount of food energy, so that it needs
particular environment, the kinds of to eat a meal at certain average inter-
simplifications that are suitable may de- vals. It has the capacity, once it sees
pend not only on the characteristics— a food heap, to proceed toward it at the
sensory, neural, and other—of the or- maximum rate of locomotion. The prob-
ganism, but equally upon the structure lem of rational choice is to choose its
of the environment. Hence, we might path in such a way that it will not
hope to discover, by a careful examina- starve.
tion of some of the fundamental struc- Now I submit that a rational way for
tural characteristics of the environment, the organism to behave is the following:
some further clues as to the nature of (a) it explores the surface at random,
the approximating mechanisms used in watching for a food heap; (b) when it
decision making. This is the line of at- sees one, it proceeds to it and eats (food
tack that will be adopted in the present getting); (c) if the total consumption
paper. of energy during the average time re-
The environment we shall discuss ini- quired, per meal, for exploration and
tially is perhaps a more appropriate one food getting is less than the energy of
for a rat than for a human. For the the food consumed in the meal, it can
term environment is ambiguous. We spend the remainder of its time in
are not interested in describing some resting.8
physically objective world in its to- There is nothing particularly remark-
tality, but only those aspects of the to- able about this description of rational
tality that have relevance as the "life choice, except that it differs so sharply
space" of the organism considered. from the more sophisticated models of
Hence, what we call the "environment" human rationality that have been pro-
will depend upon the "needs," "drives," posed by economists and others. Let
or "goals" of the organism, and upon us see what it is about the organism
its perceptual apparatus. and its environment that makes its
choice so simple.
THE ENVIRONMENT OF THE ORGANISM 1. It has only a single goal: food.
We consider first a simplified (per- It does not need to weigh the respective
haps "simple-minded") organism that advantages of different goals. It re-
has a single need—food—and is ca- quires no "utility function" or set of
pable of three kinds of activity: resting, "indifference curves" to permit it to
exploration, and food getting. The pre- choose between alternatives.
cise nature of these activities will be 2. It has no problem .of maximiza-
explained later. The organism's life tion. It needs only to maintain a cer-
space may be described as a surface tain average rate of food intake, and
over which it can locomote. Most of additional food is of no use to it. In
the surface is perfectly bare, but at iso- 8
lated, widely scattered points there are A reader who is familiar with W. Grey
Walter's mechanical turtle, Machina specu-
little heaps of food, each adequate for latrix (8), will see as we proceed that the de-
a meal. scription of our organism could well be used
The organism's vision permits it to as a set of design specifications to assure the
see, at any moment, a circular portion survival of his turtle in an environment
of the surface about the point in which sparsely provided with battery chargers. Since
I was not familiar with the structure of the
it is standing. It is able to move at turtle when I developed this model, there are
some fixed maximum rate over the sur- some differences in their behavior—but the
face. It metabolizes at a given aver- resemblance is striking.
 RATIONAL CHOICE AND ENVIRONMENT STRUCTURE 131
the psychologist's language, it has a which food is found. Let d be the aver-
definite, fixed aspiration level, and its age number of paths diverging from each
successes or failures do not change its branch point. Let » be the number of
aspirations. moves ahead the organism can see.
3. The nature of its perceptions and That is, if there is food at any of the
its environment limit sharply its plan- branch points within v moves of the
ning horizon. Since the food heaps are organism's present position, it can select
distributed randomly, there is no need the proper paths and reach it. Finally
for pattern in its searching activities. let B be the maximum number of moves
Once it sees a food heap, it can follow the organism can make between meals
a definite "best" path until it reaches it. without starving.
4. The nature of its needs and envi- At any given moment, the organism
ronment create a very natural separa- can see d branch points at a distance of
tion between "means" and "ends." Ex- one move from his present position, d2
cept for the food heaps, one point on points two moves away, and in general,
the surface is as agreeable to it as an- dh points k moves away. In all, it can
other. Locomotion has significance only d
see d + d? + h d' = -(d'-l)
as it is a means to reaching food.4 d- 1
We shall see that the first point is not points. When it chooses a branch and
essential. As long as aspirations are makes a move, d" new points become
fixed, the planning horizon is limited, visible on its horizon. Hence, in the
and there is a sharp distinction between course of m moves, md" new points
means and ends, the existence of multi- appear. Since it can make a maximum
ple goals does not create any real diffi- of H moves, and since » of these will be
culties in choice. The real complica- required to reach food that it has dis-
tions ensue only when we relax the covered on its horizon, the probability,
last three conditions; but to see clearly Q = 1 — P, that it will not survive will
what is involved, we must formulate be equal to the probability that no food
the model a little more precisely. points will be visible in (H — v) moves.
(If p is small, we can disregard the
PERCEPTUAL POWERS, STORAGE possibility that food will be visible inside
CAPACITY, AND SURVIVAL its planning horizon on the first move.)
It is convenient to describe the organ- Let p be lie probability that none of
ism's life space not as a continuous the d" new points visible at the end of a
surface, but as a branching system of particular move is a food point.
paths, like a maze, each branch point [2.1]
representing a choice point. We call Then:
the selection of a branch and locomotion
to the next branch point a "move." At 1- P
a small fraction of the branch points are [2.2]
heaps of food.
Let p, 0<p<l, be the percentage of We see that the survival chances, from
branch points, randomly distributed, at meal to meal, of this simple organism
4 depend on four parameters, two that
It is characteristic of economic models
of rationality that the distinction between describe the organism and two the en-
"means" and "ends" plays no essential role in vironment : p, the richness of the environ-
them. This distinction cannot be identified ment in food; d, the richness of the en-
with the distinction between behavior alterna-
tives and utilities, for reasons that are set
vironment in paths; H, the storage
forth at some length in the author's Adminis- capacity of the organism; and v, the
trative Behavior, Ch. 4 and S (5). range of vision of the organism.
132 HERBERT A. SIMON

To give some impression of the magni- ism metabolizes at the rate of a units
tudes involved, let us assume that p is per move, then a storage of aH food
1/10,000, (H - v) is 100, d is 10 and units, where H is given by Equation 4,
v is 3. Then the probability of seeing would be required to provide survival at
a new food point after a move is 1 — p the specified risk level, e.
= 1 - (1 - £)1M°~ 880/10,000, and the Further insight into the meaning of H
probability of survival is P = 1 — p100 can be gained by considering the average
~ 9999/10,000. Hence there is in this number of moves, M, required to dis-
case only one chance in 10,000 that the cover food. From Equation 1, the prob-
organism will fail to reach a food point ability of making (k — 1) moves without
before the end of the survival interval. discovering food, and then discovering
Suppose now that the survival time it on the k& is:
(H — v) is increased one-third, that is,
from 100 to 133. Then a similar com- Pk = (1 - P)p<*-». [2.5]
putation shows that the chance of starva- Hence, the average number of moves,
tion is reduced to less than one chance M, required to discover food is:
in 100,000. A one-third increase in v
will, of course, have an even greater
effect, reducing the chance of starvation
from one in 10~4 to one in 10~40. (1-p) 2 [2.6]
Using the same values, p = .0001, and
(H — v) = 100, we can compute the
(1 - p) (1 -
probability of survival if the organ- Since (1 — p) is the probability of dis-
ism behaves completely randomly. In covering food in any one move, M is the
this case P' = [1 - (1 - £)100] = .009. reciprocal of this probability. Combin-
From these computations, we see that ing [2.3] and [2.6], we obtain:
the organism's modest capacity to per- M logp 1
form purposive acts over a short planning [2.7]
horizon permits it to survive easily in an H-v ( l - p ) l Q g ( l-P)'
environment where random behavior Since p is close to one, loge p ^ (1 — p),
would lead to rapid extinction. A sim- and [2.7] reduces approximately to:
ple computation shows that its percep-
tual powers multiply by a factor of 880 M 1
[2.8]
the average speed with which it discovers
food.
For example, if we require (1 — P)
If p, d, and v are given, and in addition
= t < 10~4 (one chance in 10,000 of
we specify that the survival probability
must be greater than some number close starvation), then M/(H — ») < .11. For
to unity (P > 1 — e), we can compute this survival level we require food storage
approximately equal to a(v + 9M)—
from [2.23 the corresponding minimum
value of H: food enough to sustain the organism for
nine times the period required, on the
log (1 - P) = (H - t>) log p [2.3] average, to discover food, plus the period
loge required to reach the food.6
H> » logp'
[2.4] 6
1 have not discovered any very satisfactory
data on the food storage capacities of animals,
For example, if p = .95 and « = 10"10, but the order of magnitude suggested above for
then logp = - .022, log e = - 10 andthe ratio of average search time to storage capac-
ity is certainly correct. It may be noted that,
(H — v) > 455. The parameter, H, can in some cases at least, where the "food" sub-
be interpreted as the "storage capacity" stance is ubiquitous, and hence the search time
of the organism. That is, if the organ- negligible, the storage capacity is also small.
 RATIONAL CHOICE AND ENVIRONMENT STRUCTURE 133
CHOICE MECHANISMS FOR Hence, if a hunger threshold is estab-
MULTIPLE GOALS lished that leads the organism to begin
to explore /* periods after feeding, we
We consider now a more complex
will have :
organism capable of searching for and
responding to two or more kinds of goal
objects. In doing this we could intro-
duce any desired degree of complexity Hence, by making n sufficiently large,
into the choice process; but the interest- we can make X as small as we please.
ing problem is how to introduce multiple Parenthetically, it may be noted that
goals with a minimum complication of we have here a close analogue to the very
the process—that is, to construct an common two-bin system of controlling
organism capable of handling its decision industrial inventories. The primary
problems with relatively primitive choice storage, H, is a buffer stock to meet
mechanisms. demands pending the receipt of new
At the very least, the presence of two orders (with risk, I — P, of running
goals will introduce a consistency require- out); the secondary storage, n, defines
ment—the time consumed in attaining the "order point" ; and p + M is the
one goal will limit the time available for average order quantity. The storage ju
pursuit of the other. But in an environ- is fixed to balance storage "costs" against
ment like the one we have been consider- the cost (in this case, time pressure) of
ing, there need be no further relationship too frequent reordering.
between the two goals. In our original If food and the second goal object
formulation, the only essential stipula- (water, let us say) are randomly and
tion was that H, the storage capacity, be independently distributed, then there are
adequate to maintain the risk of starva- no important complications resulting
tion below a stipulated level (1 — P). from interference between the two activ-
Now we introduce the additional stipula- ities. Designate by the subscript 1 the
tion that the organism should only de- variables and parameters referring to
vote a fraction, X, of its time to food- food getting (e.g., MI is the food threshold
seeking activities, leaving the remaining in periods), and by the subscript 2 the
fraction, 1 — X, to other activities. This quantities referring to water seeking.
new stipulation leads to a requirement The organism will have adequate time
of additional storage capacity. for both activities if Xi + X2 < 1.
In order to control the risk of starving, Now when the organism reaches either
the organism must begin its exploration its hunger or thirst threshold, it will
for food whenever it has reached a level begin exploration. We assume that if
of H periods of food storage. If it has either of the goal objects becomes visible,
a total storage of (M + H) periods of it will proceed to that object and satisfy
food, and if the food heaps are at least its hunger or thirst (this will not increase
a(M + -ff) in size, then it need begin the the number of moves required, on the
search for food only M periods after its average, to reach the other object) ; but
last feeding. But the food research will if both objects become visible at the
require, on the average, M periods. same time, and if Si and S% are the
respective quantities remaining in stor-
Thus, in terrestrial animals there is little oxygen age at this time, then it will proceed to
storage and life can be maintained in the absence food or water as Mi/Si is greater or less
of air for only a few minutes. I am not arguing than Mz/S}. This choice will maximize
as to which way the causal arrow runs, but only
that the organisms, in this respect, are adapted its survival probability. What is re-
to their environments and do not provide storage quired, then, is a mechanism that pro-
that is superfluous. duces a drivejproportional to Mi/Si.
134 HERBERT A. SIMON

A priority mechanism of the kind just In particular, if p is close to one, that

described is by no means the only or is, if need-satisfying points are rare, we
simplest one that can be postulated. will have:
An even simpler rule is for the organism
to persist in searching for points that M „—
will satisfy the particular need that first (1-
reached its threshold and initiated ex-
ploratory behavior. This is not usually (1 - P) v
an efficient procedure, from the stand-
point of conserving goal-reaching time, and
but it may be entirely adequate for an
[3.5]
organism generously endowed with stor-
age capacity.
We see that an organism can satisfy Now substituting particular values for
a number of distinct needs without re- n in [3.5] we get: Mt — 3/2 M\;
quiring a very elaborate mechanism for Ms = 11/6 My, Mt = 25/12 Mi, etc.
choosing among them. In particular, We see that if the organism has two
we do not have to postulate a utility separate needs, its exploration time will
function or a "marginal rate of substitu- be only 50 per cent greater than—and
tion." not twice as great as—if it has only one
We can go even further, and assert need ; for four needs the exploration time
that a primitive choice mechanism is will be only slightly more than twice as
adequate to take advantage of important great as for a single need, and so on.
economies, if they exist, which are deriv- A little consideration of the program just
able from the interdependence of the described will show that the joint ex-
activities involved in satisfying the dif- ploratory process does not reduce the
ferent needs. For suppose the organism primary storage capacity required by the
has n needs, and that points at which he organism but does reduce the secondary
can satisfy each are distributed randomly storage capacity required. As a matter
and independently through the environ- of fact, there would be no necessity at
ment, each with the same probability, p. all for secondary storage.
Then the probability that no points This conclusion holds only if the need-
satisfying any of the needs will be visible satisfying points are independently dis-
on a particular move is />", and the mean tributed. If there is a negative correla-
number of moves for discovery of the tion in the joint distribution of points
first need-satisfying point is: satisfying different needs, then it may
be economical for the organism to pursue
'"•n — /« [3.2] its needs separately, and hence to have
ny
U P ) a simple signaling mechanism, involving
Suppose that the organism begins to secondary storage, to trigger its several
explore, moves to the first need-satisfying exploration drives. This point will be
point it discovers, resumes its explora- developed further in the next section.
tion, moves to the next point it discovers A word may be said here about
that satisfies a need other than the one "avoidance needs." Suppose that cer-
already satisfied, and so on. Then the tain points in the organism's behavior
mean time required to search for all n space are designated as "dangerous."
goals will be: Then it will need to avoid those paths
that lead to these particular points. If
M „ = mn + + r per cent of all points, randomly dis-
1 tributed, are dangerous, then the number
of available paths, among those visible
 RATIONAL CHOICE AND ENVIRONMENT STRUCTURE 135
at a given move, will be reduced to with the highest probability. If only
(1 - r)d\ Hence, p' = (1 - pY1-^ certain choice points are provided with
will be smaller than p (Equation 2.1), such clues, then a combination of ran-
and M (Equation 2.6) will be correspond- dom and systematic exploration can be
ingly larger. Hence, the presence of employed. Thus the organism may be
danger points simply increases the aver- led into "regions" where the probability
age exploration time and, consequently, of goal attainment is high relative to
the required storage capacity of the other regions, but it may have to ex-
organism. plore randomly for food within a given
region.
FURTHER SPECIFICATION OF THE A concrete example of such behavior
ENVIRONMENT: CLUES in humans is the "position play" char-
In our discussion up to the present acteristic of the first phase of a chess
point, the range of the organism's an- game. The player chooses moves on
ticipations of the future has been lim- the basis of certain characteristics of re-
ited by the number of behavior alterna- sulting positions (e.g., the extent to
tives available to it at each move (d), which his pieces are developed). Cer-
and the length of the "vision" ( v ) . It tain positions are adjudged richer in at-
is a simple matter to introduce into the tacking and defensive possibilities than
model the consequences of several types others, but the original choice may in-
of learning. An increase in the reper- volve no definite plan for the subsequent
toire of behavior alternatives or in the action after the "good" position has
length of vision can simply be repre- been reached.
sented by changes in d and v, respec- Next, we turn to the problem of
tively. choice that arises when those regions of
A more interesting possibility arises if the behavior space that are rich in
the food points are not distributed com- points satisfying one need (pi is high
pletely at random, and if there are clues in these regions) are poor in points
that indicate whether a particular inter- satisfying another need (p% is low in
mediate point is rich or poor in paths these same regions). In the earlier case
leading to food points. First, let us of goal conflict (two or more points
suppose that on the path leading up to simultaneously visible mediating differ-
each food point the k preceding choice ent needs), we postulated a priority
points are marked with a food clue. mechanism that amounted to a mecha-
Once the association between the clue nism for computing relative need in-
and the subsequent appearance of the tensity and for responding to the more
food point is learned by the organism, intense need. In the environment with
its exploration can terminate with the clues, the learning process would need
discovery of the clue, and it can follow to include a conditioning mechanism
a determinate path from that point on. that would attach the priority mecha-
This amounts to substituting v' = (v nism to the response to competing clues,
+ k) for v. as well as to the response to competing
A different kind of clue might operate visible needs.
in the following fashion. Each choice Finally, we have thus far specified
point has a distinguishable characteristic the environment in such a way that
that is associated with the probability there is only one path leading to each
of encountering a food point if a path is point. Formally, this condition can al-
selected at random leading out of this ways be satisfied by representing as two
choice point. The organism can then or more points any point that can be
select at each move the choice point reached by multiple paths. For $ome
136 HERBERT A. SIMON

purposes, it might be preferable to satisfaction of any particular need is

specify an environment in which paths not critical.
converge as well as diverge. This can We have introduced other assump-
be done without disturbing the really tions that represent characteristics of
essential conditions of the foregoing the environment, the most important
analysis. For behavior of the sort we being that need satisfaction can take
have been describing, we require of the place only at "rare" points which (with
environment only: some qualifications we have indicated)
1. that if a path is selected com- are distributed randomly.
pletely at random the probability of The most important conclusion we
survival is negligible; have reached is that blocks of the or-
2. that there exist clues in the envi- ganism's time can be allocated to activi-
ronment (either the actual visibility of ties related to individual needs (sepa-
need-satisfying points or anticipatory rate means-end chains) without creat-
clues) which permit the organism, suffi- ing any problem of over-all allocation
ciently frequently for survival, to select or coordination or the need for any gen-
specific paths that lead with certainty, eral "utility function." The only scarce
or with very high probability, to a need- resource in the situation is time, and its
satisfying point. scarcity, measured by the proportion of
the total time that the organism will
CONCLUDING COMMENTS ON need to be engaged in some activity,
MULTIPLE GOALS can be reduced by the provision of
generous storage capacity.
The central problem of this paper has This does not mean that a more effi-
been to construct a simple mechanism cient procedure cannot be constructed,
of choice that would suffice for the be- from the standpoint of the total time
havior of an organism confronted with required to meet the organism's needs.
multiple goals. Since the organism, like We have already explored some simple
those of the real world, has neither the possibilities for increasing efficiency by
senses nor the wits to discover an recognizing complementarities among ac-
"optimal" path—even assuming the con- tivities (particularly the exploration ac-
cept of optimal to be. clearly denned— tivity). But the point is that these
we are concerned only with finding a complications are not essential to the
choice mechanism that will lead it to survival of an organism. Moreover, if
pursue, a "satisficing" path, a path that the environment is so constructed (as
will permit satisfaction at some speci- it often is in fact) that regions rich in
fied level of all of its needs. possibilities for one kind of need satis-
Certain of the assumptions we have faction are poor in possibilities for other
introduced to make this possible repre- satisfactions, such efficiencies may not
sent characteristics of the organism, be available.
(a) It is able to plan short purposive It may be objected that even rela-
behavior sequences (of length not ex- tively simple organisms appear to con-
ceeding v), but not long sequences, (b) form to efficiency criteria in their be-
Its needs are not insatiable, and hence havior, and hence that their choice
it does not need to balance marginal in- mechanisms are much more elaborate
crements of satisfaction. If all its needs than those we have described. A rat,
are satisfied, it simply becomes inactive. for example, learns to take shorter
(c) It possesses sufficient storage ca- rather than longer paths to food. But
pacity so that the exact moment of this observation does not affect the cen-
 RATIONAL CHOICE AND ENVIRONMENT STRUCTURE 137

tral argument. We can introduce a organisms, elaborate mechanisms for

mechanism that leads the organism to choosing among diverse needs. Com-
choose time-conserving paths, where mon denominators among needs may
multiple paths are available for satis- simply not exist, or may exist only in
fying a given need, without any assump- very rudimentary form; and the na-
tion of a mechanism that allocates time ture of the organism's needs in relation
among different needs. The former to the environment may make their
mechanism simply increases the "slack" nonexistence entirely tolerable.
in the whole system, and makes it even There is some positive evidence bear-
more feasible to ignore the comple- ing on this point in the work that has
mentarities among activities in program- been done on conflict and frustration.
ming the over-all behavior of the or- A common method of producing con-
ganism. flict in the laboratory is to place the
This is not the place to discuss at organism in a situation where: (a) it is
length the application of the model to stimulated to address itself simultane-
human behavior, but a few general ously to alternative goal-oriented be-
statements may be in order. First, the haviors, or (b) it is stimulated to a
analysis has been a static one, in the goal-oriented behavior, but restricted
sense that we have taken the organ- from carrying out the behaviors it usu-
ism's needs and its sensing and plan- ally evinces in similar natural situa-
ning capacities as given. Except for a tions. This suggests that conflict may
few comments, we have not considered arise (at least in a large class of situa-
how the organism develops needs or tions) from presenting the animal with
learns to meet them. One would con- situations with which it is not "pro-
jecture, from general observation and grammed" to deal. Conflict of choice
from experimentation with aspiration may often be equivalent to an absence
levels, that in humans the balance be- of a choice mechanism in the given
tween the time required to meet needs situation. And while it may be easy to
and the total time available is main- create such situations in the labora-
tained by the raising and lowering of tory, the absence of a mechanism to
aspiration levels. I have commented deal with them may simply reflect the
on this point at greater length in my fact that the organism seldom encoun-
previous paper.8 ters equivalent situations in its natural
Second, there is nothing about the environment.7
model that implies that the needs are
physiological and innate rather than CONCLUSION
sociological and acquired. Provided that
the needs of the organism can be speci- In this paper I have attempted to
fied at any given time in terms of the identify some of the structural charac-
aspiration levels for the various kinds teristics that are typical of the "psy-
of consummatory behavior, the model chological" environments of organisms.
can be applied. We have seen that an organism in an
environment with these characteristics
The principal positive implication of
requires only very simple perceptual and
the model is that we should be skeptical
choice mechanisms to satisfy its several
in postulating for humans, or other
needs and to assure a high probability
8
See (6, pp. Ill, 117-18). For an experi- of its survival over extended periods of
ment demonstrating the adjustment of the
7
rat's aspiration levels to considerations of re- See, for example, Neal E. Miller, "Experi-
alizability, see Festinger (2). mental Studies of Conflict" in (4, Ch. 14).
138 HERBERT A. SIMON

time. In particular, no "utility func- REFERENCES

tion" needs to be postulated for the or- 1. EDWARDS, W. The theory of decision mak-
ganism, nor does it require any elabo- ing. Psychol, Bull., 1954, 51, 380-417.
rate procedure for calculating marginal 2. FESTINGER, L. Development of differential
appetite in the rat. /. exp. Psychol.,
rates of substitution among different 19S3, 32, 226-234.
wants. 3. DE GROOT, A. flet Denken van den Schaker.
The analysis set forth here casts seri- Amsterdam: Noord-Hollandsche Uit-
ous doubt on the usefulness of current gevers Maatschapij, 1946,
4. HUNT, J. McV. Personality and the be-
economic and statistical theories of ra- havior disorders. New York: Ronald,
tional behavior as bases for explaining 1944.
the characteristics of human and other 5. SIMON, H. A. Administrative behavior.
New York: Macmillan, 1947.
organismic rationality. It suggests an 6. SIMON, H. A. A behavioral model of ra-
alternative approach to the description tional choice. Quart. J. Econ., 1955,
of rational behavior that is more closely 59, 99-118.
related to psychological theories of per- 7. THRALL, R. M., COOMBS, C. H., & DAVIS,
R. L. (Eds.). Decision processes. New
ception and cognition, and that is in York: Wiley, 1954.
closer agreement with the facts of be- 8. WALTER, W. G. The living brain. New
havior as observed in laboratory and York: Norton, 1953.
field. (Received March 28, 1955)