A PROPOSAL FOR THE
DARTMOUTH SUMMER RESEARCH PROJECT
ON ARTIFICIAL INTELLIGENCE
J. McCarthy, Dartmouth College
M. L. Minsky, Harvard University
N. Ro chester, I.B.M. Corp oration
C.E. Shannon, Bell Telephone Lab oratories
August 31, 1955
1

We prop ose that a 2 month, 10 man study of articial intelligence b e carried out during the
summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to pro ceed on
the basis of the conjecture that every asp ect of learning or any other feature of intelligence can in
principle b e so precisely describ ed that a machine can b e made to simulate it. An attempt will b e
made to nd how to make machines use language, form abstractions and concepts, solve kinds of
problems now reserved for humans, and improve themselves. We think that a signicant advance
can b e made in one or more of these problems if a carefully selected group of scientists work on it
together for a summer.
The following are some asp ects of the articial intelligence problem:
1 Automatic Computers
If a machine can do a job, then an automatic calculator can b e programmed to simulate the
machine. The sp eeds and memory capacities of present computers may b e insucient to simulate
many of the higher functions of the human brain, but the ma jor obstacle is not lack of machine
capacity, but our inability to write programs taking full advantage of what we have.
2. How Can a Computer b e Programmed to Use a Language
It may b e sp eculated that a large part of human thought consists of manipulating words accord-
ing to rules of reasoning and rules of conjecture. From this p oint of view, forming a generalization
consists of admitting a new word and some rules whereby sentences containing it imply and are
implied by others. This idea has never b een very precisely formulated nor have examples b een
worked out.
3. Neuron Nets
How can a set of (hyp othetical) neurons b e arranged so as to form concepts. Considerable
theoretical and exp erimental work has b een done on this problem by Uttley, Rashevsky and his
group, Farley and Clark, Pitts and McCullo ch, Minsky, Ro chester and Holland, and others. Partial
results have b een obtained but the problem needs more theoretical work.
4. Theory of the Size of a Calculation
If we are given a well-dened problem (one for which it is p ossible to test mechanically whether
or not a prop osed answer is a valid answer) one way of solving it is to try all p ossible answers in
order. This metho d is inecient, and to exclude it one must have some criterion for eciency of
calculation. Some consideration will show that to get a measure of the eciency of a calculation
it is necessary to have on hand a metho d of measuring the complexity of calculating devices which
in turn can b e done if one has a theory of the complexity of functions. Some partial results on this
problem have b een obtained by Shannon, and also by McCarthy.
5. Self-lmprovement
Probably a truly intelligent machine will carry out activities which may b est b e describ ed as
self-improvement. Some schemes for doing this have b een prop osed and are worth further study.
It seems likely that this question can b e studied abstractly as well.
6. Abstractions
A numb er of typ es of \abstraction" can b e distinctly dened and several others less distinctly.
A direct attempt to classify these and to describ e machine metho ds of forming abstractions from
sensory and other data would seem worthwhile.
7. Randomness and Creativity
A fairly attractive and yet clearly incomplete conjecture is that the dierence b etween creative
thinking and unimaginative comp etent thinking lies in the injection of a some randomness. The
randomness must b e guided by intuition to b e ecient. In other words, the educated guess or the
hunch include controlled randomness in otherwise orderly thinking.
2

In addition to the ab ove collectively formulated problems for study, we have asked the indi-
viduals taking part to describ e what they will work on. Statements by the four originators of the
pro ject are attached.
We prop ose to organize the work of the group as follows.
Potential participants will b e sent copies of this prop osal and asked if they would like to work
on the articial intelligence problem in the group and if so what they would like to work on. The
invitations will b e made by the organizing committee on the basis of its estimate of the individual's
p otential contribution to the work of the group. The memb ers will circulate their previous work
and their ideas for the problems to b e attacked during the months preceding the working p erio d of
the group.
During the meeting there will b e regular research seminars and opp ortunity for the memb ers
to work individually and in informal small groups.
The originators of this prop osal are:
1. C. E. Shannon , Mathematician, Bell Telephone Lab oratories. Shannon develop ed the
statistical theory of information, the application of prop ositional calculus to switching circuits,
and has results on the ecient synthesis of switching circuits, the design of machines that learn,
cryptography, and the theory of Turing machines. He and J. McCarthy are co-editing an Annals
of Mathematics Study on \The Theory of Automata" .
2. M. L. Minsky , Harvard Junior Fellow in Mathematics and Neurology. Minsky has built
a machine for simulating learning by nerve nets and has written a Princeton PhD thesis in math-
ematics entitled, \Neural Nets and the Brain Mo del Problem" which includes results in learning
theory and the theory of random neural nets.
3. N. Ro chester , Manager of Information Research, IBM Corp oration, Poughkeepsie, New
York. Ro chester was concerned with the development of radar for seven years and computing
machinery for seven years. He and another engineer were jointly resp onsible for the design of the
IBM Typ e 701 which is a large scale automatic computer in wide use to day. He worked out some of
the automatic programming techniques which are in wide use to day and has b een concerned with
problems of how to get machines to do tasks which previously could b e done only by p eople. He
has also worked on simulation of nerve nets with particular emphasis on using computers to test
theories in neurophysiology,
4. J. McCarthy , Assistant Professor of Mathematics, Dartmouth College. McCarthy has
worked on a numb er of questions connected with the mathematical nature of the thought pro cess
including the theory of Turing machines, the sp eed of computers, the relation of a brain mo del to
its environment, and the use of languages by machines. Some results of this work are included in
the forthcoming \Annals Study" edited by Shannon and McCarthy. McCarthy's other work has
b een in the eld of dierential equations.
The Ro ckefeller Foundation is b eing asked to provide nancial supp ort for the pro ject on the
following basis:
1. Salaries of $1200 for each faculty level participant who is not b eing supp orted by his own
organization. It is exp ected, for example, that the participants from Bell Lab oratories and IBM
Corp oration will b e supp orted by these organizations while those from Dartmouth and Harvard
will require foundation supp ort.
2. Salaries of $700 for up to two graduate students.
3. Railway fare for participants coming from a distance.
4. Rent for p eople who are simultaneously renting elsewhere.
5. Secretarial exp enses of $650, $500 for a secretary and $150 for duplicating exp enses.
3

6. Organization exp enses of $200. (Includes exp ense of repro ducing preliminary work by
participants and travel necessary for organization purp oses.
7. Exp enses for two or three p eople visiting for a short time.
Estimated Exp enses
6 salaries of 1200 $7200
2 salaries of 700 1400
8 traveling and rent exp enses averaging 300 2400
Secretarial and organizational exp ense 850
Additional traveling exp enses 600
Contingencies 550
||{
$13,500
4

PROPOSAL FOR RESEARCH BY C. E. SHANNON
I would like to devote my research to one or b oth of the topics listed b elow. While I hop e to
do so, it is p ossible that b ecause of p ersonal considerations I may not b e able to attend for the
entire two months. I, nevertheless, intend to b e there for whatever time is p ossible,
1. Application of information theory concepts to computing machines and brain mo dels.
A basic problem in information theory is that of transmitting information reliably over a noisy
channel. An analogous problem in computing machines is that of reliable computing using unreliable
elements. This problem has b een studies by von Neumann for Sheer stroke elements and by
Shannon and Mo ore for relays; but there are still many op en questions. The problem for several
elements, the development of concepts similar to channel capacity, the sharp er analysis of upp er
and lower b ounds on the required redundancy, etc. are among the imp ortant issues. Another
question deals with the theory of information networks where information ows in many closed
lo ops (as contrasted with the simple one-way channel usually considered in communication theory).
Questions of delay b ecome very imp ortant in the closed lo op case, and a whole new approach seems
necessary. This would probably involve concepts such as partial entropies when a part of the past
history of a message ensemble is known.
2. The matched environment - brain mo del approach to automata. In general a machine or
animal can only adapt to or op erate in a limited class of environments. Even the complex human
brain rst adapts to the simpler asp ects of its environment, and gradually builds up to the more
complex features. I prop ose to study the synthesis of brain mo dels by the parallel development
of a series of matched (theoretical) environments and corresp onding brain mo dels which adapt to
them. The emphasis here is on clarifying the environmental mo del, and representing it as a math-
ematical structure. Often in discussing mechanized intelligence, we think of machines p erforming
the most advanced human thought activities{proving theorems, writing music, or playing chess.
I am prop osing here to start at the simple and when the environment is neither hostile (merely
indierent) nor complex, and to work up through a series of easy stages in the direction of these
advanced activities.
5

PROPOSAL FOR RESEARCH BY M. L. MINSKY
It is not dicult to design a machine which exhibits the following typ e of learning. The
machine is provided with input and output channels and an internal means of providing varied
output resp onses to inputs in such a way that the machine may b e rained" by a rial and
error" pro cess to acquire one of a range of input-output functions. Such a machine, when placed
in an appropriate environment and given a criterior of \success" or ailure" can b e trained to
exhibit \goal-seeking" b ehavior. Unless the machine is provided with, or is able to develop, a way
of abstracting sensory material, it can progress through a complicated environment only through
painfully slow steps, and in general will not reach a high level of b ehavior.
Now let the criterion of success b e not merely the app earance of a desired activity pattern
at the output channel of the machine, but rather the p erformance of a given manipulation in a
given environment. Then in certain ways the motor situation app ears to b e a dual of the sensory
situation, and progress can b e reasonably fast only if the machine is equally capable of assembling
an ensemble of \motor abstractions" relating its output activity to changes in the environment.
Such \motor abstractions" can b e valuable only if they relate to changes in the environment which
can b e detected by the machine as changes in the sensory situation, i.e., if they are related, through
the structure of the environrnent, to the sensory abstractions that the machine is using.
I have b een studying such systems for some time and feel that if a machine can b e designed
in which the sensory and motor abstractions, as they are formed, can b e made to satisfy certain
relations, a high order of b ehavior may result. These relations involve pairing, motor abstractions
with sensory abstractions in such a way as to pro duce new sensory situations representing the
changes in the environment that might b e exp ected if the corresp onding motor act actually to ok
place.
The imp ortant result that would b e lo oked for would b e that the machine would tend to build
up within itself an abstract mo del of the environment in which it is placed. If it were given a
problem, it could rst explore solutions within the internal abstract mo del of the environment
and then attempt external exp eriments. Because of this preliminary internal study, these external
exp eriments would app ear to b e rather clever, and the b ehavior would have to b e regarded as rather
\imaginative"
A very tentative prop osal of how this might b e done is describ ed in my dissertation and I
intend to do further work in this direction. I hop e that by summer 1956 I wi11 have a mo del of
such a machine fairly close to the stage of programming in a computer.
6

PROPOSAL FOR RESEARCH BY N. ROCHESTER
Originality in Machine Performance
In writing a program for an automatic calculator, one ordinarily provides the machine with
a set of rules to cover each contingency which may arise and confront the machine. One exp ects
the machine to follow this set of rules slavishly and to exhibit no originality or common sense.
Furthermore one is annoyed only at himself when the machine gets confused b ecause the rules he
has provided for the machine are slightly contradictory. Finally, in writing programs for machines,
one sometimes must go at problems in a very lab orious manner whereas, if the machine had just
a little intuition or could make reasonable guesses, the solution of the problem could b e quite
direct. This pap er describ es a conjecture as to how to make a machine b ehave in a somewhat more
sophisticated manner in the general area suggested ab ove. The pap er discusses a problem on which
I have b een working sp oradically for ab out ve years and which I wish to pursue further in the
Articial Intelligence Pro ject next summer.
The Pro cess of Invention or Discovery
Living in the environment of our culture provides us with pro cedures for solving many problems.
Just how these pro cedures work is not yet clear but I shall discuss this asp ect of the problem
in terms of a mo del suggested by Craik
1
. He suggests that mental action consists basically of
constructing little engines inside the brain which can simulate and thus predict abstractions relating
to environment. Thus the solution of a problem which one already understands is done as follows:
1. The environment provides data from which certain abstractions are formed.
2. The abstractions together with certain internal habits or drives provide:
2.1 A denition of a problem in terms of desired condition to b e achieved in the future, a goal.
2.2 A suggested action to solve the problem.
2.3 Stimulation to arouse in the brain the engine which corresp onds to this situation.
3. Then the engine op erates to predict what this environmental situation and the prop osed reac-
tion will lead to.
4. If the prediction corresp onds to the goal the individual pro ceeds to act as indicated.
The prediction will corresp ond to the goal if living in the environment of his culture has
provided the individual with the solution to the problem. Regarding the individual as a stored
program calculator, the program contains rules to cover this particular contingency.
For a more complex situation the rules might b e more complicated. The rules might call
for testing each of a set of p ossible actions to determine which provided the solution. A still
more complex set of rules might provide for uncertainty ab out the environment, as for example in
playing tic tac to e one must not only consider his next move but the various p ossible moves of the
environment (his opp onent).
Now consider a problem for which no individual in the culture has a solution and which has
resisted eorts at solution. This might b e a typical current unsolved scientic problem. The
individual might try to solve it and nd that every reasonable action led to failure. In other words
the stored program contains rules for the solution of this problem but the rules are slightly wrong.
In order to solve this problem the individual will have to do something which is unreasonable
or unexp ected as judged by the heritage of wisdom accumulated by the culture. He could get
such b ehavior by trying dierent things at random but such an approach would usually b e to o
inecient. There are usually to o many p ossible courses of action of which only a tiny fraction are
acceptable. The individual needs a hunch, something unexp ected but not altogether reasonable.
Some problems, often those which are fairly new and have not resisted much eort, need just a little
7

randomness. Others, often those which have long resisted solution, need a really bizarre deviation
from traditional metho ds. A problem whose solution requires originality could yield to a metho d
of solution which involved randomness.
In terms of Craik's
1
S mo del, the engine which should simulate the environment at rst fails
to simulate correctly. Therefore, it is necessary to try various mo dications of the engine until one
is found that makes it do what is needed.
Instead of describing the problem in terms of an individual in his culture it could have b een
describ ed in terms of the learning of an immature individual. When the individual is presented
with a problem outside the scop e of his exp erience he must surmount it in a similar manner.
So far the nearest practical approach using this metho d in machine solution of problems is
an extension of the Monte Carlo metho d. In the usual problem which is appropriate for Monte
Carlo there is a situation which is grossly misundersto o d and which has to o many p ossible factors
and one is unable to decide which factors to ignore in working out analytical solution. So the
mathematician has the machine making a few thousand random exp eriments. The results of these
exp eriments provide a rough guess as to what the answer may b e. The extension of the Monte
Carlo Metho d is to use these results as a guide to determine what to neglect in order to simplify
the problem enough to obtain an approximate analytical solution.
It might b e asked why the metho d should include randomness. Why shouldn't the metho d
b e to try each p ossibility in the order of the probability that the present state of knowledge would
predict for its success? For the scientist surrounded by the environment provided by his culture, it
may b e that one scientist alone would b e unlikely to solve the problem in his life so the eorts of
many are needed. If they use randomness they could all work at once on it without complete du-
plication of eort. If they used system they would require imp ossibly detailed communication. For
the individual maturing in comp etition with other individuals the requirements of mixed strategy
(using game theory terminology) favor randomness. For the machine, randomness will probably b e
needed to overcome the shortsightedness and prejudices of the programmer. While the necessity
for randomness has clearly not b een proven, there is much evidence in its favor.
The Machine With Randomness
In order to write a program to make an automatic calculator use originality it will not do to
intro duce randomness without using forsight. If, for example, one wrote a program so that once in
every 10,000 steps the calculator generated a random numb er and executed it as an instruction the
result would probably b e chaos. Then after a certain amount of chaos the machine would probably
try something forbidden or execute a stop instruction and the exp eriment would b e over.
Two approaches, however, app ear to b e reasonable. One of these is to nd how the brain
manages to do this sort of thing and copy it. The other is to take some class of real problems which
require originality in their solution and attempt to nd a way to write a program to solve them on
an automatic calculator. Either of these approaches would probably eventually succeed. However,
it is not clear which would b e quicker nor how many years or generations it would take. Most of my
eort along these lines has so far b een on the former approach b ecause I felt that it would b e b est
to master all relevant scientic knowledge in order to work on such a hard problem, and I already
was quite aware of the current state of calculators and the art of programming them.
The control mechanism of the brain is clearly very dierent from the control mechanism in
to day's calculators. One symptom of the dierence is the manner of failure. A failure of a cal-
culator characteristically pro duces something quite unreasonable. An error in memory or in data
transmission is as likely to b e in the most signicant digit as in the least. An error in control
can do nearly anything. It might execute the wrong instruction or op erate a wrong input-output
unit. On the other hand human errors in sp eech are apt to result in statements which almost make
sense (consider someone who is almost asleep, slightly drunk, or slightly feverish). Perhaps the
8

mechanism of the brain is such that a slight error in reasoning intro duces randomness in just the
right way. Perhaps the mechanism that controls serial order in b ehavior
2
guides the random factor
so as to improve the eciency of imaginative pro cesses over pure randomness.
Some work has b een done on simulating neuron nets on our automatic calculator. One purp ose
was to see if it would b e thereby p ossible to intro duce randomness in an appropriate fashion. It
seems to have turned out that there are to o many unknown links b etween the activity of neurons
and problem solving for this approach to work quite yet. The results have cast some light on the
b ehavior of nets and neurons, but have not yielded a way to solve problems requiring originality.
An imp ortant asp ect of this work has b een an eort to make the machine form and manipulate
concepts, abstractions, generalizations, and names. An attempt was made to test a theory
3
of how
the brain do es it. The rst set of exp eriments o ccasioned a revision of certain details of the theory.
The second set of exp eriments is now in progress. By next summer this work will b e nished and
a nal rep ort will have b een written.
My program is to try next to write a program to solve problems which are memb ers of some
limited class of problems that require originality in their solution. It is to o early to predict just what
stage I will b e in next summer, or just; how I will then dene the immediate problem. However,
the underlying problem which is describ ed in this pap er is what I intend to pursue. In a single
sentence the problem is: how can I make a machine which will exhibit originality in its solution of
problems?
REFERENCES
1. K.J.W. Craik, The Nature of Explanation , Cambridge University Press, 1943 (reprinted
1952), p. 92.
2. K.S. Lashley, \The Problem of Serial Order in Behavior", in Cerebral Mechanism in
Behavior, the Hixon Symp osium , edited by L.A. Jeress, John Wiley & Sons, New York, pp. 112{
146, 1951.
3. D. O. Hebb, The Organization of Behavior , John Wiley & Sons, New York, 1949
9

PROPOSAL FOR RESEARCH BY JOHN MCCARTHY
During next year and during the Summer Research Pro ject on Articial Intelligence, I prop ose
to study the relation of language to intelligence. It seems clear that the direct application of trial
and error metho ds to the relation b etween sensory data and motor activity will not lead to any
very complicated b ehavior. Rather it is necessary for the trial and error metho ds to b e applied at
a higher level of abstraction. The human mind apparently uses language as its means of handling
complicated phenomena. The trial and error pro cesses at a higher level frequently take the form of
formulating conjectures and testing them. The English language has a numb er of prop erties which
every formal language describ ed so far lacks.
1. Arguments in English supplemented by informal mathematics can b e concise.
2. English is universal in the sense that it can set up any other language within English and
then use that language where it is appropriate.
3. The user of English can refer to himself in it and formulate statements regarding his progress
in solving the problem he is working on.
4. In addition to rules of pro of, English if completely formulated would have rules of conjecture.
The logical languages so far formulated have either b een instruction lists to make computers
carry out calculations sp ecied in advance or else formalization of parts of mathematics. The latter
have b een constructed so as:
1. to b e easily describ ed in informal mathematics,
2. to allow translation of statements from informal mathematics into the language,
3. to make is easy to argue ab out whether pro ofs of (???)
No attempt has b een made to make pro ofs in the articial languages as short as informal
pro ofs. It therefore seems to b e desirable to attempt to construct an articial language which a
computer can b e programmed to use on problems requiring conjecture and self-reference. It should
corresp ond to English in the sense that short English statements ab out the given sub ject matter
should have short corresp ondents in the language and so should short arguments or conjectural
arguments. I hop e to try to formulate a language having these prop erties and in addition to
contain the notions of physical ob ject, event, etc., with the hop e that using this language it will b e
p ossible to program a machine to learn to play games well and do other tasks .
10

PEOPLE INTERESTED IN THE
ARTIFICIAL INTELLIGENCE PROBLEM
The purp ose of the list is to let those on it know who is interested in receiving do cuments on
the problem. The p eople on the 1ist wlll receive copies of the rep ort of the Dartmouth Summer
Pro ject on Articial Intelligence.
The list consists of p eople who particlpated in or visited the Dartmouth Summer Research
Pro ject on Articlal Intelligence, or who are known to b e interested in the sub ject. It is b eing sent
to the p eople on the 1ist and to a few others.
For the present purp ose the articial intelligence problem is taken to b e that of making a
machine b ehave in ways that would b e called intelligent if a human were so b ehaving.
A revised list will b e issued so on, so that anyone else interested in getting on the list or anyone
who wishes to change his address on it should write to:
John McCarthy
Department of Mathematics
Dartmouth College
Hanover, NH
The list consists of:
Adelson, Marvin Ashby, W. R.
Hughes Aircraft Company Barnwo o d House
Airp ort Station, Los Angeles, CA Gloucester, England
Backus, John Bernstein, Alex
IBM Corp oration IBM Corp oration
590 Madison Avenue 590 Madison Avenue
New York, NY New York, NY
Bigelow, J. H. Elias, Peter
Institute for Advanced Studies R. L. E., MIT
Princeton, NJ Cambridge, MA
Duda, W. L. Davies, Paul M.
IBM Research Lab oratory 1317 C. 18th Street
Poughkeepsie, NY Los Angeles, CA.
Fano, R. M. Farley, B. G.
R. L. E., MIT 324 Park Avenue
Cambridge, MA Arlington, MA.
Galanter, E. H. Gelernter, Herb ert
University of Pennsylvania IBM Research
Philadelphia, PA Poughkeepsie, NY
Glashow, Harvey A. Go ertzal, Herb ert
1102 Olivia Street 330 West 11th Street
Ann Arb or, MI. New York, New York
Hagelbarger, D. Miller, George A.
Bell Telephone Lab oratories Memorial Hall
Murray Hill, NJ Harvard University
Cambridge, MA.
11

Harmon, Leon D. Holland, John H.
Bell Telephone Lab oratories E. R. I.
Murray Hill, NJ University of Michigan
Ann Arb or, MI
Holt, Anatol Kautz, William H.
7358 Rural Lane Stanford Research Institute
Philadelphia, PA Menlo Park, CA
Luce, R. D. MacKay, Donald
427 West 117th Street Department of Physics
New York, NY University of London
London, WC2, England
McCarthy, John McCullo ch, Warren S.
Dartmouth College R.L.E., M.I.T.
Hanover, NH Cambridge, MA
Melzak, Z. A. Minsky, M. L.
Mathematics Department 112 Newbury Street
University of Michigan Boston, MA
Ann Arb or, MI
More, Trenchard Nash, John
Department of Electrical Engineering Institute for Advanced Studies
MIT Princeton, NJ
Cambridge, MA
Newell, Allen Robinson, Abraham
Department of Industrial Administration Department of Mathematics
Carnegie Institute of Technology University of Toronto
Pittsburgh, PA Toronto, Ontario, Canada
Ro chester, Nathaniel Rogers, Hartley, Jr.
Engineering Research Lab oratory Department of Mathematics
IBM Corp oration MIT
Poughkeepsie, NY Cambridge, MA.
Rosenblith, Walter Rothstein, Jerome
R.L.E., M.I.T. 21 East Bergen Place
Cambridge, MA. Red Bank, NJ
Sayre, David Schorr-Kon, J.J.
IBM Corp oration C-380 Lincoln Lab oratory
590 Madison Avenue MIT
New York, NY Lexington, MA
Shapley, L. Schutzenb erger, M.P.
Rand Corp oration R.L.E., M.I.T.
1700 Main Street Cambridge, MA
Santa Monica, CA
Selfridge, O. G. Shannon, C. E.
Lincoln Lab oratory, M.I.T. R.L.E., M.I.T.
Lexington, MA Cambridge, MA
12

Shapiro, Norman Simon, Herb ert A.
Rand Corp oration Department of Industrial Administration
1700 Main Street Carnegie Technology
Santa Monica, CA Pittsburgh, PA
Solomono, Raymond J. Steele, J. E., Capt. USAF
Technical Research Group Area B., Box 8698
17 Union Square West Wright-Patterson AFB
New York, NY Ohio
Webster, Frederick Mo ore, E. F.
62 Co olidge Avenue Bell Telephone Lab oratory
Cambridge, MA Murray Hill, NJ
Kemeny, John G.
Dartmouth College
Hanover, NH
13