|
原文如下,來自http://www-formal.stanford.edu/j ... outh/dartmouth.html
Next:About this document
A PROPOSAL FOR THEDARTMOUTH SUMMER RESEARCH PROJECT ON ARTIFICIAL INTELLIGENCE
J. McCarthy, Dartmouth College
M. L. Minsky, Harvard University
N. Rochester, I.B.M. Corporation
C.E. Shannon, Bell Telephone Laboratories
August 31, 1955
We propose that a 2 month, 10 man study of artificial intelligence be carried outduring the summer of 1956 at Dartmouth College in Hanover, New Hampshire. Thestudy is to proceed on the basis of the conjecture that every aspect oflearning or any other feature of intelligence can in principle be so preciselydescribed that a machine can be made to simulate it. An attempt will be made tofind how to make machines use language, form abstractions and concepts, solvekinds of problems now reserved for humans, and improve themselves. We thinkthat a significant advance can be made in one or more of these problems if acarefully selected group of scientists work on it together for a summer.
The following are some aspects of the artificial intelligenceproblem:
1. AutomaticComputers
If amachine can do a job, then an automatic calculator can be programmed to simulatethe machine. The speeds and memory capacities of present computers may beinsufficient to simulate many of the higher functions of the human brain, butthe major obstacle is not lack of machine capacity, but our inability to writeprograms taking full advantage of what we have.
2. HowCan a Computer be Programmed to Use a Language
It maybe speculated that a large part of human thought consists of manipulating wordsaccording to rules of reasoning and rules of conjecture. From this point ofview, forming a generalization consists of admitting a new word and some ruleswhereby sentences containing it imply and are implied by others. This idea hasnever been very precisely formulated nor have examples been worked out.
3. NeuronNets
How cana set of (hypothetical) neurons be arranged so as to form concepts.Considerable theoretical and experimental work has been done on this problem byUttley, Rashevsky and his group, Farley and Clark, Pitts and McCulloch, Minsky,Rochester and Holland, and others. Partial results have been obtained but theproblem needs more theoretical work.
4. Theoryof the Size of a Calculation
If weare given a well-defined problem (one for which it is possible to testmechanically whether or not a proposed answer is a valid answer) one way ofsolving it is to try all possible answers in order. This method is inefficient,and to exclude it one must have some criterion for efficiency of calculation.Some consideration will show that to get a measure of the efficiency of a calculationit is necessary to have on hand a method of measuring the complexity ofcalculating devices which in turn can be done if one has a theory of thecomplexity of functions. Some partial results on this problem have beenobtained by Shannon, and also by McCarthy.
5. Self-lmprovement
Probablya truly intelligent machine will carry out activities which may best bedescribed as self-improvement. Some schemes for doing this have been proposedand are worth further study. It seems likely that this question can be studiedabstractly as well.
6. Abstractions
Anumber of types of ``abstraction'' can be distinctly defined and several othersless distinctly. A direct attempt to classify these and to describe machinemethods of forming abstractions from sensory and other data would seemworthwhile.
7. Randomnessand Creativity
Afairly attractive and yet clearly incomplete conjecture is that the differencebetween creative thinking and unimaginative competent thinking lies in theinjection of a some randomness. The randomness must be guided by intuition tobe efficient. In other words, the educated guess or the hunch includecontrolled randomness in otherwise orderly thinking.
Inaddition to the above collectively formulated problems for study, we have askedthe individuals taking part to describe what they will work on. Statements bythe four originators of the project are attached.
Wepropose to organize the work of the group as follows.
Potentialparticipants will be sent copies of this proposal and asked if they would liketo work on the artificial intelligence problem in the group and if so what theywould like to work on. The invitations will be made by the organizing committeeon the basis of its estimate of the individual's potential contribution to thework of the group. The members will circulate their previous work and theirideas for the problems to be attacked during the months preceding the workingperiod of the group.
Duringthe meeting there will be regular research seminars and opportunity for themembers to work individually and in informal small groups.
The originators of this proposal are:
1. C.E. Shannon, Mathematician, Bell Telephone Laboratories. Shannon developedthe statistical theory of information, the application of propositional calculusto switching circuits, and has results on the efficient synthesis of switchingcircuits, the design of machines that learn, cryptography, and the theory ofTuring machines. He and J. McCarthy are co-editing an Annals of MathematicsStudy on ``The Theory of Automata'' .
2. M.L. Minsky, Harvard Junior Fellow in Mathematics and Neurology. Minsky hasbuilt a machine for simulating learning by nerve nets and has written a PrincetonPhD thesis in mathematics entitled, ``Neural Nets and the Brain Model Problem''which includes results in learning theory and the theory of random neural nets.
3. N. Rochester, Manager of Information Research,IBM Corporation, Poughkeepsie, New York. Rochester was concerned with thedevelopment of radar for seven years and computing machinery for seven years.He and another engineer were jointly responsible for the design of the IBM Type701 which is a large scale automatic computer in wide use today. He worked outsome of the automatic programming techniques which are in wide use today andhas been concerned with problems of how to get machines to do tasks whichpreviously could be done only by people. He has also worked on simulation ofnerve nets with particular emphasis on using computers to test theories inneurophysiology.
4. J. McCarthy, Assistant Professor of Mathematics,Dartmouth College. McCarthy has worked on a number of questions connected withthe mathematical nature of the thought process including the theory of Turingmachines, the speed of computers, the relation of a brain model to itsenvironment, and the use of languages by machines. Some results of this workare included in the forthcoming ``Annals Study'' edited by Shannon and McCarthy.McCarthy's other work has been in the field of differential equations.
TheRockefeller Foundation is being asked to provide financial support for theproject on the following basis:
1.Salaries of $1200 for each faculty level participant who is not being supportedby his own organization. It is expected, for example, that the participantsfrom Bell Laboratories and IBM Corporation will be supported by theseorganizations while those from Dartmouth and Harvard will require foundationsupport.
2. Salariesof $700 for up to two graduate students.
3.Railway fare for participants coming from a distance.
4. Rentfor people who are simultaneously renting elsewhere.
5.Secretarial expenses of $650, $500 for a secretary and $150 for duplicatingexpenses.
6.Organization expenses of $200. (Includes expense of reproducing preliminarywork by participants and travel necessary for organization purposes.
7.Expenses for two or three people visiting for a short time.
EstimatedExpenses
6 salaries of 1200 $7200
2 salaries of 700 &1400
8 traveling and rent expenses averaging 300 &2400
Secretarial and organizational expense &850
Additional traveling expenses &600
Contingencies &550
&----&
$13,500
PROPOSAL FOR RESEARCH BY C.E.SHANNON
I would like to devote my research to one or both of the topicslisted below. While I hope to do so, it is possible thatbecause of personal considerations I may not be able to attend for the entiretwo months. I, nevertheless, intend to be there for whatever time is possible.
1.Application of information theory concepts to computing machines and brainmodels. A basic problem in information theory is that of transmittinginformation reliably over a noisy channel. An analogous problem in computingmachines is that of reliable computing using unreliable elements. This problemhas been studies by von Neumann for Sheffer stroke elements and by Shannon andMoore for relays; but there are still many open questions. The problem forseveral elements, the development of concepts similar to channel capacity, thesharper analysis of upper and lower bounds on the required redundancy, etc. areamong the important issues. Another question deals with the theory ofinformation networks where information flows in many closed loops (ascontrasted with the simple one-way channel usually considered in communicationtheory). Questions of delay become very important in the closed loop case, anda whole new approach seems necessary. This would probably involve concepts suchas partial entropies when a part of the past history of a message ensemble isknown.
2. Thematched environment - brain model approach to automata. In general a machine oranimal can only adapt to or operate in a limited class of environments. Eventhe complex human brain first adapts to the simpler aspects of its environment,and gradually builds up to the more complex features. I propose to study thesynthesis of brain models by the parallel development of a series of matched(theoretical) environments and corresponding brain models which adapt to them.The emphasis here is on clarifying the environmental model, and representing itas a mathematical structure. Often in discussing mechanized intelligence, wethink of machines performing the most advanced human thought activities-provingtheorems, writing music, or playing chess. I am proposing here to start at thesimple and when the environment is neither hostile (merely indifferent) norcomplex, and to work up through a series of easy stages in the direction ofthese advanced activities.
PROPOSAL FOR RESEARCH BY M.L.MINSKY
It isnot difficult to design a machine which exhibits the following type oflearning. The machine is provided with input and output channels and aninternal means of providing varied output responses to inputs in such a waythat the machine may be ``trained'' by a ``trial and error'' process to acquireone of a range of input-output functions. Such a machine, when placed in anappropriate environment and given a criterior of ``success'' or ``failure'' canbe trained to exhibit ``goal-seeking'' behavior. Unless the machine is providedwith, or is able to develop, a way of abstracting sensory material, it canprogress through a complicated environment only through painfully slow steps,and in general will not reach a high level of behavior.
Now letthe criterion of success be not merely the appearance of a desired activitypattern at the output channel of the machine, but rather the performance of agiven manipulation in a given environment. Then in certain ways the motorsituation appears to be a dual of the sensory situation, and progress can bereasonably fast only if the machine is equally capable of assembling anensemble of ``motor abstractions'' relating its output activity to changes in theenvironment. Such ``motor abstractions'' can be valuable only if they relate tochanges in the environment which can be detected by the machine as changes inthe sensory situation, i.e., if they are related, through the structure of theenvironrnent, to the sensory abstractions that the machine is using.
I havebeen studying such systems for some time and feel that if a machine can bedesigned in which the sensory and motor abstractions, as they are formed, canbe made to satisfy certain relations, a high order of behavior may result.These relations involve pairing, motor abstractions with sensory abstractionsin such a way as to produce new sensory situations representing the changes inthe environment that might be expected if the corresponding motor act actuallytook place.
Theimportant result that would be looked for would be that the machine would tendto build up within itself an abstract model of the environment in which it isplaced. If it were given a problem, it could first explore solutions within theinternal abstract model of the environment and then attempt externalexperiments. Because of this preliminary internal study, these externalexperiments would appear to be rather clever, and the behavior would have to beregarded as rather ``imaginative''
A verytentative proposal of how this might be done is described in my dissertationand I intend to do further work in this direction. I hope that by summer 1956 Iwi11 have a model of such a machine fairly close to the stage of programming ina computer.
PROPOSAL FOR RESEARCH BY N. ROCHESTER
Originality in Machine Performance
Inwriting a program for an automatic calculator, one ordinarily provides themachine with a set of rules to cover each contingency which may arise andconfront the machine. One expects the machine to follow this set of rulesslavishly and to exhibit no originality or common sense. Furthermore one isannoyed only at himself when the machine gets confused because the rules he hasprovided for the machine are slightly contradictory. Finally, in writingprograms for machines, one sometimes must go at problems in a very laboriousmanner whereas, if the machine had just a little intuition or could makereasonable guesses, the solution of the problem could be quite direct. Thispaper describes a conjecture as to how to make a machine behave in a somewhatmore sophisticated manner in the general area suggested above. The paperdiscusses a problem on which I have been working sporadically for about fiveyears and which I wish to pursue further in the ArtificialIntelligence Projectnext summer.
|
|