Reprinted fe Orssasione Reaeaaee ‘ntl ie 8 HEURISTIC PROBLEM SOLVING: THE NEXT ADVANCE IN OPERATIONS RESEARCH* Herbert A. Simon and Allen Newell Carnesie Institute of Tususlogy, Pittsburgh, Ponnouleania, and ‘The Hand Corporation, Aunts Montes, California , I ‘HE IDEA ‘THAT the development of science and its application to human affairs often requires the eouperation of many professions will not surprise the members of this audience, iplines and Operations research and management svienee are young professions that are ouly now ‘> doginning to develop their own programs of training; and they have mean. while drawn their practitioners from the whole spectrum of intellectual disciplines. We ure mathematicians, physical scientists, biologists statisticians, economists, and political seientiate, In some ways it is a very new idea to draw upon the techniques and fundamental kuc ledge of these fields in order to improve the everyday operation of administrative organizations, The terms ‘operatious re search’ and ‘management science’ have evolved in the past fifteen years, as have the organized activities associated with them. But of course, our professional activity, the application of intelligence in a systematic way to administration, has a history Uhat extends much farther into the past One of its obvious antecedents is the scien * fathered by Fanousick W. Tavzon. management movement. But for an appropriate patron saint for our profession, we can most appropriately look back a full half century before Taylor to the remarkable figure of Cuauues BaBpace. Perhaps more than any mau since Leonardo da Vinci he exemplified in his life and work the powerful ways in which * Addr ab the banquot of the Twelfth National Meeting of the Orwxnioxs aseanct Bocinrr or Awnnsen, Pittsburgh, Pennsylvania, November 14, Tal Ne, Simon presented the paper, is omen Jelnt prod of the autho, fa thoy rely on the precede 3 of Genesis 27:22, 1 hm vive is Jacob's vee, but the2 Simon and Newell fundamental science would contribute to practical faire, and practical affairs to ecience. He was one of the strongest mathematicians of his generation, but he dovoted his carer (othe improvement of manufacturing farts, and—most remarkable of allto the invention ofthe digital eamputer in something very close Lo its modem form, ‘The spirit of the operations researcher, his curiosity, his impalieneo with Inefficiency in uny aspeat of human affairs, shows forth from every payy of Babbage’s writing. give you just one example Clocks octupy o very high pluce simongst Instrumente by means of which hhuman time is economisel: and their multiplication in eouspieuous plagce in large towns ix attended with many advantages, helt position, neverthe: Jess, in London, is often very ill chosen; and the usual place, hallway up oo 2 high steeple, in Uhe midst of narrow streets, in x erowdod city, is very ‘wofavoursble, unless the church happen Us stand out from the hones whieh form the street. ‘The nwt elgibio station fr a clock is, that should projest considerably into the strect at some elevatiou, with ' dial-plate on earl side, like that which belong! to the old church of St, Dunstan, in Fleet strest, 80 that passengers iu both direotions would have their uttention directod te the hour. 82 T have mentioned Babbyxe as the inventor of the computer, Since Babbage and the computer are going to be the heroes of my talk. Lonight, 1 should lke to tell vou a true story, culled froin Babbage's writings, about the history of the computer. 1 like this story because it illustrates not only say earlier point about the mauy mutual relations of the professions in our field, but also because it gives the underdogs like myself trained in ‘sult ficlds like economies und political stience—something we ean point to when the superior accomplislnents of the natural sciences become too ‘embarrassing for us, As yon will see, thie story chows that physicists aod electrical engineers had little to do with the iuvention of the digital rom= Puter—that the real inveulor was the economist Adam Smith, whose ides was translated into hardware throuyh successive stages of development by two mathematicians, Prony and Hahbage. (I should pethaps, mention that the developers owed a debt also to the French weavers and mechanies ‘responsible for the Jacquard! loom, and consequently for the punched eard.) ‘The slory comes from a French docuuent, which Babbage repreaiuces fn the original lauguage. I give it here in translatin: Horo is the ancedote:-M_ de Prony was employed 1ittee to construct, for the decimal graduation af the circle, logarithmic and trigouometrie tables which would not only leave nothing to be desired from the standpoint of arruracy, but which would constitute the must vast and imposing monument of ealoulation that had ever buen executed or cven con- ved. Tho logarithms from 1 11 200,000 aro 8 nesessary and exentinl supple ment to this work. Tt was eaay for M. Prony to comvinee hiusslf that even ifHeuristic Problem Sotsing 3B hho associated with hineulf three or four experienced collaborators the longest reasonable expectation of the daration of his life woul nal suffice Wo complete ‘he undertaking. He was preoecupied with this unhappy thought when, find- ing himecif before a bookstore, he saw the beautiful edition of Adam Saith published in London in 1776. "Te opal Use book at random and chanced ‘upon the first chaptor, which treats of the division of labor and where the manufacture of pins is cited as example. ranily Tad he perused Uhe frst pages when, by a stroke of inepiration he cconcoived the expedient of putting his logurithmas into production like pins. Fle was giving, at this time, at the Henle Plytechniques, some leclures ou topic in analysis related to this kind of work—the method of differences and ite applications to interpolation, Hie went to spend some tune in the conntry and returnecl to Paris with the pla of manufacture (hat bas been followed in the execution. Ho organized twvo workshops which performed the same caleu- lations separately, and served as reciprocal cheeks He was Prony’s mass production of the mathematical tables, in tur, that suggested to Babbage that; machinery could replace humats labor in the clerical phases of the task, and that started him on the undertaking of clesigning and coustructing an automatie calculating engine. Although the complete absence of electrical and electronic components, aud his eonse- ‘quent dependence on mechanical devices, robbed him of full success in the uudertaking, (here is no doubt that he understood and invented the digital compnter—inchnding the evitieally iimporlaat idea of x condition! transfer operation, Tt would be hard to imagine a more appropriate illustration of the un- expected ways in whieh human kuowledge develops, and of the con- tribution ofall the sciences and arts to this development that:is so character- istie of operations research and management science. S WE TURN our gaze now from past to future, I should like to outline my main thesis quite bluntly. Operations recearch has made large contributions ta those management deeisious Usal exn be reduced to system- atic computational routines. To date, comparable progress has not beets ‘made in applying scientific techniques to the judgmental decisions that cannot be so reduced. Research of the past three years into the nature ‘of complex information processes in general, and human judguental or heuristic thinking proceszes in particular, is about to change this state of allsirs radically. We are now poised for a great advance that will bring the digital compnter and the tools of mathematies aud he behavioral seienees to bear on the very core of managerial activity—on the exercise ‘of judgment und intuition; on the processes of making complex decisions, et me spell out this thesis, frst describing the present situation in operations research as I sec it, then indicating why I think this situation is going to change drastically,4 Simon and Neseolt HE RAPID GROWTH of operations research over the past, two decades has brought to industry and goverument an important kit of tools for grappling with the complesities of managing large unganizations. These tools have been collected from the far corners of the intellectual world—from mathematics, from stalisties and probability theory, from econometrics, from electrical engineering, and even fim biology. | Such exotic techniques as lineur programming, queuing theory, servomechanisi theory, game theory, dynamie prograsuning, marginal analysis, the calculus of variations, and information theory are now at work helping to solve practical problens uf business operation. Skeptical—and sensibly skepticul—managements have come to see that, even if not all the blue-sky claims for the new approuches have been hacked by solid fuet, there is a large core of valid technique and applies tion, The tools have produced Langible resulte in a substantial number of demonstration installations, and the question is less und less ‘Are they here to stay?’ and more and more ‘How and where can we nse them effectively?” The traditional areas of production and inventory control, of scheduling, and of markcting research are undergoing « substantial and tapid evolution, Having observed this important change, we can nole with equal ac- enrary that large areas of managerial activity it would be correct ta say ‘most areas have harelly been touched by operations research or the ad- vances in management science. Operations researcls lux demonstrated its effectiveness in deuling with the kinds of management problems that we ‘might call ‘well structured,’ but il hus left pretty much untouched the re- ‘maining, ‘ll structured,’ problems. ‘The trouble, as executives are fond of pointing ont to operations searchers, is that there ure no known formal techniques for finding answers to most of the important: top-level management problems, Nor do these problems seem to be of the same kind as the more tangible iddle-ssanage- ‘ment situations in which existing operations research techniques have been ‘most effective. Unarmed with formnal lechniques, operations researchers have to resort to the same common sense and human eleverness that has served managements these many years, Executives still find a vast sphere of activity in which they are secure from the depredations of mathema- ticians and computers. Let me Uy lo make a litle more precice this distinction between well structured and ill-structured problems thut today establishes the juris dictional boundary beyond which formal tools do not reach, A problem is woll structured to the extent that it satisfies the following. criteria: 1. Te ean be described in terms of numerical variables, sealar amd vector ‘quantities, ~Houristie Problem Solving 3 2. ‘The goals to be attained ean be specified in terms of a wellalefined ubjective fSunction—for example, the maximization of profit or Uhe minimiaation of cost. 3. There exist compntational routines (algorithms) that permit the solution to tw: found snd stated in actual numerical terms. Common examples of such algorithms, which have played an important rule iw operations eetarch, are maxitn- ation procedures in the exleulus and ealeulus of variations, linear-programming slyorithms liko the etepping ctone and suplex methevs, Moule Curlo techniquen, ‘and eo on. In short, well-structured problems are those that can be formulated ex plicitly and quantitatively, and Usst can then be solved by known and feasible computational techniques. What, then, ure ill-structured problems? Problems are ill-structured when they are not wellstructured. In some eases, for example, the essen- tial variables are not numerical at all, but symbolic or verbal. An executive who is drafting a sick-leave policy is searching for words, nul auibers. Second, there are many important situations in everyday life where the objective function, the goal, is vague aud nonquantitative. How, for example, do we evaluate the quality of an educational system or the effectiveness of @ public relutions department? Third, there are many practical problems —it, would he aceuralw Lo say ‘most practical probleme? for which computational algorithms simply are not availuble, If we face the fucts of organizational life, we are forced to admit thal the majority of decisions that executives face every day and certainly a ujority of the very most important decisions—Iie much eloser to the ill structured than to the well-structured end of the spectrum. And yet, operations research and management science, for all their solid contritn tions to management, have mot yot made such headway in the arca of i structured problems. These are still almost exclusively the pro the experienced manager with his ‘judgment and intuition.’ ‘The hasie decisions about the design of organization structures are still made by judgment rather than science; business policy at top-nauagement levels is still more often x matter of hunch than of calculation. Operations 1 search has had more to do with the factory manager and the production- scheduling clerk than it has with the vier-president aud the Board of Directors, Tam not unaware that operations researchers are often called in to advise management at top levels and regarding problems of the kinds I have calted ill-structured. But T think we all recognize that when we are ‘asked by management. to advise on such decisions, we are asked because we are thought to possess a certain amount of experience and common sense, and ol because of any belie that our specialized tonls, imatheistical oF otherwise, have much lw do with the task at hand, [ think mast. of us ‘ean distinguich pretty clearly helween the eases in which we are working of6 Simon and Newell as operations researchers, and those ia which we are performing as general ‘management consultants. And I am sure that most of us look forward to the day when our science will enable us to handle with appropriate unlytie tools those problems that we now tackle with judganent and guess ‘The basic fuel we ave to recognize is that-no matter how strongly we ‘wish to treat problema with the tools our seienee provides us, we ean only oo when the situations thst confront us ie in the ares Wu which the tools apply. Techniques are Uke aems and hands of science, and the reach of a science is measired by their range. The telescope made sunspots and Tupiter's moons a part of Galllo's science, just as particle wevelerators and the mathematical machiuery of quantum mechanies bring theinterior of the som within the reach of the nuclear physicist, Tn dealing with the ill structured problems of maunagewent we have not had the mathematical lls we have needed—we have not had ‘judg- teal mechanics’ to match quantum mechanies. We have not had the fengines—no executive eontrifuges. Wo havo had only the rudiaeuts of experimental techniques for observing organizational behavior in the laboratory, allhough we have made great strides in the last’ deeade in developing thes F OUR SCIENCE, then, is to be coextensive with the ficld of manage: ment, we must have the tools and techniques that will extend its range to that whole field, I think there is good reason to believe that we are acquiring thee Lools and techniques at this very point in history. ‘Even while operations research is solving well-structured problems, fundamental research is dissolving the mystery of how humans solve ill- structured problems. Moreover, we have begun to learn hove to u pulers to solve these problems, where we do not have systems ‘efficient computational algorithms. And wenow know, at least in limited area, not only how to program computers to perform sich problem-solving activities successfully; we know also how to program computers to learn to do these thi In short, we now have the elements uf « theory of henristie (as co trasted wilh algorithmic) problem aalving; and we can use this theory both to understand human heurietio processes and to simulate such processes with digital computers. Intuition, insight, and learning are no longer exclusive possessions of humans: any large high-speed computer can be programmed to exhibit them also, ‘T cannot give here the detailed evidence ou which these assertions—and very strong assertions they are—are hased. I must warn you that ex: suuples of sueressful computer programs for heuristic problem solving are still very few, One pioneering effort was a program written hy 0. G.Heuristic Problem Solving 1 Seurnipor and G. P, Dinxrew that permitted « computer to learn to Listinguish between figures representing the letter 0 and figures represent ing A presented to it ‘visually."® The program that has been described ‘most completely in the literature gives « computer the ability to discover proofs for mathematical theorems not to verify proof, it should be noted, for a simple algoritlun could be devised for that, but to perform the ‘creative’ and ‘intuitive? activities of w scientist seeking the proof of Ueorem. ‘The program is also being used to predict the beluvior of humans when solving such problems, This program is the product of work carried on jointly at the Carnegie Institute of Technology and the Rand Corpora- tion, by Allen Newell, J. C. Shave, and myself! ‘A number of investigations in the same general direetion—involving such human activities as language Lrunslution, chess playing, engineering design, musical composition, and pattern recoguilion—are undor Way at other reswurch centers. At least one computer now designs small standard clectric motors (From customer specifications to the final design) for a manufacturing concer, one playsa pretty fair gaue of checkers, and several vollers know the rudiments of chess. ‘The ttzrac, at the Univenity of Mlinois, composes snusic, using I believe, the counterpoint of Palestrina; ‘and Tam told by a competent. judge that the resulting product is acs. Ghetically interesting, lot me sunumatine ws concretely ae poeeible my accessment of the present ‘and future state of the art and theory of heuristic problem solving. As of the presont—1057: 1, Digital computers can perfurm certains heuristic problem solving tasks for algorithms are available 2, In doing so, they use process that ate closely parallel to human problem- solving proves, 3. Within limite, these machines Ica to improve their performance on the basis of experience (not merely hy memuriaing specific patterns of succesful be- havior, but by reprogramming themgelves in ways that parallel at least some ‘human learning procedures), On the hasis of these developments, und the speed with which research in this field is progressing, I am willing ta make the following predictions, to be realized within the next ten years 1, Thot within ten years a digital compu ‘unless the rales har it from eanupetition, 2. That within ten years a digital computer will discover an! prove aut im- portant new mathematical thearem. A. Tak within ten years a digital computer will write music that will be ar cepted by critios as pomscaing considerable aesthetic. value 4. That within ten years mast theories in psychology will take the form of + will be he world’s chess champion,a Simon and Neveell computer programs, or of qualitative statements about the characteristics of ‘computer programs. tis not my aim to surprise or shock you—if indeed that were possible in an age of nuclear fission and prospective interplanctary travel. Hut the simplest way T can summarize dhe situation is to say that there are now in the world machines that think, that learn, and that create, More. over, their ability to do these things is going to increase rapidly unfil—in a visible future—Uhe range of problems they ean handle will be coextensive with the range to which the human mind has heen applied ‘What are the implications of this development? ‘They are of at, least three rather distinel kinds 1.-There will be more and more applications of machines to take the plive of humans in solving i-structaree! problems; just as muuchines are now being more ‘and more used to solve well-structured probleme, ‘There will be applications of machines to tackle ikstructured problems of such mageitude snd difficulty that humans have not been able to solve them. (This ie parallel to eurrent applications of computers to the numeral solution of pactial differential equations that ie heyonnl the eapacily of huad methods.) 8. The research on houristie problom eolving will be applied to understanding ‘the human mind. With the aid of heuristic programs, we will help nan obey the ancient injumetion: Koow thyself, Aud kuowing himself, he may learn to usc advancos of knowledge to benefit, rathor than destroy, the human species, ~ In estimating the rates at which these developments will come about. it may be instructive to turn to a elose analogy in the field of atomir energy ‘The implications of atomic energy are also threefold: (1) the generation of power to replace and angment power from couveutional fuels; (2) the production of hitherto unrealizable concentrations of power (the primary pewceful application being thas far to the study of the interior of the atom); and (i) the use of radiosetive materials as avers for the study of physical ‘and biological processes, The main point in drawing the analogy is a in oth exses—computers and atomic energy—the usefulness of the frst application hinges on ecouomie valeultions, while the significance of the ‘other two rests mainly on their technical feasibility. Atomie fuels will replace conventional fuels only when the capital casts per unit of energy-generuling capacity are competitive with the eapital sts of conventional plants. Computers far heuristic problem solving will replace executives only when the costs per unit of problem-solving capacity are competitive with the costs for exeeutives. In neither ease is it, casy to make a forecast with available data, but it seems highly probable in both cases that the changeover, if it comes, will come gradually. ‘A substantial impact of heurstie problem solving on research (cither inHourtette Problem Solving 9 allowing us to tackle more difficult problems than humans now eau, or in informing us how talented humans solve problems) is probably more im= minent, Here—us in the parallel eases for atemic euergy—the question will be very little How much will it cost?" and very much ‘Can we do it? It is noither a trivial nor a costless process to transfer from a productive scientist lo a student the heuristic programsthat make (he former a power- ful problem solver. To do this generally takes same twenly years of ‘educational effort, and the undertaking is frequently unsuccessful. "T's ree produce in another computer a problem-solving. program that has been eared and heen proved effective by a first computer is a Livial matter, ‘When machines will have minds, we can create copies of these minds ss hexply as we ean now print books. If what I have said still seems distant and speculative to you, I would like to recall to yom again the precedent of Charles Babbage, who, always ‘standing on the realities of the present saw the importance also of peering into the future and forecasting its shape. Pethape to the sober eye of inductive philosophy, thee anticipations of ‘the future may sppexr too faintly eonncoted with the history of the past... Even now, the imprisoned winds which the exeliest post mado the Grecian, warrior bear for the protection of his fragile bark; oF those which, in mane ‘movievn Gime, the Lapland wizards sold to the eetnted sailors;—these, the ‘unreal creations of fancy or of fra, culled, at the command of science from their shadowy existenee, obey a holler epell: and the unruly masters of the ‘wt and the seer become the obedient slaves of eivilieed saan Nor have the wild imaginings uf Uhe satirist been quite unrivalled hy the realities of after years: as ifn mookery of the Collegeof Laut, light almost solar hae been extracted from the refuse of fish; fie has been sifted by the lump of Davy’; and machinery lus boon taught arithmetic instead of pert {6 PERHAPS this is a. sppropriate point to bring my sprenlations lo a clove and to summarize briefly the course of my argument. Up ta the present time, operations rescarch and the management seienees have becn largely limited, by the nuture of their tools, to dealing with well-structured broblems that possess algorithmic means of solution. With recent develop ments in our understanding of heuristic praceses and their eimulation by digital computers, the way is open to deal acientifically with illstruc- tured problems -to make the computer coextensive with the human mind. ‘The energy revolution of the eighteenth und nineteenth conturies {forced man to reconsider his role in a world in which his physical power and speed were outstripped Ly dhe power and speed of machines. ‘The revolution in heuristic problem solving will furce man to consider his role in 8 world in which his intellectual power and speed are outstripped by the0 Simon and Newell intelligence of machines. Fortunately, the new revolution will at the same me give him a deeper understanding of the structure and warkings of his ‘own mind. It is my personal hope that the Initer developawent will outstrip the former—that man will lear where he wants to travel before he acquires the capability of leaving Ue planet. REFEREN 1. Cuanuts Bassace, On the Hemuomy of Machinery and Manufacturers p45. 2. Lbid.,p- 188. [Quoted hy Dalim fron a Note sur la publication, propoee parle ‘ouaernement Anglcis les yrands tables logarithniquee et trigonomatriques de ‘M-de Prony, (1820)] A. 0G. Seismmer, “Palvera Recognition and Modem Computers,” and G. P. Tiwwsen, “Programming Patter Reoognition,” both in Proceedings of the 1055 Western Joint Computer Conference, IRE, 4. "The Logic Theory Machine,” [RE Trangactiona 1V-2, 61-79 (September 1950); ‘Empirical Explorations of the Logic Theory Machine” and “Programming the Logic Theory Muchine,” Proceedings of the 1957 Western aint. Comper Conference, IRE; “The Elements of a ‘Theory of Muman Problem Selvin,” Pouch, Ree. in press. 5, Chtanuns Bassace, op. cit, p. 289,