Platt

NNoottee:: WWhhiillee ggrreeaatt ccaarree wwaass ggiivveenn ttoo rreepprroodduuccee tthhiiss aarrttiiccllee eexxaaccttllyy aass tthhee oorriiggiinnaall,, tthhiiss ddooccuummeenntt wwaass ttrraannssccrriibbeedd ffrroomm aann eexxttrreemmeellyy ssccrraattcchheedd mmiiccrrooffiillmm ccooppyy..DDiisscceerrnniinngg ppuunnccttuuaattiioonn wwaass eessppeecciiaallllyy ddiiffffiiccuulltt,, aanndd iitt iiss,, qquuiittee ppoossssiibbllyy,, ddiiffffeerreenntt ffrroomm tthhee oorriiggiinnaall..

16 OCTOBER 1964 347

16 October 1964, Volume 146, Number 3642 SCIENCE

Strong InferenceCertain systematic methods of scientific thinking may produce much more rapid progress thanothers.

John R. Platt

Scientists these days tend to keep up apolite fiction that all science is equal.Except for the work of the misguidedopponent whose arguments we happen tobe refuting at the time, we speak asthough every scientist's field andmethods of study are as good as everyother scientist's, and perhaps a littlebetter. This keeps us all cordial when itcomes to recommending each other forgovernment grants.

But I think anyone who looks at thematter closely will agree that some fieldsof science are moving forward verymuch faster than others, perhaps by anorder of magnitude, if numbers could beput on such estimates. The discoveriesleap from the headlines--and they arereal advances in complex and difficultsubjects, like molecular biology andhigh-energy physics. As Alvin Weinbergsays1, "Hardly a month goes by withouta stunning success in molecular biologybeing reported in the Proceedings of theNational Academy of Sciences."

Why should there be such rapidadvances in some fields and not inothers? I think the usual explanationsthat we tend to think of -- such as thetractability of the subject, or the qualityor education of the men drawn into it, orthe size of research contracts--areimportant but inadequate. I have begunto believe that the primary factor inscientific advance is an intellectual one.These rapidly moving fields are fieldswhere a particular method of doingscientific research is systematically usedand taught, an accumulative method ofinductive inference that is so effectivethat I think it should be given the nameof "strong inference." I believe it is

important to examine this methoditsuse and history and rationale, and to seewhether other groups and individualsmight learn to adopt it profitably in theirown scientific and intellectual work.

In its separate element, stronginference is just the simple and old-fashioned method of inductive inferencethat goes back to Francis Bacon. Thesteps are familiar to every collegestudent and are practiced, off and on, byevery scientist. The difference comes intheir systematic application. Stronginference consists of applying thefollowing steps to every problem inscience, formally and explicitly andregularly:1) Devising alternative hypotheses;2) Devising a crucial experiment (or

several of them), with alternativepossible outcomes, each of whichwill, as nearly as possible, excludeone or more of the hypotheses;

3) Carrying out the experiment so as toget a clean result;

1') Recycling the procedure, makingsubhypotheses or sequential hypothesesto refine the possibilities that remain;and so on.

It is like climbing a tree. At the firstfork, we choose--or, in this case,"nature" or the experimental outcomechooses--to go to the right branch or theleft; at the next fork, to left or right; andso on. There are similar branch points ina "conditional computer program,"where the next move depends on theresult of the last calculation. And there isa "conditional inductive tree" or "logicaltree" of this kind written out in detail inmany first-year chemistry books, in thetable of steps for qualitative analysis of

an unknown sample, where the student isled through a real problem ofconsecutive inference: Add reagent A; ifyou get a red precipitate, it is sub-groupalpha and you filter and add reagent B; ifnot, you add the other reagent, B; and soon.On any new problem, of course,inductive inference is not as simple andcertain as deduction, because it involvesreaching out into the unknown. Steps 1and 2 require intellectual inventions,which must be cleverly chosen so thathypothesis, experiment, outcome, andexclusion will be related in a rigoroussyllogism; and the question of how togenerate such inventions is one whichhas been extensively discussedelsewhere2 3. What the formal schemareminds us to do is to try to make theseinventions, to take the next step, toproceed to the next fork, withoutdawdling or getting tied up inirrelevancies.

It is clear why this makes for rapidand powerful progress. For exploring theunknown, there is no faster method; thisis the minimum sequence of steps. Anyconclusion that is not an exclusion isinsecure and must be rechecked. Anydelay in recycling to the next set ofhypotheses is only a delay. Stronginference, and the logical tree itgenerates, are to inductive reasoningwhat the syllogism is to deductivereasoning, in that it offers a regularmethod for reaching firm inductiveconclusions one after the other as rapidlyas possible.

"But what is so novel about this?"someone will say. This is the method ofscience and always has been; why give it

348 SCIENCE, VOL. 146

a special name? The reason is that manyof us have almost forgotten it. Science isnow an everyday business. Equipment,calculations, lectures become ends inthemselves. How many of us write downour alternatives and crucial experimentsevery day, focusing on the exclusion of ahypothesis? We may write our scientificpapers so that it looks as if we had steps1, 2, and 3 in mind all along. But inbetween, we do busywork. We become"method-oriented" rather than "problem-oriented." We say we prefer to "feel ourway" toward generalizations. We fail toteach our students how to sharpen uptheir inductive inferences. And we donot realize the added power that theregular and explicit use of alternativehypotheses and sharp exclusions couldgive us at every step of our research.

The difference between the averagescientist's informal methods and themethods of the strong-inference users issomewhat like the difference between agasoline engine that fires occasionallyand one that fires in steady sequence. Ifour motorboat engines were as erratic asour deliberate intellectual efforts, mostof us would not get home for supper.

Molecular Biology

The new molecular biology is a fieldwhere I think this systematic method ofinference has become wide-spread andeffective. It is a complex field; yet asuccession of crucial experiments overthe past decade has given us asurprisingly detailed understanding ofhereditary mechanisms and the controlof enzyme formation and proteinsynthesis.

The logical structure shows in everyexperiment. In 1953 James Watson andFrancis Crick proposed that the DNAmolecule-- the "hereditary substance" ina cell--is a long two-stranded helicalmolecule4. This suggested a number ofalternatives for crucial test. Do the twostrands of the helix stay together when acell divides, or do they separate?Matthew Neselson and Franklin Stahlused an ingenious isotope-density-labeling technique which showed thatthey separate5. Does the DNA helixalways have two strands, or can it have

three, as atomic models suggest?Alexander Rich showed it can haveeither, depending on the ionicconcentration6. These are the kinds ofexperiments John Dalton would haveliked, where the combining entities arenot atoms but long macromolecularstrands.

Or take a different sort of question:Is the "genetic map"--showing thestatistical relationship of differentgenetic characteristics in recombinationexperiments--a one-dimensional maplike the DNA molecule (that is, a linearmap), as T.H. Morgan proposed in 1911,or does it have two-dimensional loops orbranches? Seymour Benzer showed thathis hundreds of fine micro-geneticexperiments on bacteria would fit onlythe mathematical matrix for the one-dimensional case7.

But of course, selected crucialexperiments of this kind can be found inevery field. The real difference inmolecular biology is that formalinductive inference is so systematicallypracticed and taught. On any givenmorning at the Laboratory of MolecularBiology in Cambridge, England, theblackboards of Francis Crick or SidneyBrenner will commonly be foundcovered with logical trees. On the topline will be the hot new result just upfrom the laboratory or just in by letter orrumor. On the next line will be two orthree alternative explanations, or a littlelist of "What he did wrong."Underneath will be a series of suggestedexperiments or controls that can reducethe number of possibilities. And so on.The tree grows during the day as oneman or another comes in and arguesabout why one of the experimentswouldn't work, or how it should bechanged.

The strong-inference attitude isevident just in the style and language inwhich the papers are written. Forexample, in analyzing theories ofantibody formation, Joshua Lederberg8

gives a list of nine propositions "subjectto denial," discussing which ones wouldbe "most vulnerable to experimentaltest."

The papers of the French leadersFrancois Jacob and Jacques Monod are

also celebrated for their high "logicaldensity," with paragraph after paragraphof linked "inductive syllogisms." Butthe style is widespread. Start with thefirst paper in the Journal of MolecularBiology for 19649, and you immediatelyfind: "Our conclusions might beinvalid if(i)(ii)or (iii)We shalldescribe experiments which eliminatethese alternatives." The averagephysicist or chemist or scientist in anyfield accustomed to less closely reasonedarticles and less sharply stated inferenceswill find it a salutary experience to dipinto that journal almost at random.

Resistance to Analytical Methodology

This analytical approach to biologyhas sometimes become almost a crusade,because it arouses so much resistance inmany scientists who have grown up in amore relaxed and diffuse tradition. Atthe 1958 Conference on Biophysics, atBoulder, there was a dramaticconfrontation between the two points ofview. Leo Szilard said: "The problemsof how enzymes are induced, of howproteins are synthesized, of howantibodies are formed, are closer tosolution than is generally believed. Ifyou do stupid experiments, and finishone a year, it can take 50 years. But ifyou stop doing experiments for a littlewhile and think how proteins canpossibly be synthesized, there are onlyabout 5 different ways, not 50! And itwill take only a few experiments todistinguish these."

One of the young men added: "It isessentially the old question: How smalland elegant an experiment can youperform?"

These comments upset a number ofthose present. An electron microscopistsaid, "Gentlemen, this is off the track.This is philosophy of science."

Szilard retorted, "I was notquarreling with third-rate scientists: Iwas quarreling with first-rate scientists."

A physical chemist hurriedly asked,"Are we going to take the officialphotograph before lunch or after lunch?"

But this did not deflect the dispute.A distinguished cell biologist rose andsaid, "No two cells give the same


16 OCTOBER 1964

properties. Biology is the science ofheterogeneous systems." And he addedprivately, "You know there are scientists,and there are people in science who arejust working with these over-simplifiedmodel systems -- DNA chains and invitro systems -- who are not doingscience at all. We need their auxiliarywork: they build apparatus, they makeminor studies, but they are notscientists."

To which Cy Levinthal replied:"Well, there are two kinds of biologists,those who are looking to see if there isone thing that can be understood, andthose who keep saying it is verycomplicated and that nothing can beunderstood. You must study thesimplest system you think has theproperties you are interested in."

As they were leaving the meeting,one man could be heard muttering,"What does Szilard expect me to do --shoot myself?"

Any criticism or challenge toconsider changing our methods strikes ofcourse at all our ego-defenses. But inthis case the analytical method offers thepossibility of such great increases ineffectiveness that it is unfortunate that itcannot be regarded more often as achallenge to learning rather than as achallenge to combat. Many of therecent triumphs in molecular biologyhave in fact been achieved on just such"oversimplified model systems," verymuch along the analytical lines laiddown in the 1958 discussion. Theyhave not fallen to the kind of men whojustify themselves by saying, "No twocells are alike," regardless of how truethat may ultimately be. The triumphsare in fact triumphs of a new way ofthinking.

High-Energy Physics

This analytical thinking is rare, butit is by no means restricted to the newbiology. High-energy physics is anotherfield where the logic of exclusions isobvious even in the newspaper accounts.For example, in the famous discovery ofC. N. Yang and T. D. Lee, the questionthat was asked was: Do the fundamentalparticles conserve mirror-symmetry or

"parity" in certain reactions, or do theynot? The crucial experiments weresuggested: within a few months theywere done, and conservation of paritywas found to be excluded. RichardGarwin, Leon Lederman, and MarcelWeinrich did one of the crucialexperiments. It was thought of oneevening at suppertime: by midnight theyhad arranged the apparatus for it; and by4 am they had picked up the predictedpulses showing the non-conservation ofparity.10 The phenomena had just beenwaiting, so to speak, for the explicitformulation of the alternativehypotheses.

The theorists in this field take pridein trying to predict new properties ornew particles explicitly enough so that ifthey are not found the theories will fall.As the biologist W. A. H. Rushton hassaid,11 "A theory which cannot bemortally endangered cannot be alive."Murray Gell-Mann andYuval Ne'emanrecently used the particle groupingwhich they call "The Eightfold Way" topredict a missing particle, the Omega-Minus, which was then looked for andfound.12 But one alternative branch ofthe theory would predict a particle withone-third the usual electronic charge, andit was not found in the experiments, sothis branch must be rejected.

The logical tree is so much a part ofhigh-energy physics that some stages ofit are commonly built, in fact, into theelectronic coincidence circuits that detectthe particles and trigger the bubble-chamber photographs. Each kind ofparticle should give a different kind ofpattern in the electronic counters, and thecircuits can be set to exclude or includewhatever types of events are desired. Ifthe distinguishing criteria are sequential,they may even run through a completelogical tree in a microsecond or so. Thiselectronic preliminary analysis, likehuman preliminary analysis ofalternative outcomes, speeds up progressby sharpening the criteria. It eliminateshundreds of thousands of the irrelevantpictures that formerly had to be scanned,and when it is carried to its limit, a fewoutput pulses, hours apart, may beenough to signal the existence of theantiproton or the fall of a theory.

I think the emphasis on stronginference in the two fields I havementioned has been partly the result ofpersonal leadership, such as that of theclassical geneticists in molecularbiology, or of Szilard with his "MidwestChowder and Bacteria Society" atChicago in 1948-50, or of Max Delbruckwith his summer courses in phagegenetics at Cold Spring Harbor. But itis also partly due to the nature of thefields themselves. Biology, with its vastinformational detail and complexity, is a"high-information" field, where yearsand decades can easily be wasted on theusual type of "low-information"observations or experiments if one doesnot think carefully in advance aboutwhat the most important and conclusiveexperiments would be. And in high-energy physics, both the "informationflux" of particles from the newaccelerators and the million-dollar costsof operation have forced a similaranalytical approach. It pays to have atop-notch group debate every experimentahead of time; and the habit spreadsthroughout the field.

Induction and Multiple Hypothesis

Historically, I think there have beentwo main contributions to thedevelopment of a satisfactory strong-inference method. The first is that ofFrancis Bacon13. He wanted a "surermethod" of "finding out nature" thaneither the logic-chopping or all-inclusivetheories of the time or the laudable butcrude attempts to make inductions "bysimple enumeration." He did notmerely urge experiments, as somesuppose; he showed the fruitfulness ofinterconnecting theory and experimentso that the one checked the other. Ofthe many inductive procedures hesuggested, the most important, I think,was the conditional inductive tree, whichproceeded from alternative hypotheses(possible "causes," as he calls them),through crucial experiments ("Instancesof the Fingerpost"), to exclusions ofsome alternatives and adoption of whatis left ("establishing axioms"). HisInstances of the Fingerpost are explicitlyat the forks in the logical tree, the term


being borrowed "from the fingerpostswhich are set up where roads part, toindicate the several directions."

Many of his crucial experimentsproposed in Book II of The NewOrganon are still fascinating. Forexample, in order to decide whether theweight of a body is due to its "inherentnature," as some had said, or is due toattraction of the earth, which woulddecrease with distance, he proposescomparing the rate of a pendulum clockand a spring clock and then lifting themfrom the earth to the top of a tall steeple.He concludes that if the pendulum clockon the steeple "goes more slowly than itdid on account of the diminished virtueof its weights we may take theattraction of the mass of the earth as thecause of the weight."

Here was a method that couldseparate off the empty theories.

Bacon said the inductive methodcould be learned by anybody, just likelearning to "draw a straighter line ormore perfect circle with the help of aruler or a pair of compasses." "My wayof discovering sciences goes far to levelmen's wit and leaves but little toindividual excellence, because itperforms everything by the surest rulesand demonstrations." Even occasionalmistakes would not be fatal. "Truth willsooner come out from error than fromconfusion."

It is easy to see why young mindsleaped to try it.

Nevertheless there is a difficultywith this method. As Baconemphasizes, it is necessary to make"exclusions." He says, "The inductionwhich is to be available for the discoveryand demonstration of sciences and arts,must analyze nature by proper rejectionsand exclusions and then, after asufficient number of negatives, come toa conclusion on the affirmativeinstances. "[To man] it is granted only toproceed at first by negatives, and at lastto end in affirmatives after exclusion hasbeen exhausted."

Or, as the philosopher Karl Poppersays today, there is no such thing asproof in science -- because some lateralternative explanation may be as goodor better -- so that science advances only

by disproofs. There is no point inmaking hypotheses that are notfalsifiable, because such hypotheses donot say anything: "it must be possible foran empirical scientific system to berefuted by experience" 14.

The difficulty is that disproof is ahard doctrine. If you have a hypothesisand I have another hypothesis, evidentlyone of them must be eliminated. Thescientist seems to have no choice but tobe either soft-headed or disputatious.Perhaps this is why so many tend toresist the strong analytical approach --and why some great scientists are sodisputatious.

Fortunately, it seems to me, thisdifficulty can be removed by the use of asecond great intellectual invention, the"method of multiple hypotheses," whichis what was needed to round out theBaconian scheme. This is a method thatwas put forward by T. C. Chamberlin 15,a geologist at Chicago at the turn of thecentury who is best known for hiscontribution to the Chamberlin-Moultonhypothesis of the origin of the solarsystem.

Chamberlin says our trouble is thatwhen we make a single hypothesis, webecome attached to it.

"The moment one has offered anoriginal explanation for a phenomenonwhich seems satisfactory, that momentaffection for his intellectual child springsinto existence and as the explanationgrows into a definite theory his parentalaffections cluster about his offspring andit grows more and more dear to him.There springs up also unwittingly apressing of the theory to make it fit thefacts and a pressing of the facts to makethem fit the theory.

"To avoid this grave danger, themethod of multiple hypotheses is urged.It differs from the simple workinghypothesis in that it distributes the effortand divides the affections.

Each hypothesis suggests its owncriteria, its own means of proof, its ownmethod of developing the truth, and if agroup of hypotheses encompass thesubject on all sides, the total outcome ofmeans and of methods is full and rich.

Chamberlin thinks the method"leads to certain distinct habits of mind"

and is of prime value in education."When faithfully followed for asufficient time, it develops a mode ofthought of its own kind which may bedesignated the habit of complexthought.

This charming paper deserves to bereprinted in some more accessiblejournal today, where it could be requiredreading for every graduate student -- andfor every professor.

It seems to me that Chamberlin hashit on the explanation -- and the cure --for many of our problems in thesciences. The conflict and exclusion ofalternatives that is necessary to sharpinductive inference has been all too oftena conflict between men, each with hissingle Ruling Theory. But whenevereach man begins to have multipleworking hypotheses, it becomes purely aconflict between ideas. It becomesmuch easier then for each of us to aimevery day at conclusive disproofs -- atstrong inference -- without eitherreluctance or combativeness. In fact,when there are multiple hypotheseswhich are not anyone's "personalproperty" and when there are crucialexperiments to test them, the daily life inthe laboratory takes on an interest andexcitement it never had, and the studentscan hardly wait to get to work to see howthe detective story will come out. Itseems to me that this is the reason for thedevelopment of those "distinctive habitsof mind" and the "complex thought" thatChamberlin described, the reason for thesharpness, the excitement, the zeal, theteamwork -- yes, even internationalteamwork -- in molecular biology andhigh-energy physics today. What elsecould be so effective?

When multiple hypotheses becomecoupled to strong inference, the scientificsearch becomes an emotionalpowerhouse as well as an intellectualone.

Unfortunately, I think, there areother areas of science today that are sickby comparison, because they haveforgotten the necessity for alternativehypotheses and disproof. Each man hasonly one branch -- or none -- on thelogical tree; and it twist at randomwithout ever coming to the need for a


16 OCTOBER 1964

crucial decision at any point. We cansee from the external symptoms thatthere is something scientifically wrong.The Frozen Method. The EternalSurveyor. The Never Finished. TheGreat Man With A Single Hypothesis.The Little Club of Dependents. TheVendetta. The All-EncompassingTheory Which Can Never Be Falsified.

Some cynics tell a story, which maybe apocryphal, about the theoreticalchemist who explained to his class, "Andthus we see that the C-Cl bond is longerin the first compound than in the secondbecause the percent of ionic character issmaller."

A voice from the back of the roomsaid, "But Professor X, according to theTable, the C-Cl bond is shorter in thefirst compound."

"Oh, is it?" said the professor."Well, that's easy to understand, becausethe double-bond character is higher inthat compound."

To the extent that this kind of storyis accurate, a "theory" of this sort is not atheory at all, because it does not excludeanything. It predicts everything, andtherefore does not predict anything. Itbecomes simply a verbal formula whichthe graduate student repeats and believesbecause the professor has said it so often.This is not science but faith; not theory,but theology. Whether it is hand-waving or number-waving or equation-waving, a theory is not a theory unless itcan be disproved. That is, unless it canbe falsified by some possibleexperimental outcome.

In chemistry, the resonance theoristswill of course suppose that I amcriticizing them, while the molecular-orbital theorists will suppose I amcriticizing them. But their actions -- ouractions, for I include myself among them-- speak for themselves. A failure toagree for 30 years is publicadvertisement of a failure to disprove.

My purpose here, however, is not tocall names but rather to say that we areall sinners, and that in every field and inevery laboratory we need to try toformulate multiple alternativehypotheses sharp enough to be capableof disproof.

Systematic Application

I think the work methods of anumber of scientists have been testimonyto the power of strong inference. Issuccess not due in many cases tosystematic use of Bacon's "surest rulesand demonstrations" as much as to rareand unattainable intellectual power?Faraday's famous diary16, or Fermi'snotebooks3,17 show how these menbelieved in the effectiveness of dailysteps in applying formal inductivemethods to one problem after another.

Within 8 weeks after the discoveryof x-rays, Roentgen had identified 17 oftheir major properties. Every studentshould read his first paper18. Eachdemonstration in it is a little jewel ofinductive inference. How else could theproofs have gone so fast, except by amethod of maximum effectiveness?

Organic chemistry has been thespiritual home of strong inference fromthe beginning. Do the bonds alternate inbenzene or are they equivalent? If thefirst, there should be five disubstitutedderivatives; if the second, three. Andthree it is 19. This is a strong-inferencetest -- not a matter of measurement, ofwhether there are grams or milligrams ofthe products, but a matter of logicalalternatives. How else could thetetrahedral carbon atom or the hexagonalsymmetry of benzene have beeninferences could be confirmed by x-rayand infrared measurement?

We realize that it was out of thiskind of atmosphere that Pasteur came tothe field of biology. Can anyone doubtthat he brought with him a completelydifferent method of reasoning? Everytwo or three years he moved to onebiological problem after another, fromoptical activity to the fermentation ofbeet sugar to the "diseases" of wine andbeer, to the disease of silkworms, to theproblem of "spontaneous generation", tothe anthrax disease of sheep, to rabies.In each of these fields there were expertsin Europe who knew a hundred times asmuch as Pasteur, yet each time he solvedproblems in a few months that they hadnot been able to solve. Obviously it wasnot encyclopedic knowledge thatproduced his success, and obviously it

was not simply luck, when it wasrepeated over and over again; it can onlyhave been the systematic power of aspecial method of exploration. Arebacteria falling in? Make the necks ofthe flasks S-shaped. Are bacteriasucked in by the partial vacuum? Put ina cotton plug. Week after week hiscrucial experiments build up the logicaltree of exclusions. The drama of stronginference in molecular biology today isonly a repetition of Pasteur's story.

The grand scientific syntheses, likethose of Newton and Maxwell, are rareand individual achievements that standoutside any rule or method.Nevertheless it is interesting to note thatseveral of the great synthesizers havealso shown the strong-inference habit ofthought in their other work, as Newtondid in the inductive proofs of his Opticksand Maxwell did in his experimentalproof that three and only three colors areneeded in color vision.

A Yardstick of Effectiveness

I think the evident effectiveness ofthe systematic use of strong inferencesuddenly gives us a yardstick forthinking about the effectiveness ofscientific methods in general. Surveys,taxonomy, design of equipment,systematic measurements and tables,theoretical computations -- all have theirproper and honored place, provided theyare parts of a chain of precise inductionof how nature works. Unfortunately, alltoo often they become ends inthemselves, mere time-serving from thepoint of view of real scientific advance, ahypertrophied methodology that justifiesitself as a lore of respectability.

We praise the "lifetime of study,"but in dozens of cases, in every field,what was needed was not a lifetime, butrather a few short months or weeks ofanalytical inductive inference. In anynew area we should try, like Roentgen,to see how fast we can pass from thegeneral survey to analytical inferences.We should try, like Pasteur, to seewhether we can reach strong inferencesthat encyclopedism could not discern.

We speak piously of takingmeasurements and making small studies


that will "add another brick to the templeof science." Most such bricks just liearound the brickyard 20. Tables ofconstants have their place and value, butthe study of one spectrum after another,if not frequently re-evaluated, maybecome a substitute for thinking, a sadwaste of intelligence in a researchlaboratory, and a mistraining whosecrippling effects may last a lifetime.

To paraphrase an old saying,Beware of the man of one method or oneinstrument, either experimental ortheoretical. He tends to becomemethod-oriented rather than problem-oriented. The method-oriented man isshackled: the problem-oriented man is atleast reaching freely toward what is mostimportant. Strong inference redirects aman to problem-orientation, but itrequires him to be willing repeatedly toput aside his last methods and teachhimself new ones.

On the other hand, I think thatanyone who asks the question aboutscientific effectiveness will alsoconclude that much of themathematicizing in physics andchemistry today is irrelevant if notmisleading.

The great value of mathematicalformulation is that when an experimentagrees with a calculation to five decimalplaces, a great many alternativehypotheses are pretty well excluded(though the Bohr theory and theSchrdinger theory both predict exactlythe same Rydberg constant!) But whenthe fit is only to two decimal places, ormaybe one, it man be no better than anyrule-of-thumb extrapolation, and someother kind of qualitative exclusion mightbe more rigorous for testing theassumptions and more important toscientific understanding than quantitativefit.

I know that this is like saying thatthe emperor has no clothes. Today wepreach that science is not science unlessit is quantitative. We substitutecorrelations for causal studies, andphysical equations for organic reasoning.Measurements and equations aresupposed to sharpen thinking, but, in myobservation, they more often tend tomake the thinking noncausal and fuzzy.

They tend to become the object ofscientific manipulation instead ofauxiliary tests or crucial inferences.

Many -- perhaps most -- of the greatissues of science are qualitative, notquantitative, even in physics andchemistry. Equations andmeasurements are useful when and onlywhen they are related to proof: but proofor disproof comes first and is in factstrongest when it is absolutelyconvincing without any quantitativemeasurement.

Or to say it another way, you cancatch phenomena in a logical box or in amathematical box. The logical box iscoarse but strong. The mathematicalbox is fine-grained but flimsy. Themathematical box is a beautiful way ofwrapping up a problem, but it will nothold the phenomena unless they havebeen caught in a logical box to beginwith.

What I am saying is that, innumerous areas that we call science, wehave come to like our habitual ways, andour studies that can be continuedindefinitely. We measure, we define,we compute, we analyze, but we do notexclude. And this is not the way to useour minds most effectively or to makethe fastest progress in solving scientificquestions.

Of course it is easy -- and all toocommon -- for one scientist to call theothers unscientific. My point is not thatmy particular conclusions here arenecessarily correct, but that we havelong needed some absolute standard ofpossible scientific effectiveness bywhich to measure how well we aresucceeding in various areas -- a standardthat many could agree on and one thatwould be undistorted by the scientificpressures and fashions of the times andthe vested interests and busywork thatthey develop. It is not public evaluationthat I am interested in so much as aprivate measure by which to compareone's own scientific performance withwhat it might be. I believe that stronginference provides this kind of standardof what the maximum possible scientificeffectiveness could be -- as well as arecipe for reaching it.

Aids to Strong Inference

How can we learn the method andteach it? It is not difficult. The mostimportant thing is to keep in mind thatthis kind of thinking is not a lucky knackbut a system that can be taught andlearned. The molecular biologists todayare living proof of it. The second thingis to be explicit and formal and regularabout it, to devote a half hour or an hourto analytical thinking every day, writingout the logical tree and the alternativesand crucial experiments in a permanentnotebook. I have discussed elsewhere3

the value of Fermi's notebook method,the effect it had on his colleagues andstudents, and the testimony that it "canbe adopted by anyone with profit."

It is true that it takes great courtesyto teach the method, especially to one'speers--or their students. The stronginference point of view is so resolutelycritical of methods of work and values inscience that any attempt to comparespecific cases is likely to sound bothsmug and destructive. Mainly one shouldtry to teach it by example and byexhorting to self-analysis and self-improvement only in general terms, as Iam doing here.

But I will mention one severe butuseful private test-- a touchstone ofstrong inference--that removes thenecessity for third person criticism,because it is a test that anyone can learnto carry with him for use as needed. It isour old friend the Baconian "exclusion"but I call it "The Question." Obviously itshould be applied as much to one's ownthinking as to others. It consists ofasking in your own mind, on hearing anyscientific explanation or theory putforward, "But sir, what experiment coulddisprove your hypothesis?" or on hearinga scientific experiment described, "Butsir, what hypothesis does yourexperiment disprove?"

This goes straight to the heart of thematter. It forces everyone to refocus onthe central question of whether there isor is not a testable scientific stepforward.

If such a question were asked aloud,many a supposedly great scientist wouldsputter and turn livid and would want to


16 OCTOBER 1964

throw the questioner out as a hostilewitness! Such a man is less than heappears, for he is obviously notaccustomed to think in terms ofalternative hypotheses and crucialexperiments for himself; and one mightalso wonder about the state of science inthe field he is in. But who knows?--thequestion might educate him, and his fieldtoo!

On the other hand, I think thatthroughout most of molecular biologyand nuclear physics the response to TheQuestion would be to outlineimmediately not one but several tests todisprove the hypothesis--and it wouldturn out that the speaker already had twoor three graduate students working onthem!

I almost think that governmentagencies could make use of this kind oftouchstone. It is not true that all scienceis equal, or that we cannot justlycompare the effectiveness of scientistsby any method other than a mutual-recommendation system. The man towatch, the man to put your money on, isnot the man who wants to make "asurvey" or a "more detailed study" butthe man with the notebook; the man withthe alternative hypotheses and the crucialexperiments, the man who knows how toanswer your Question of disproof and isalready working on it.

There are some really hardproblems, some high-informationproblems, ahead of us in several fields,problems of photosynthesis, of cellularorganization, of the molecular structureand organization of the nervous system,not to mention some of our social andinternational problems. It seems to methat the method of most rapid progress insuch complex areas, the most efectiveway of using our brains, is going to be toset down explicitly at each step just whatthe question is, and what all thealternatives are, and then to set upcrucial experiments to try to disprovesome. Problems of this complexity, ifthey can be solved at all, can be solvedonly by men generating and excludingpossibilities with maximumeffectiveness, to obtain a high degree ofinformation per unit time--men willingto work a little bit at thinking.

When whole groups of us begin toconcentrate like that, I believe we maysee the molecular-biology phenomenonrepeated over and over again, with orderof magnitude increase in the rate ofscientific understanding in almost everyfield. 1 A.M.Weinberg. Minerva 1963. 159(winter 1963): Phys. Today 17, 42(1964).2 2. G. Polya. Mathematics and PlausibleReasoning (Princeton Univ. Press,Princeton, N.J. 1954), vol. 1, Inductionand Analogy in Mathematics; vol. 2,Patterns of Plausible Inference.3 J.R. Platt. The Excitement of Science(Houghton Mifflin, Boston, 1962): seeespecially chapters 7 and 8.4 J.D. Watson and F.H.C. Crick. Nature171:737 (1953).5 M. Meselson and F. Stahl. Proc. Natl.Acad. Sci. U.S. 44:671 (1958).6 A. Rich, in Biophysical Science; AStudy Program. J.L. Oncley et al., Eds.(Wiley, New York, 1959), p. 191.7 S. Benzer, Proc. Natl. Acad. Sci. U.S.45:1607 (1959).8 J. Lederberg. Science 129, 1649 (1959).9 P.F. Davison, D. Freifelder, B.W.Holloway. J. Mol. Biol. 8, 1 (1964).10 R.L. Garwin, L.M. Lederman, M.Weinrich. Phys. Rev. 105, 1415 (1957).11 W.A.H. Rushton, personalcommunication.12 See G.F. Chew, M. Gell-Mann, A.H.Rosenfeld. Sci. Am. 210, 74 (Feb 1964);ibid. 210, 60 (Apr. 1964); ibid. 210, 54(June 1964).13 F. Bacon. The New Organon andRelated Writings (Liberal Arts Press,New York, 1960), especially pp. 98, 112,151, 156, 196.14 K.R. Popper. The Logic of ScientificDiscovery (Basic Books, New York,1959), p.41. A modified view is given byT.S. Kuhn, The Structure of ScientificRevolutions (Univ of Chicago Press,Chicago, 1962), p.146; it does not, Ibelieve, invalidate any of theseconclusions.15 T.C. Chamberlin, J. Geol. 5, 837(1897). I am indebted to ProfessorsPreston Cloud and Bryce Crawford Jr.,of the University of Minnesota for

correspondence on this article and aclassroom reprint of it.16 M. Faraday, Faraday's Diary 1820-62(Bell, London, 1932-36).

17 H.L. Anderson and S.K. Allison, Rev.Mod. Phys. 27, 273 (1955).18 E.C. Watson (Am. J. Phys. 13, 281(1945) gives an English translation ofboth of Roentgen's first papers on x-rays.19 See G.W. Whelan, Advanced OrganicChemistry (Wiley, New York, 1949),chapter 4, for numerous such examples.20 B.K. Forscher, Science 142, 339(1963).

Documents

Platt