CHAPTER 29 Scientific Thinking and Reasoningdunbarlab/pubpdfs... · 2010-05-17 · CHAPTER 29 Scientific Thinking and Reasoning Kevin Dunbar Jonathan Fugelsang What is Scientific

P1 : GFZ0521824176c29.xml CB798B/Holyoak 0 521 82417 6 October 31 , 2004 18:5

C H A P T E R 2 9

Scientific Thinking and Reasoning

Kevin DunbarJonathan Fugelsang

What is Scientific Thinkingand Reasoning?

Scientific thinking refers to the mentalprocesses used when reasoning about thecontent of science (e.g., force in physics),engaged in typical scientific activities (e.g.,designing experiments), or specific types ofreasoning that are frequently used in sci-ence (e.g., deducing that there is a planetbeyond Pluto). Scientific thinking involvesmany general-purpose cognitive operationsthat human beings apply in nonscientific do-mains such as induction, deduction, anal-ogy, problem solving, and causal reason-ing. These cognitive processes are coveredin many chapters of this handbook (seeSloman & Lagnado, Chap. 5 on induction;Holyoak, Chap. 6 on analogy; Buehner andCheng, Chap. 7 on causality; Evans, Chap.8 on deduction; Novick and Bassok, Chap.1 4 on problem solving; Chi and Ohllson,Chap. 16 on conceptual change). What dis-tinguishes research on scientific thinkingfrom general research on cognition is thatresearch on scientific thinking typically in-

volves investigating thinking that has scien-tific content. A number of overlapping re-search traditions have been used to investi-gate scientific thinking. We cover the historyof research on scientific thinking and the dif-ferent approaches that have been used, high-lighting common themes that have emergedover the past fifty years of research.

A Brief History of Researchon Scientific Thinking

Science is often considered one of the hall-marks of the human species, along withart, music, and literature. Illuminating thethought processes used in science there-fore reveals key aspects of the human mind.The thought processes underlying scientificthinking have fascinated both scientists andnonscientists because the products of sci-ence have transformed our world and be-cause the process of discovery is shroudedin mystery. Scientists talk of the chance dis-covery, the flash of insight, the years of per-spiration, and the voyage of discovery. These

705


706 the cambridge handbook of thinking and reasoning

images of science have helped make themental processes underlying the discoveryprocess intriguing to cognitive scientists asthey attempt to uncover what really goeson inside the scientific mind and how sci-entists really think. Furthermore, the ques-tions, “Can scientists be taught to think bet-ter, avoiding mistakes of scientific thinking?”and “Could the scientific process be auto-mated such that scientists are no longer nec-essary?” make scientific thinking a topic ofenduring interest. One of the most com-pelling accounts of science that makes thereader want to understand science and whyscience is interesting recently appeared inthe journal Popular Science. In this article,Charles Hirshberg discusses his mother, sci-entist Joan Feynman, and her scientific con-tributions as well as the difficulties of being awoman scientist. The following excerpt cap-tures the excitement and thrill that even ahousehold encounter with science can gen-erate and that is thought to be at the rootof many scientists’ desire to conduct science(Hirschberg, 2003).

My introduction to chemistry came in1 970, on a day when my mom was bak-ing challah bread for the Jewish New Year.I was about ten, and though I felt cookingwas unmanly for a guy who played short-stop for Village Host Pizza in the MenloPark, California, Little League, she hadpersuaded me to help. When the bread wasin the oven, she gave me a plastic pill bot-tle and a cork. She told me to sprinkle alittle baking soda into the bottle, then a lit-tle vinegar, and cork the bottle as fast asI could. There followed a violent and com-pletely unexpected pop as the cork flew offand walloped me in the forehead. Explod-ing food: I was ecstatic! “That’s called achemical reaction,” she said, rubbing myshirt clean. “The vinegar is an acid and thesoda is a base, and that’s what happenswhen you mix the two.” After that, I neverunderstood what other kids meant whenthey said that science was boring.

The cognitive processes underlying sci-entific discovery and day-to-day scientificthinking have been a topic of intensescrutiny and speculation for almost 400

years (e.g., Bacon, 1620; Galilei, 1638; Klahr,2000; Tweney, Doherty, & Mynatt, 1981 ).Understanding the nature of scientific think-ing has been an important and central issuenot only for our understanding of science,but also for our understating of what it is tobe human. Bacon’s Novumm Organum, in1620, sketched out some of the key featuresof the ways that experiments are designedand data interpreted. Over the ensuing 400

years, philosophers and scientists vigorouslydebated the appropriate methods that scien-tists should use (see Giere, 1993). These de-bates over the appropriate methods for sci-ence typically resulted in the espousal of aparticular type of reasoning method such asinduction or deduction. It was not until theGestalt psychologists began working on thenature of human problem solving, duringthe 1940s, that experimental psychologistsbegan to investigate the cognitive processesunderlying scientific thinking and reasoning.

The Gestalt Psychologist Max Werthei-mer initiated the first investigations of sci-entific thinking in his landmark book, Pro-ductive Thinking (Wertheimer, 1945 ; seeNovick & Bassok, Chap. 1 4). Wertheimerspent a considerable amount of time corre-sponding with Albert Einstein (Figure 29.1 ),attempting to discover how Einstein gener-ated the concept of relativity. Wertheimerargued that Einstein had to overcome thestructure of Newtonian physics at each stepin his theorizing and the ways that Einsteinactually achieved this restructuring were ar-ticulated in terms of Gestalt theories. For arecent and different account of how Einsteinmade his discovery see Galison (2003). Wewill see later how this process of overcomingalternative theories is an obstacle with whichboth scientists and nonscientists need todeal when evaluating and theorizing aboutthe world.

One of the first investigations of the cog-nitive abilities underlying scientific think-ing was the work of Jerome Bruner and hiscolleagues at Harvard (Bruner, Goodnow, &Austin, 1956). They argued that a key ac-tivity in which scientists engage is to deter-mine whether or not a particular instanceis a member of a category. For example, a


scientific thinking and reasoning 707

scientist might want to discover which sub-stances undergo fission when bombarded byneutrons and which substances do not. Here,scientists have to discover the attributes thatmake a substance undergo fission. Bruneret al. (1956) saw scientific thinking as thetesting of hypotheses and collecting of datawith the end goal of determining whethersomething is a member of a category or not.They invented a paradigm in which peoplewere required to formulate hypotheses andcollect data that tests their hypotheses. Us-ing this approach, Bruner et al. identified anumber of strategies people use to formu-late and test hypotheses. They found thata key factor determining which hypothesistesting strategy people use is the amount ofmemory capacity the strategy takes up (seealso Morrison, Chap. 19. on working mem-ory). Another key factor they discovered wasthat it was much more difficult for people todiscover negative concepts (e.g., not blue)than positive concepts (e.g., blue). Althoughthe Bruner et al. research is most com-monly thought of as work on concepts, theysaw their work as uncovering a key compo-nent of scientific thinking.

A second early line of research on scien-tific thinking was developed by Peter Wa-son and his colleagues. Like Bruner et al.,Wason (1968) saw a key component of sci-entific thinking as being the testing of hy-potheses. Whereas Bruner et al. focused onthe different types of strategies people useto formulate hypotheses, Wason focused onwhether people adopt a strategy of tryingto confirm or disconfirm their hypotheses.Using Popper’s (1959) theory that scien-tists should try and falsify rather than con-firm their hypotheses, Wason devised a de-ceptively simple task in which participantswere given three numbers, such as 2-4-6,and were asked to discover the rule under-lying the three numbers. Participants wereasked to generate other triads of numbersand the experimenter would tell the partic-ipant whether the triad was consistent orinconsistent with the rule. They were toldthat when they were sure they knew whatthe rule was they should state it. Most par-ticipants began the experiment by thinking

that the rule was even numbers increasing bytwo. They then attempted to confirm theirhypothesis by generating a triad like 8-10-1 2 , then 1 4-16-1 8. These triads are consis-tent with the rule and the participants weretold yes, that the triads were indeed con-sistent with the rule. However, when theyproposed the rule, even numbers increas-ing by two, they were told that the rulewas incorrect. The correct rule was num-bers of increasing magnitude. From this re-search Wason concluded that people try andconfirm their hypotheses, whereas norma-tively speaking, they should try and discon-firm their hypotheses. One implication ofthis research is that confirmation bias is notjust restricted to scientists, but is a generalhuman tendency.

It was not until the 1970s that a generalaccount of scientific reasoning was proposed.Herbert Simon, often in collaboration withAllan Newell (e.g., Newell & Simon, 1972),proposed that scientific thinking is a formof problem solving. He proposed that prob-lem solving is a search in a problem space.Newell and Simon’s (1972) theory of prob-lem solving is discussed in many places inthis Volume, usually in the context of spe-cific problems (see especially Novick & Bas-sok, Chap. 1 4 , on problem solving). HerbertSimon (1977), however, devoted consider-able time to understanding many differentscientific discoveries and scientific reason-ing processes. The common thread in his re-search was that scientific thinking and dis-covery is not a mysterious magical process,but a process of problem solving in whichclear heuristics are used. Simon’s goal was toarticulate the heuristics that scientists use intheir research at a fine-grained level. He builtmany programs that simulated the process ofscientific discovery and articulated the spe-cific computations that scientists use in theirresearch (see subsequent section on compu-tational approaches to scientific thinking).Particularly important was Simon and Lea’s(1974) work demonstrating that concept for-mation and induction consist of a search intwo problem spaces; a space of instances anda space of rules. This idea has been highlyinfluential on problem solving accounts of


708 the cambridge handbook of thinking and reasoning

scientific thinking that will be discussed inthe next section.

Overall, the work of Bruner, Wason, andSimon laid the foundations for contempo-rary research on scientific thinking. Earlyresearch on scientific thinking is conve-niently summarized in Tweney, Doherty,and Mynatt’s 1981 book, On Scientific Think-ing, in which they sketched out many of thethemes that have dominated research on sci-entific thinking over the past few decades.Other, more recent books, such as RonaldGiere’s Cognitive models of Science (1993);David Klahr’s Explaining Science (2000); Pe-ter Carruthers, Steven Stich, and MichaelSiegal’s Cognitive Basis of Science (2002); andGorman and colleagues’ New Directions inScientific and Technical Thinking (2004) pro-vide detailed analyses of different aspects ofscientific discovery. In this chapter, we dis-cuss the main approaches that have beenused to investigate scientific thinking.

One of the main features of investigationsof research on the scientific mind has beento take one aspect of scientific thinking thatis thought to be important and investigateit in isolation. How does one go about in-vestigating the many different aspects of sci-entific thinking? Numerous methodologieshave been used to analyze the genesis of sci-entific concepts, theories, hypotheses, andexperiments. Researchers have used experi-ments, verbal protocols, computer programs,and analysis of particular scientific discover-ies. A recent development has been to inves-tigate scientists as they reason “live” (in vivostudies of scientific thinking) in their ownlaboratories (Dunbar, 1995 , 2002). From a“Thinking and Reasoning” standpoint, themajor aspects of scientific thinking that havebeen most actively investigated are prob-lem solving, analogical reasoning, hypothe-sis testing, conceptual change, collaborativereasoning, inductive reasoning, and deduc-tive reasoning.

Scientific Thinking as Problem Solving

One important goal for accounts of scien-tific thinking has been to provide an over-

arching framework to understand the scien-tific mind. One framework that has had agreat influence in cognitive science is thatscientific thinking and scientific discoverycan be conceived as a form of problem solv-ing. Simon (1977) argued that both scientificthinking in general and problem solving inparticular could be thought of as a search ina problem space (see Chapter 1 1 ). A prob-lem space consists of all the possible statesof a problem and all the operations that aproblem solver can use to get from one stateto the next (see problem solving entry). Ac-cording to this view, by characterizing thetypes of representations and procedures peo-ple use to get from one state to another, itis possible to understand scientific thinking.Scientific thinking therefore can be charac-terized as a search in various problem spaces(Simon, 1977). Simon investigated a num-ber of scientific discoveries by bringing par-ticipants into the laboratory, providing theparticipants with the data to which a sci-entist had access, and getting the partici-pants to reason about the data and rediscovera scientific concept. He then analyzed theverbal protocols participants generated andmapped out the types of problem spaces inwhich the participants search (e.g., Qin &Simon, 1990). Kulkarni and Simon (1988)used a more historical approach to uncoverthe problem-solving heuristics that Krebsused in his discovery of the urea cycle. Kulka-rni and Simon analyzed Krebs’s diaries andproposed a set of problem-solving heuristicsthat he used in his research. They then built acomputer program incorporating the heuris-tics and biological knowledge that Krebs hadbefore he made his discoveries. Of particularimportance are the search heuristics the pro-gram uses, such as the experimental proposalheuristics and the data interpretation heuris-tics built into the program. A key heuristicwas an unusualness heuristic that focusedon unusual findings and guided the searchthrough a space of theories and a space ofexperiments.

Klahr and Dunbar (1988) extended thesearch in a problem space approach and pro-posed that scientific thinking can be thoughtof as a search through two related spaces –an hypothesis space, and an experiment


scientific thinking and reasoning 709

space. Each problem space that a scientistuses will have its own types of representa-tions and operators used to change the rep-resentations. Search in the hypothesis spaceconstrains search in the experiment space.Klahr and Dunbar found that some partic-ipants move from the hypothesis space tothe experiment space, whereas others movefrom the experiment space to the hypothesisspace. These different types of searches leadto the proposal of different types of hypothe-ses and experiments. More recent workhas extended the dual-space approach toinclude alternative problem-solving spaces,including those for data, instrumentation,and domain-specific knowledge (Schunn &Klahr, 1995 , 1996; Klahr & Simon, 1999).

Scientific Thinking as HypothesisTesting

Many researchers have regarded testing spe-cific hypotheses predicted by theories as oneof the key attributes of scientific thinking.Hypothesis testing is the process of evalu-ating a proposition by collecting evidenceregarding its truth. Experimental cognitiveresearch on scientific thinking that specifi-cally examines this issue has tended to fallinto two broad classes of investigations. Thefirst class is concerned with the types ofreasoning that lead scientists astray, block-ing scientific ingenuity (see also Sternberg,Chapter 1 5 on creativity). A large amountof research has been conducted on the po-tentially faulty reasoning strategies that bothparticipants in experiments and scientistsuse, such as considering only one favoredhypothesis at a time and how this preventsscientists from making discoveries. The sec-ond class is concerned with uncovering themental processes underlying the generationof new scientific hypotheses and concepts.This research has tended to focus on the useof analogy and imagery in science as well asthe use of specific types of problem-solvingheuristics (see also Holyoak, Chapter 6

on analogy).Turning first to investigations of what di-

minishes scientific creativity, philosophers,

historians, and experimental psychologistshave devoted a considerable amount of re-search to “confirmation bias.” This is wherescientists consider only one hypothesis (typ-ically the favored hypothesis) and ignorealternative hypotheses or other potentiallyrelevant hypotheses. This important phe-nomenon can distort the design of exper-iments, formulation of theories, and inter-pretation of data. Beginning with the workof Wason (1968) and as discussed previ-ously, researchers have repeatedly shownthat when participants are asked to designan experiment to test a hypothesis, they pre-dominantly design experiments they thinkwill yield results consistent with the hypoth-esis. Using the 2-4-6 task mentioned ear-lier, Klayman and Ha (1987) showed thatin situations in which one’s hypothesis islikely to be confirmed, seeking confirmationis a normatively incorrect strategy, whereaswhen the probability of confirming one’shypothesis is low, then attempting to con-firm ones hypothesis can be an appropri-ate strategy. Historical analyses by Tweney(1989), on the way that Faraday made hisdiscoveries, and experiments investigatingpeople testing hypotheses have revealed thatpeople use a confirm early–disconfirm latestrategy: When people initially generate orare given hypotheses, they try to gather evi-dence that is consistent with the hypothesis.Once enough evidence has been gathered,then people attempt to find the boundariesof their hypothesis and often try to discon-firm their hypotheses.

In an interesting variant on the confir-mation bias paradigm, Gorman (1989) hasshown that when participants are told thereis the possibility of error in the data theyreceive, they assume any data inconsistentwith their favored hypothesis is attributableto error. The possibility of error therefore in-sulates hypotheses against disconfirmation.This hypothesis has not been confirmed byother researchers (Penner & Klahr, 1996),but is an intriguing one that warrants furtherinvestigation.

Confirmation bias is very difficult to over-come. Even when participants are askedto consider alternate hypotheses, they of-ten fail to conduct experiments that could


71 0 the cambridge handbook of thinking and reasoning

potentially disconfirm their hypothesis.Tweney and his colleagues provide an excel-lent overview of this phenomenon in theirclassic monograph, “On Scientific Think-ing” (1981 ). The precise reasons for thistype of block are still widely debated. Re-searchers such as Michael Doherty have ar-gued that limitations in working memorymake it difficult for people to consider morethan one hypothesis. Consistent with thisview, Dunbar and Sussman (1995) showedthat when participants are asked to holdirrelevant items in working memory whiletesting hypotheses, participants are unableto switch hypotheses in the face of incon-sistent evidence (see also Morrison, Chap-ter 19 on working memory). Although lim-itations of working memory are involved inthe phenomenon of confirmation bias, evengroups of scientists can display confirmationbias. The recent controversies over cold fu-sion are an example of confirmation bias.Here, large groups of scientists had otherhypotheses available to explain their data,yet maintained their hypotheses in the faceof other, more standard alternative hypothe-ses. Mitroff (1974) provides some interestingexamples of scientists at the National Aero-nautical and Space Administration demon-strating confirmation bias that highlightsthe roles of commitment and motivation inthis process.

Causal Thinking in Science

Much of scientific thinking and scientifictheory building pertains to the developmentof causal models between variables of inter-est. For example, does smoking cause cancer,Prozac relieve depression, or aerosol spraydeplete the ozone layer? (See also Buehner &Cheng. Chap. 7, on causality.) Scientists andnonscientists alike are constantly bombardedwith statements regarding the causal rela-tionship between such variables. How doesone evaluate the status of such claims? Whatkinds of data are informative? How do sci-entists and nonscientists deal with data thatare inconsistent with their theory?

One important issue in the causal rea-soning literature that is directly relevant toscientific thinking is the extent to which sci-entists and nonscientists are governed by thesearch for causal mechanisms (i.e., the chainof events that lead from a cause to an effect)versus the search for statistical data (i.e., howoften variables co-occur). This dichotomycan be boiled down to the search for quali-tative versus quantitative information aboutthe paradigm the scientist is investigating.Researchers from a number of cognitive psy-chology laboratories have found that peo-ple prefer to gather more information aboutan underlying mechanism than covariationbetween a cause and an effect (e.g., Ahnet al., 1995). That is, the predominant strat-egy that students in scientific thinking simu-lations use is to gather as much informationas possible about how the objects under in-vestigation work rather than collecting largeamounts of quantitative data to determinewhether the observations hold across mul-tiple samples. These findings suggest that acentral component of scientific thinking maybe to formulate explicit mechanistic causalmodels of scientific events.

One place where causal reasoning hasbeen observed extensively is when scientistsobtain unexpected findings. Both historicaland naturalistic research has revealed thatreasoning causally about unexpected find-ings has a central role in science. Indeed,scientists themselves frequently state that afinding was attributable to chance or was un-expected. Given that claims of unexpectedfindings are such a frequent component ofscientists’ autobiographies and interviewsin the media, Dunbar (1995 , 1997, 1999;Dunbar & Fugelsang, 2004 ; Fugelsang et al.,2004) decided to investigate the ways thatscientists deal with unexpected findings. In1991–1992 Dunbar spent one year in threemolecular biology laboratories and one im-munology laboratory at a prestigious U.S.university. He used the weekly laboratorymeeting as a source of data on scientific dis-covery and scientific reasoning. (This type ofstudy, he has called InVivo cognition). Whenhe examined the types of findings the sci-entists made, he found that more than 50%


scientific thinking and reasoning 71 1

Figure 2 9.1 . Causal thinking in science. Potential mechanisms of humanimmunodeficiency virus integration into host DNA. The diagram shows twopotential causal mechanisms – cellular (left branch) and viral (right branch).

were unexpected and that these scientistshad evolved a number of important strate-gies for dealing with such findings. One clearstrategy was to reason causally about thefindings: Scientists attempted to build causalmodels of their unexpected findings. Thiscausal model building results in the exten-sive use of collaborative reasoning, analog-ical reasoning, and problem-solving heuris-tics (Dunbar, 1997; 2001 ).

Many of the key unexpected findingsthat scientists reasoned about in the InVivostudies of scientific thinking were inconsis-tent with the scientists’ pre-existing causalmodels. A laboratory equivalent of the bi-ology labs therefore was to create a situa-tion in which students obtained unexpectedfindings that were inconsistent with theirpre-existing theories. Dunbar and Fugelsang(2004 ; see also Fugelsang et al., 2004) ex-amined this issue by creating a scientificcausal thinking simulation in which exper-imental outcomes were either expected orunexpected. (Dunbar [1995] called this typeof study of people reasoning in a cognitivelaboratory InVitro Cognition). They foundthat students spent considerably more timereasoning about unexpected findings thanexpected findings. Second, when assessingthe overall degree to which their hypoth-

esis was supported or refuted, participantsspent the majority of their time consid-ering unexpected findings. An analysis ofparticipants’ verbal protocols indicates thatmuch of this extra time is spent formu-lating causal models for the unexpectedfindings.

Scientists are not merely the victims ofunexpected findings, but plan for unex-pected events to occur. An example of theways that scientists plan for unexpected con-tingencies in their day-to-day research isshown in Figure 29.1 . Figure 29.1 is an ex-ample of a diagram in which the scientist isbuilding causal models about the ways thathuman immunodeficiency virus (HIV) inte-grates itself into the host deoxyribonucleicacid (DNA) taken from a presentation ata lab meeting. The scientist proposes twomain causal mechanisms by which HIV in-tegrates into the host DNA. The main eventthat must occur is that gaps in the DNAmust be filled. In the left-hand branch ofDiagram 2 , he proposes a cellular mech-anism whereby cellular polymerase fills ingaps as the two sources of DNA integrate.In the right-hand branch, he proposes thatinstead of cellular mechanisms filling in thegaps, viral enzymes fill in the gap and jointhe two pieces of DNA. He then designs



an experiment to distinguish between thesetwo causal mechanisms. Clearly, visual anddiagrammatic reasoning is used here andis a useful way of representing differentcausal mechanisms (see also Tversky, Chap-ter 10 on visuospatial reasoning). In this case,the visual representations of different causalpaths are used to design an experiment andpredict possible results. Thus, causal reason-ing is a key component of the experimentaldesign process.

When designing experiments, scientistsknow that unexpected findings occur of-ten and have developed many strategies totake advantage of them (Baker & Dunbar,2000). Scientists build different causal mod-els of their experiments incorporating manyconditions and controls. These multipleconditions and controls allow unknownmechanisms to manifest themselves. Ratherthan being the victims of the unexpected,the scientists create opportunities for unex-pected events to occur, and once these eventsdo occur, they have causal models that al-low them to determine exactly where in thecausal chain their unexpected finding arose.The results of these InVivo and InVitro stud-ies all point to a more complex and nuancedaccount of how scientists and nonscientiststest and evaluate hypotheses.

The Roles of Inductive and DeductiveThinking in the Scientific Mind

One of the most basic characteristics of sci-ence is that scientists assume that the uni-verse that we live in follows predictablerules. Very few scientists in this centurywould refute the claim that the earth ro-tates around the sun, for example. Scien-tists reason from these rules using a varietyof different strategies to make new scien-tific discoveries. Two frequently used typesof reasoning strategies are inductive (seeSloman & Lagnado, Chap. 5) and deductivereasoning (see Evans, Chap. 8). In the caseof inductive reasoning, a scientist may ob-serve a series of events and try to discover a

rule that governs them. Once a rule is dis-covered, scientists can extrapolate from therule to formulate theories of the observedand yet to be observed phenomena. One ex-ample is using inductive reasoning in the dis-covery that a certain type of bacterium is acause of many ulcers (Thagard, 1999). In afascinating series of articles, Thagard docu-ments the reasoning processes that Marshalland Warren went through in proposing thisnovel hypothesis. One key reasoning pro-cess was the use of induction by generaliza-tion. Marshall and Warren noted that almostall patients with gastric enteritis had a spi-ral bacterium in their stomachs and formedthe generalization that this bacterium is thecause of many stomach ulcers. There are nu-merous other examples of induction by gen-eralization in science, such as Tycho Braheinduction about the motion of planets fromhis observations, Dalton’s use of induction inchemistry, and the discovery of prions as thesource of mad cow disease. Many theoriesof induction have used scientific discoveryand reasoning as examples of this importantreasoning process.

Another common type of inductive rea-soning is to map a feature of one memberof a category to another member of a cate-gory. This is called categorical induction. Thistype of induction projects a known prop-erty of one item onto another item from thesame category. Thus, knowing that the RousSarcoma virus is a retrovirus that uses RNArather than DNA, a biologist might assumethat another virus that is thought to be aretrovirus also uses RNA rather than DNA.Although research on this type of inductiontypically has not been discussed in accountsof scientific thinking, this type of inductionis common in science. For an important con-tribution to this literature see Smith, Shafir,and Osherson (1993), and for a review ofthis literature see Heit (2000).

Turning now to deductive thinking, manythinking processes to which scientists adherefollow traditional rules of deductive logic.These processes correspond to conditions inwhich a hypothesis may lead to, or is de-ducible to, a conclusion. Although they are



not always phrased in syllogistic form, de-ductive arguments can usually be phrased as“syllogisms,” or as brief, mathematical state-ments in which the premises lead to the con-clusion. Deductive reasoning is an extremelyimportant aspect of scientific thinking be-cause it underlies a large component of howscientists conduct their research. By lookingat many scientific discoveries, we can oftensee that deductive reasoning is at work. De-ductive reasoning statements all contain in-formation or rules that state an assumptionabout how the world works, and a conclu-sion that would necessarily follow from therule. A classic example, that is still receiv-ing much scientific investigation today, is thecase of Planet X. In the early twentieth cen-tury, Percival Lowell coined the term “PlanetX” when referring to any planet yet to be dis-covered. Around that time and continuing tothis day, based on rather large residual orbitalperturbations of Uranus and Neptune manyscientists are convinced there exists a yet tobe discovered planet in our solar system. Be-cause it is assumed as fact that only large ob-jects that possess a strong gravitational forcecan cause such perturbations, the search forsuch an object ensued. Given Pluto’s rathermeager stature, it has been dismissed as acandidate for these perturbations. We canapply these statements to deductive logicas follows:

Premise 1: The gravitational force of largeplanetary bodies causes perturbations in or-bits of planetary bodiesPremise 2: Uranus and Neptune have per-turbations in their orbitsConclusion: The gravitational force of alarge planetary body influences the orbitsof Uranus and Neptune

Of course, the soundness of the logical de-duction is completely dependent on theaccuracy of the premises. If the premisesare correct, then the conclusion willbe correct.

Inductive and deductive reasoning, evenby successful scientists, is not immune toerror. Two classes of errors commonly foundin deductive reasoning are context and con-

tent errors. A common context error thatpeople often make is to assume that con-ditional relationships are, in fact, bicondi-tional. The conditional statement “if some-one has AIDS then they also have HIV,”for example, does not necessarily imply that“if someone has HIV then they also haveAIDS.” This is a common error in deduc-tive reasoning that can result in logically in-correct conclusions being drawn. A commoncontent error people often make is to modifythe interpretation of a conclusion based onthe degree to which the conclusion is plau-sible. Here, scientists may be more likely toaccept a scientific discovery as valid if theoutcome is plausible. You can see how thissecond class of errors in deductive logic canhave profound implications for theory de-velopment. Indeed, if scientists are overlyblinded by the plausibility of an outcome,they may fail to objectively evaluate thesteps in their deductive process.

The Roles of Analogy in ScientificThinking

One of the most widely mentioned rea-soning processes used in science is analogy.Scientists use analogies to form a bridgebetween what they already know and whatthey are trying to explain, understand, or dis-cover. In fact, many scientists have claimedthat the use of certain analogies was instru-mental in their making a scientific discoveryand almost all scientific autobiographies andbiographies feature an important analogythat is discussed in depth. Coupled with thefact that there has been an enormous re-search program on analogical thinking andreasoning (see Holyoak, Chapter 6), we nowhave a number of models and theories of ana-logical reasoning – that show exactly howanalogy can play a role in scientific discovery(see Gentner, Holyoak, & Kokinov, 2001 ).By analyzing the use of analogies in sci-ence, Thagard and Croft (1999), Nersessian(1999), Gentner and Jeziorski (1993), andDunbar and Blanchette (2001 ) all have



shown that analogical reasoning is a key as-pect of scientific discovery.

Traditional accounts of analogy distin-guish between two components of analog-ical reasoning – the target and the source.The target is the concept or problem thata scientist is attempting to explain or solve.The source is another piece of knowledgethat the scientist uses to understand the tar-get, or to explain the target to others. Whatthe scientist does when he or she makes ananalogy is to map features of the source ontofeatures of the target. By mapping the fea-tures of the source onto the target, new fea-tures of the target may be discovered, or thefeatures of the target can be rearranged sothat a new concept is invented and a sci-entific discovery is made. A common anal-ogy used with computers is to describe aharmful piece of software as a computervirus. Once a piece of software is called avirus, people can map features of biologicalviruses, such as it is small, spreads easily, self-replicates using a host, and causes damage.Not only do people map a single feature ofthe source onto the target, but also the sys-tems of relations between features from thesource to the target. They also make analog-ical inferences. If a computer virus is simi-lar to a biological virus, for example, an im-mune system can be created on computersthat can protect computers from future vari-ants of a virus. One of the reasons scientificanalogy is so powerful is that it can gen-erate new knowledge such as the creationof a computational immune system havingmany of the features of a real biological im-mune system. This also leads to predictionsthat there will be newer computer virusesthat are the computational equivalent ofretroviruses, lacking DNA or standard in-structions, that will elude the computationalimmune system.

The process of making an analogy in-volves a number of key steps – retrieval ofa source from memory, aligning the featuresof the source with those of the target, map-ping features of the source onto those ofthe target, and possibly making of new infer-ences about the target. Scientific discoveriesare made when the source highlights a

hitherto unknown feature of the target orrestructures the target into a new set of rela-tions. Interestingly, research on analogy hasshown that participants do not easily useanalogy (see Gentner et al., 1997; Holyoak& Thagard, 1995). Participants tend to fo-cus on the sharing of a superficial featurebetween the source and the target, ratherthan the relations among features. In hisInVivo studies of science, Dunbar (1995 ,2001 , 2002) investigated the ways that sci-entists use analogies while they are conduct-ing their research and found that scientistsuse both relational and superficial featureswhen they make analogies. The choice ofwhether use superficial or relational featuresdepends on their goals. If their goal is to fixa problem in an experiment, their analogiesare based upon superficial features. If theirgoal is to formulate hypotheses, they focuson analogies based upon sets of relations.One important difference between scien-tists and participants in experiments is thatthe scientists have deep relational knowl-edge of the processes they are investigat-ing and can use that relational knowledge tomake analogies.

Analogies sometimes lead scientists andstudents astray. Evelyn Fox-Keller (1985)shows how an analogy between the pulsingof a lighthouse and the activity of the slimemold dictyostelium led researchers astray fora number of years. Likewise, the analogybetween the solar system (the source) andthe structure of the atom (the target) hasbeen shown to be potentially misleading tostudents taking more advanced courses inphysics or chemistry. The solar system anal-ogy has a number of misalignments to thestructure of the atom, such as electrons be-ing repelled rather than attracted, by eachother, and that electrons do not have individ-ual orbits like planets, but have orbit cloudsof electron density. Furthermore, studentshave serious misconceptions of the natureof the solar system, which can compoundtheir misunderstanding of the nature of theatom (Fischler & Lichtfield, 1992). Althoughanalogy is a powerful tool in science, as is thecase with all forms of induction, incorrectconclusions can be reached.



Conceptual Change in theScientific Mind

Many researchers have noted that an im-portant component of science is the gen-eration of new concepts and modificationof existing ones. Scientific concepts, like allconcepts, can be characterized as contain-ing representations of words, thoughts, ac-tions, objects, and processes. How does one’sknowledge of scientific concepts change overtime? The large-scale changes that occur inconceptual structures have been labeled con-ceptual change (see Chi & Ohlsson, Chap.16; Nersessian, 2002 ; Thagard, 1992). The-ories of conceptual change focus on twomain types of shifts. One is the additionof knowledge to a pre-existing conceptualstructure. Here, there is no conflict betweenthe pre-existing conceptual knowledge andthe new information the student is acquir-ing. Such minor conceptual shifts are rela-tively easy to acquire and do not demandrestructuring of the underlying representa-tions of scientific knowledge. The secondtype of conceptual shift is what is known as“radical conceptual change” (see Keil, 1999,and Nersessian, 1998, for reviews of this lit-erature). In this type of situation, it is nec-essary for a new conceptual system to beacquired that organizes knowledge in newways, adds new knowledge, and results ina very different conceptual structure. Thisradical conceptual change is thought to benecessary for acquiring many new conceptsin physics and is regarded as the major sourceof difficulty for students. The factors at theroot of this conceptual shift view have beendifficult to determine, although a number ofstudies in human development (Carey, 1985 ;Chi, 1992 ; Chi & Roscoe 2002), in the his-tory of science (Nersessian, 1998; Thagard,1992), and in physics education (Clement,1982 ; Mestre, 1991 ) give detailed accountsof the changes in knowledge representationthat occur when people switch from oneway of representing scientific knowledge toanother. A beautiful example of concep-tual change is shown in Figure 29.2 . This il-lustration is taken from the first edition of

Isaac Newton’s Fluxions (1 736). It displaysthe ancient Greeks looking on in amaze-ment at an English hunter who shoots at abird using Newton’s new method of fluxions.Clearly they had not undergone the concep-tual change needed to understand Newto-nian physics.

One area in which students show greatdifficulty in understanding scientific con-cepts is in physics. Analyses of studentschanging conceptions, using interviews, ver-bal protocols, and behavioral outcome mea-sures, indicate that large-scale changes instudents’ concepts occur in physics educa-tion (see McDermott and Redish 1999 fora review of this literature). Following Kuhn(1962), researchers have noted that studentschanging conceptions are similar to the se-quences of conceptual changes in physicsthat have occurred in the history of science.These notions of radical paradigm shifts andensuing incompatibility with past knowl-edge states have drawn interesting parallelsbetween the development of particular sci-entific concepts in children and in the historyof physics.

Investigations of naıve people’s under-standing of motion indicate that studentshave extensive misunderstandings of mo-tion. This naıve physics research indicatesthat many people hold erroneous be-liefs about motion similar to a medieval“Impetus” theory (McCloskey, Caramazza,& Green, 1980). Furthermore, students ap-pear to maintain “Impetus” notions even af-ter one or two courses in physics. In fact,some authors have noted that students whohave taken one or two courses in physicsmay perform worse on physics problemsthan naıve students (Mestre, 1991 ). It isonly after extensive learning that we seea conceptual shift from “Impetus” theo-ries of motion to Newtonian scientific the-ories. How one’s conceptual representationshifts from “naıve” to Newtonian is a mat-ter of contention because some have arguedthat the shift involves a radical conceptualchange, whereas others have argued that theconceptual change is not really complete.Kozhevnikov and Hegarty (2001 ) argue thatmuch of the naıve “Impetus” notions of



motion are maintained at the expense ofNewtonian principles even with extensivetraining in physics. They argue that such“Impetus” principles are maintained at animplicit level. Thus, although students cangive the correct Newtonian answer to prob-lems, their reaction times to respond indicatethey are also using impetus theories.

Although conceptual changes are thoughtto be large-scale changes in concepts thatoccur over extensive periods of time, it hasbeen possible to observe conceptual changeusing InVivo methodologies. Dunbar (1995)reported a major conceptual shift that oc-curred in immunologists, in which they ob-tained a series of unexpected findings thatforced the scientists to propose a new con-cept in immunology that, in turn, forced thechange in other concepts. The drive behindthis conceptual change was the discovery ofa series of different unexpected findings oranomalies that required the scientists to re-vise and reorganize their conceptual knowl-edge. Interestingly, this conceptual changewas achieved by a group of scientists reason-ing collaboratively, rather than by one scien-tist working alone. Different scientists tendto work on different aspects of concepts, andalso different concepts, that, when put to-gether, lead to a rapid change in entire con-ceptual structures.

Overall, accounts of conceptual changein individuals indicate it is, indeed, similarto that of conceptual change in entire scien-tific fields. Individuals need to be confrontedwith anomalies that their pre-existing theo-ries cannot explain before entire conceptualstructures are overthrown. However, re-placement conceptual structures have to begenerated before the old conceptual struc-ture can be discarded. Often, people donot overthrow their naıve conceptual the-ories and have misconceptions in many fun-damental scientific concepts that are main-tained across the lifespan.

The Scientific Brain

In this chapter, we have provided anoverview of research into the workings of the

scientific mind. In particular, we have shownhow the scientific mind possesses many cog-nitive tools that are applied differently de-pending on the task at hand. Research inthinking and reasoning has recently been ex-tended to include a systematic analysis of thebrain areas associated with scientific reason-ing using techniques such as functional mag-netic resonance imaging (fMRI), positronemission topography, and event related po-tentials. There are two main reasons fortaking this approach. First, these approachesallow the researcher to look at the en-tire human brain, making it possible to seethe many different sites involved in sci-entific thinking and to gain a more com-plete understanding of the entire range ofmechanisms involved in scientific think-ing. Second, these brain-imaging approachesallow researchers to address fundamentalquestions in research on scientific thinking.One important question concerns the extentto which ordinary thinking in nonscientificcontexts and scientific thinking recruit sim-ilar versus disparate neural structures of thebrain. Dunbar (2002) proposed that scien-tific thinking uses the same cognitive mech-anisms all human beings possess, rather thanbeing an entirely different type of thinking.He has proposed that in scientific thinking,standard cognitive processes are used, butare combined in ways that are specific to aparticular aspect of science or a specific dis-cipline of science. By comparing the resultsof brain imaging investigations of scientificthinking with brain imaging studies of non-scientific thinking, we can see both whetherand when common versus dissociated brainsites are invoked during different cognitivetasks. This approach will make it possible toarticulate more clearly what scientific think-ing is, and how it is both similar to and differ-ent from the nonscientific thinking typicallyexamined in the cognitive laboratory (alsosee Goel, Chap. 20).

Considering the large arsenal of cogni-tive tools researchers have at their disposal,determining the neurological underpinningof scientific thinking becomes mainly amatter of dissecting the processes thoughtto be involved in the reasoning process,



Figure 2 9.2 . Conceptual change in science: The ancient Greeks look on inamazement as a hunter uses Newtonian principles to shoot down a bird. Thisfigure is taken from the frontispiece of his Method of Fluxions and Infinite Series;with its Application to the Geometry of Curve Lines. Frontispiece in BodelianLibrary.

and conducting systematic experiments onthese subprocesses. What might these sub-processes be? As the previous sections ofthis chapter show, scientific thinking in-volves many cognitive capabilities including,but not limited to, analogical reason-ing, casual reasoning, induction, deduction,and problem solving: These subprocesses

undoubtedly possess common and distinctneural signatures. A number of cognitiveneuroscientists recently examined problemsolving (Fincham et al., 2002 ; Goel &Grafman, 1995 ; Colvin, Dunbar, & Graf-man, 2001 ), analogical reasoning (Whartonet al., 2000; Kroger et al., 2002), hypothe-sis testing (Fugelsang & Dunbar, submitted),



inductive reasoning (Goel & Dolan, 2000;Seger et al., 2000), and deductive reason-ing (Parsons & Osherson, 2001 ; Oshersonet al., 1998). They all pointed to the role ofthe dorsolateral prefrontal/parietal networkfor tasks requiring these higher level cogni-tive capacities. It is important to note thatthis brain network has been implicated intasks that are highly attention and working-memory demanding.

One question cognitive neuroscience in-vestigations of scientific thinking are be-ginning to address is the neurologicalunderpinnings of conceptual change. UsingfMRI to investigate students who have andwho have not undergone conceptual changein scientific areas, it is possible to uncover theneural changes that accompany conceptualchange. Fugelsang and Dunbar (submitted)have found shifts from ventral pathways todorsal pathways in the brain when studentsshift from naıve impetus theories of motionto Newtonian theories of motion. These cog-nitive neuroscience investigations reveal theways that knowledge is organized in the sci-entific brain and provide detailed accounts ofthe nature of the representation of scientificknowledge.

The extent to which these processes arelateralized in the right or left hemisphereis a matter of recent debate, especially asit pertains to inductive and deductive rea-soning. Hemispheric differences in scientificdeductive thinking potentially can be quiterevealing about the nature of the represen-tations of the scientific mind. For exam-ple, recent cognitive neuroscience researchcan provide important new insights intoone of the most fundamental questions thathave perplexed many scientists for decades –namely, whether complex scientific think-ing processes, such as deductive and induc-tive reasoning, are represented in terms oflinguistic or visual–spatial representations.Anecdotal claims are equivocal as to the na-ture of such representations. When think-ing about scientific concepts and devisingtheoretical explanations for phenomena, forexample, scientists may verbally representtheir theories in text or visually represent

theories in graphical models. More oftenthan not, scientific theories are representedin both modalities to some degree.

Based on what we know about hemi-spheric differences in the brain, there areseveral clear predictions about how spatialand verbal thinking styles would be repre-sented in the brain. If scientific thinking werepredominantly based on verbal or linguisticrepresentations, for example, we would ex-pect activations of the basic language neu-ral structures such as the frontal and inferiortemporal regions in the left hemisphere. Ifscientific thinking were predominately basedon visual-spatial representations, one wouldexpect activation of the basic perceptionand motor control neural structures suchas those found in the parietal and occipitallobes, particularly in the right hemisphere.To date, findings from research on this issuehave been quite mixed. Goel and colleagues(e.g., Goel et al., 1998; Goel Chap. 20) havefound significant activations for deductivereasoning to occur predominantly in the lefthemisphere. Parsons and Osherson (2001 )using a similar, but different, task of deduc-tive reasoning, found that such tasks recrui-ted resource predominantly from the righthemisphere.

Much research has been conducted to de-termine the cause of these different resultsand Goel (Chap. 20) provides a detailed ac-count of recent research on the brain anddeductive reasoning. One result regardinghemispheric differences important for stud-ies of scientific thinking is that of Roser et al.,(submitted). They conducted experimentalexaminations of hemispheric differences inscientific causal thinking in a split-brain pa-tient. They found that the patient’s righthemisphere was uniquely able to detectcausality in perceptually salient events (i.e.,colliding balls), whereas his left hemispherewas uniquely able to infer causality basedon a more complex, not directly perceivable,chain of events. These data add to our grow-ing understanding of how the brain containsspecialized neural structures that contributeto the interpretation of data obtained fromthe environment. The obvious experiments



that need to be done would involve allowingscientists to think and reason naturally abouttheir own theories versus theories from dif-ferent domains while being imaged. Thiswould allow one to decompose the effectsof scientific thinking and familiarity. Clearly,research on the scientific brain is aboutto begin.

Computational Approaches toScientific Thinking

Along with recent brain imaging studies,computational approaches have provided amore complete account of the scientificmind. Computational models provide spe-cific detailed accounts of the cognitive pro-cesses underlying scientific thinking. Earlycomputational work consisted of taking ascientific discovery and building computa-tional models of the reasoning processesinvolved in the discovery. Langley et al.(1987) built a series of programs that sim-ulated discoveries such as those of Coper-nicus and Stahl. These programs have vari-ous inductive reasoning algorithms built intothem and, when given the data the scientistsused, were able to propose the same rules.Computational models make it possible topropose detailed models of the cognitivesubcomponents of scientific thinking thatspecify exactly how scientific theories aregenerated, tested, and amended (see Darden1997; Shrager & Langley, 1990, for accountsof this branch of research). More recently,the incorporation of scientific knowledgeinto the computer programs resulted in ashift in emphasis from using programs tosimulate discoveries to building programsthat help scientists make discoveries. A num-ber of these computer programs have madenovel discoveries. For example, Valdes-Perez(1994) built systems for discoveries in chem-istry, and Fajtlowicz has done this in mathe-matics (Erdos, Fajtlowicz, & Staton, 1991 ).

These advances in the fields of computerdiscovery have led to new fields, confer-ences, journals, and even departments that

specialize in the development of programsdevised to search large databases in thehope of making new scientific discoveries(Langley, 2000, 2002). This process is com-monly known as “data mining.” Not until rel-atively recently has this technique proven vi-able because of recent advances in computertechnology. An even more recent develop-ment in the area of data mining is the useof distributed computer networks that takeadvantage of thousands, or even millions, ofcomputers worldwide to jointly mine datain the hope of making significant scientificdiscoveries. This approach has shown muchpromise because of its relative cost effec-tiveness. The most powerful supercomput-ers currently cost over 100 million dollars,where as a distributed network server maycost only tens of thousands of dollars forroughly the same computational power.

Another recent shift in the use of com-puters in scientific discovery is to have com-puters and people make discoveries together,rather than expecting computers to makean entire scientific discovery. Now, insteadof using computers to mimic the entire sci-entific discovery process used by humans,computers can use powerful algorithms thatsearch for patterns on large databases andprovide the patterns to humans who canthen use the output of these computers tomake discoveries from the human genometo the structure of the universe.

Scientific Thinking and ScienceEducation

Science education has undergone manychanges over the past hundred years thatmirrored wider changes in both educationand society. In the early 1900s, science edu-cation was seen as a form of nature study,particularly in the kindergarten througheight grades. Each decade has seen a re-port on the need to improve science edu-cation. Starting in the 1930s, proponents ofthe progressive education movement begana movement that continues to this day. They



argued that children should be taught morethan just facts; should be taught methodsand general principles, as well as ways inwhich science relate to the child’s world. In1938, a report by the Progressive EducationAssociation noted that the psychology of thelearner should be at the core of science edu-cation, as well as making a link to children’severyday lives. Various reports on science ed-ucation appeared over the ensuing years, butit was the launch of the Sputnik satellite in1957 that transformed science education inthe United States. Seeing the Soviets launcha rocket before the United States galvanizedthe nation into training better scientists andidentifying the brightest students. The netresult for science education was that text-books were updated, a factually based cur-riculum was maintained, and the notion ofscience as a voyage of discovery entered thepopular imagination. By the 1980s, however,many cultural changes had occurred and sci-ence students in the United States appearedto be falling behind those in other countries.Numerous reports by science teachers andscientists recommended widespread changesin the ways that science is taught. Most im-portant in these changes was the move to aconstructivist view of education. Accordingto this view, students construct their knowl-edge rather than being the passive recipientsof scientific knowledge (see also Ritchhart &Perkins, Chap. 32 , on teaching thinking).

Beginning in the 1980s, a number of re-ports, often constructivist, stressed the needfor teaching scientific thinking skills and notjust methods and content. The addition ofscientific thinking skills to the science cur-riculum from kindergarten through adult-hood was a major shift in focus. Many ofthe particular scientific thinking skills em-phasized were covered in previous sectionsof this chapter, such as deductive and induc-tive thinking strategies. Rather than focusingon one particular skill, such as induction, re-searchers in education have focused on howthe different components of scientific think-ing are put together in science. Furthermore,science educators have focused on situationsin which science is conducted collabora-tively, rather than being the product of one

person thinking alone. These changes in sci-ence education parallel changes in method-ologies used to investigate science, such asanalyzing the ways that scientists think andreason in their laboratories.

By looking at science as a complex, multi-layered, and group activity, many researchersin science education have adopted a con-structivist approach. This approach seeslearning as an active rather than a passiveprocess, and proposes that students learnthrough constructing their scientific knowl-edge. The goal of constructivist science edu-cation often is to produce conceptual changethrough guided instruction in which theteacher or professor acts as a guide to dis-covery rather than the keeper of all the facts.One recent and influential approach to sci-ence education is the inquiry-based learningapproach. Inquiry-based learning focuses onposing a problem or a puzzling event to stu-dents and asking them to propose a hypoth-esis that can be used to explain the event.Next, students are asked to collect data thattest the hypotheses, reach conclusions, andthen reflect upon both the original problemand the thought processes they used to solvethe problem. Students often use computersthat aid in their construction of new knowl-edge. The computers allow students to learnmany of the different components of scien-tific thinking. For example, Reiser and hiscolleagues have developed a learning envi-ronment for biology, where students are en-couraged to develop hypotheses in groups,codify the hypotheses, and search databasesto test them (Reiser et al., 2001 ).

One of the myths of science is the lonescientist toiling under a naked lightbulb,suddenly shouting “Eureka, I have made adiscovery.” Instead, InVivo studies of scien-tists (e.g., Dunbar, 1995 , 2002), historicalanalyses of scientific discoveries (Nersessian,1999), and InVivo studies of children learn-ing science at museums all point to collab-orative scientific discovery mechanisms asbeing one of the driving forces of science(Crowley et al., 2001 ). What happens duringcollaborative scientific thinking is that thereis usually a triggering event, such as an unex-pected result or situation that a student does



not understand. This results in other mem-bers of the group adding new informationto the person’s representation of knowledge,often adding new inductions and deduc-tions that both challenge and transformthe reasoner’s old representations of knowl-edge (Dunbar, 1998). This means that socialmechanisms play a key component in foster-ing changes in concepts that have been ig-nored in traditional cognitive research, butare crucial for both science and science edu-cation. In science education, there has been ashift to collaborative learning, particularly atthe elementary level, but in university edu-cation, the emphasis is still on the individualscientist. Because many domains of sciencenow involve collaborations across scientificdisciplines, we expect the explicit teach-ing of collaborative science heuristics toincrease.

What is the best way to teach andlearn science? Surprisingly, the answer tothis question has been difficult to un-cover. Although there has been consider-able research on the benefits of using aparticular way of learning science, few com-parative studies of different methods havebeen conducted. Following Seymour Pa-pert’s book MindStorms, for example, (1980)many schools moved to discovery learningin which children discover aspects of pro-gramming and mathematics through writ-ing their own computer programs in theLOGO programming language. This discov-ery learning approach, which thousands ofschools have adopted, has been presented asan alternative to more didactic approachesto teaching and learning. By allowing stu-dents to discover principles on their ownand to set their own goals, students are pur-ported to have deeper knowledge that trans-fers more appropriately. Although there ismuch anecdotal evidence on the benefits ofdiscovery learning, only recently has a di-rect comparison of discovery learning withmore traditional methods been conducted.Klahr and Nigam (2004) conducted a studyof third and fourth grade children learningabout experimental design. They found thatmany more children learned from direct in-struction than from discovery learning. Fur-

thermore, they found that discovery learn-ing children did not have richer or deeperknowledge than direct instruction children.This type of finding suggests that pure dis-covery learning, although intuitively appeal-ing, benefits only a few children and thatguided discovery coupled with explicit in-struction is one of the most effective educa-tional strategies in science.

Conclusions and Future Directions

Although much is known regarding certaincomponents of scientific thinking, much re-mains to be discovered. In particular, therehas been little contact among cognitive, neu-roscience, social, personality, and motiva-tional accounts of scientific thinking. Clearly,the relations among these different aspectsof scientific thinking need to be combinedto produce a comprehensive picture of thescientific mind. One way to achieve this isby using converging multiple methodolo-gies as outlined previously, such as natu-ralistic observation, controlled experimentsin the cognitive laboratory, and functionalbrain imaging techniques. Theoretical devel-opments into the workings of the scientificmind would greatly benefit from more un-constrained analyses of the neuroanatomicalcorrelates of the scientific reasoning process.We, as scientists, are beginning to get a rea-sonable grasp of the inner workings of thesubcomponents of the scientific mind (i.e.,problem solving, analogy, induction) and sci-entific thought. However, great advances re-main to be made concerning how these pro-cesses interact so scientific discoveries canbe made. Future research will focus on boththe collaborative aspects of scientific think-ing and the neural underpinnings of the sci-entific mind.

Acknowledgments

The authors would like to thank the fol-lowing organizations: Dartmouth College,McGill University, The Spencer Foundation,The National Science Foundation, and the



Engineering Research Council of Canada forfunding research discussed in this chapter.The comments of Keith Holyoak, Vimla Pa-tel, and an anonymous reviewer were allhelpful in making this a better chapter.

References

Ahn, W., Kalish, C. W., Medin, D. L., & Gelman,S. A. (1995). The role of covariation versusmechanism information in causal attribution,Cognition, 54 , 299–352 .

Azmitia, M. A. & Crowley, K. (2001 ). Therhythms of scientific thinking: A study of col-laboration in an earthquake microworld. In K.Crowley, C. Schunn, & T. Okada (Eds.) De-signing for science: Implications from everyday,classroom, and professional settings. Mawah,NJ: Erlbaum.

Bacon, F. (1620/1 854). Novum Organum. (B.Monatgue Trans.). Philadelphia: Parry &McMillan.

Baker, L. M., & Dunbar, K. (2000). Experimentaldesign heuristics for scientific discovery: Theuse of baseline and known controls. Interna-tional Journal of Human Computer Studies, 53 ,335–349.

Bruner, J. S., Goodnow, J. J., & Austin, G. A.(1956). A Study of Thinking. New York: NY Sci-ence Editions.

Carey, S. (1985). Conceptual Change in Child-hood. Cambridge, MA: MIT Press.

Carruthers, P., Stich, S, & Siegal, M. (2002). TheCognitive Basis of Science. New York, NY: Cam-bridge University Press.

Chi, M. (1992). Conceptual change within andacross ontological categories: Examples fromlearning and discovery in science. In R. Giere(Ed.), Cognitive Models of Science. Minneapolis,MN: University of Minnesota Press.

Chi, M. T. H., & Roscoe, R. D. (2002). The pro-cesses and challenges of conceptual change. InM. Limon, & L. Mason (Eds), ReconsideringConceptual Change: Issues in Theory and Prac-tice (pp 3–27). The Netherlands: Kluwer Aca-demic Publishers.

Clement, J. (1982). Students’ preconceptions inintroductory mechanics. American Journal ofPhysics, 50, 66–71 .

Colvin, M. K., Dunbar, K., & Grafman, J. (2001 ).The effects of frontal lobe lesions on goal

achievement in the water jug task. Journal ofCognitive Neuroscience, 1 3 , 1 1 29–1 1 47.

Darden, L. (1997). Strategies for discoveringmechanisms: Schema instantiation, modularsubassembly, forward chaining/backtracking.Proceedings of the 1 997 Biennial Meeting of thePhilosophy of Science Association.

Dunbar, K. (1995). How scientists really reason:Scientific reasoning in real-world laboratories.In R. J. Sternberg, & J. Davidson (Eds.), Mech-anisms of Insight. Cambridge, MA: MIT press.

Dunbar, K. (1997). How scientists think: Onlinecreativity and conceptual change in science.In T. B. Ward, S. M. Smith, & S. Vaid (Eds.),Conceptual Structures and Processes: Emergence,Discovery and Change. Washington, DC: APAPress.

Dunbar, K. (1998). Problem solving. In W.Bechtel, & G. Graham (Eds.). A companion toCognitive Science. London, England: Blackwell.

Dunbar, K. (1999). The scientist InVivo: How sci-entists think and reason in the laboratory. In L.Magnani, N. Nersessian, & P. Thagard (Eds.),Model-based Reasoning in Scientific Discovery.New York, NY: Plenum.

Dunbar, K. (2001 ). The analogical paradox: Whyanalogy is so easy in naturalistic settings, yetso difficult in the psychology laboratory. In D.Gentner, K. J. Holyoak, & B. Kokinov, Analogy:Perspectives from Cognitive Science. Cambridge,MA: MIT press.

Dunbar, K. & Blanchette, I. (2001 ). The in-vivo/invitro approach to cognition: the case ofanalogy. Trends in Cognitive Sciences, 5 , 334–339.

Dunbar, K. (2002). Science as category: Implica-tions of InVivo science for theories of cognitivedevelopment, scientific discovery, and the na-ture of science. In P. Caruthers, S. Stich, & M.Siegel (Eds.), Cognitive Models of Science. NewYork, NY: Cambridge University Press.

Dunbar, K., & Fugelsang, J. (2004). Causal think-ing in science: How scientists and students in-terpret the unexpected. To appear in M. E.Gorman, A. Kincannon, D. Gooding, & R. D.Tweney (Eds.), New Directions in Scientific andTechnical Thinking. Lawrence Erlbaum. Hills-dale, NJ.

Dunbar, K. & Sussman, D. (1995). Toward a cog-nitive account of frontal lobe function: Simu-lating frontal lobe deficits in normal subjects.Annals of the New York Academy of Sciences,769, 289–304 .



Erdos, P., Fajtlowicz, S., & Staton, W. (1991 ). De-gree sequences in the triangle-free graphs. Dis-crete Mathematics, 92 (91 ), 85–88.

Fincham, J. M., Carter, C. S., van Veen, V.,Stenger, V. A., & Anderson, J. R. (2002). Neu-ral mechanisms of planning: A computationalanalysis using event-related fMRI. Proclama-tions of the National Academy of Science, 99,3346–3351 .

Fischler, H., & Lichtfeldt, M. (1992). Modernphysics and students conceptions. InternationalJournal of Science Education, 1 4 , 1 81–190.

Fox-Keller, E. (1985). Reflections on Gender andScience. New Haven, CT: Yale University Press.

Fugelsang, J., & Dunbar, K. (submitted). How thebrain uses theory to interpret data.

Fugelsang, J., & Dunbar, K. (in preparation). Howthe brain learns physics.

Fugelsang, J., Stein, C., Green, A., & Dunbar, K.(2004). Theory and data interactions of the sci-entific mind: Evidence from the molecular andthe cognitive laboratory. Canadian Journal ofExperimental Psychology, 58, 1 32–1 41 .

Galilei, G. (1638/1991 ). A. de Salvio, & H. Crewtrans. Dialogues Concerning Two New Sciences.Amherst, NY: Prometheus Books.

Galison, P. (2003). Einstein’s Clocks, Poincare’sMaps: Empires of Time. New York, NY: W. W.Norton.

Gentner, D., Brem, S., Ferguson, R. W.,Markman, A. B., Levidow, B. B., Wolff, P., etal. (1997). Analogical reasoning and concep-tual change: A case study of Johannes Kepler.The Journal of the Learning Sciences, 6(1 ), 3–40.

Gentner, D., Holyoak, K. J., & Kokinov, B. (2001 ).The Analogical Mind: Perspectives from CognitiveScience. Cambridge, MA: MIT Press.

Gentner, D., & Jeziorski, M. (1993). The shiftfrom metaphor to analogy in western science.In A. Ortony (Ed.), Metaphor and Thought (2nded., pp. 447–480). Cambridge, England: Cam-bridge University Press.

Giere, R. (1993). Cognitive Models of Science. Min-neapolis, MN: University of Minnesota Press.

Goel, V., & Dolan, R. J. (2000). Anatomical seg-regation of component processes in an induc-tive inference task. Journal of Cognitive Neuro-science, 1 2 , 1 10–1 19.

Goel, V., Gold, B., Kapur, S., & Houle, S. (1998).Neuroanatomical correlates of human reason-ing. Journal of Cognitive Neuroscience, 1 0, 293–302 .

Goel, V., & Grafman, J. (1995). Are the frontallobes implicated in “planning” functions? In-terpreting data from the Tower of Hanoi. Neu-ropsychologia, 33 , 623–642 .

Gorman, M. E. (1989). Error, falsification andscientific inference: An experimental inves-tigation. Quarterly Journal of ExperimentalPsychology: Human Experimental Psychology,41 A, 385–41 2 .

Gorman, M. E., Kincannon, A., Gooding, D., &Tweney, R. D. (2004). New Directions in Sci-entific and Technical Thinking. Lawrence Erl-baum. Hillsdale, NJ.

Heit, E. (2000). Properties of inductive reasoning.Psychonomic Bulletin & Review, 7, 569–592 .

Hirschberg, C. (2003). My mother, the scientist.In R. Dawkins, & T. Folger (Eds.), The BestAmerican Science and Nature Writing 2 003 .New York: Houghton Mifflin.

Holyoak, K. J., & Thagard, P. (1995). MentalLeaps. Cambridge, MA: MIT Press.

Keil, F. C. (1999). Conceptual change. In R.Wilson., & F. Keil (Eds.) The MIT Encyclope-dia of Cognitive Science, Cambridge, MA: MITpress.

Klahr, D. (2000). Exploring Science: The Cognitionand Development of Discovery Processes. Cam-bridge, MA: MIT Press.

Klahr, D., & Dunbar, K. (1988). Dual space searchduring scientific reasoning. Cognitive Science,1 2 , 1–48.

Klahr, D., & Nigam, M. (2004). The equivalenceof learning paths in early science instruction:Effects of direct instruction and discoverylearning Psychological Science, 1 5 , 661–667.

Klahr, D., & Simon, H. (1999). Studies of sci-entific discovery: Complementary approachesand convergent findings. Psychological Bulletin,1 2 5 , 524–543 .

Klayman, J., & Ha, Y. (1987). Confirmation, dis-confirmation, and information in hypothesistesting. Psychological Review, 94 , 21 1–228.

Kozhevnikov, M., & Hegarty, M. (2001 ). Impe-tus beliefs as default heuristic: Dissociation be-tween explicit and implicit knowledge aboutmotion. Psychonomic Bulletin and Review, 8,439–453 .

Kroger, J. K., Sabb, F. W., Fales, C. L.,Bookheimer, S. Y., Cohen, M. S., & Holyoak, K.J. (2002). Recruitment of anterior dorsolateralprefrontal cortex in human reasoning: A para-metric study of relational complexity. CerebralCortex, 1 2 , 477–485 .



Kuhn, T. (1962). The Structure of Scientific Revolu-tions. Chicago, IL: University of Chicago Press.

Kulkarni, D., & Simon, H. A. (1988). The pro-cesses of scientific discovery: The strategy ofexperimentation. Cognitive Science, 1 2 , 1 39–1 76.

Langley, P. (2000). Computational support ofscientific discovery. International Journal ofHuman-Computer Studies, 53 , 393–410.

Langley, P. (2002). Lessons for the computationaldiscovery of scientific knowledge. In the Pro-ceedings of the First International Workshop onData Mining Lessons Learned.

Langley, P., Simon, H. A., Bradshaw, G. L., &Zytkow, J. M. (1987). Scientific Discovery: Com-putational Explorations of the Creative Processes.Cambridge, MA: MIT Press.

McCloskey, M., Caramazza, A., & Green, B.(1980). Curvilinear motion in the absence ofexternal forces: Naive beliefs about the motionof objects. Science, 2 1 0, 1 1 39–1 1 41 .

McDermott, L. C., & Redish, L. (1999). Researchletter on physics education research. AmericanJournal of Psychics, 67, 755 .

Mestre, J. P. (1991 ). Learning and instruction inpre-college physical science. Physics Today, 44 ,56–62 .

Mitroff, I. (1974). The Subjective Side of Science.Amsterdam. Elsevier.

Nersessian, N. (1998). Conceptual change. In W.Bechtel, & G. Graham (Eds.). A Companion toCognitive Science. London, England: Blackwell.

Nersessian, N. (1999). Models, mental mod-els, and representations: Model-based reason-ing in conceptual change. In L. Magnani, N.Nersessian, & P. Thagard (Eds.). Model-basedReasoning in Scientific Discovery. New York,NY: Plenum.

Nersessian, N. J. (2002). The cognitive basisof model-based reasoning in science In P.Carruthers, S. Stich, & M. Siegal (Eds). TheCognitive Basis of Science. New York, NY: Cam-bridge University Press.

Newton, I. (1 736). Method of Fluxions and InfiniteSeries; with its Application to the Geometry ofCurve Lines. Frontispiece in Bodleian Library.

Newell, A., & Simon, H. A. (1972). Human Prob-lem Solving. Oxford, England: Prentice-Hall.

Osherson, D., Perani, D., Cappa, S., Schnur, T.,Grassi, F., & Fazio, F. (1998). Distinct brainloci in deductive versus probabilistic reasoning.Neuropsychologia, 36, 369–376.

Papert, S. (1980). Mindstorms: Children Comput-ers and Powerful ideas. New York, N. Y.: BasicBooks Inc.

Parsons, L. M., & Osherson, D. (2001 ). New ev-idence for distinct right and left brain systemsfor deductive versus probabilistic reasoning.Cerebral Cortex, 1 1 , 954–965 .

Penner, D. E., & Klahr, D. (1996). When to trustthe data: Further investigations of system er-ror in a scientific reasoning task. Memory &Cognition, 24(5), 655–668.

Popper, K. R. (1959). The Logic of Scientific Dis-covery. London England: Hutchinson.

Qin, Y., & Simon, H. A. (1990). Laboratory repli-cation of scientific discovery processes. Cogni-tive Science, 1 4 , 281–31 2 .

Reiser, B. J., Tabak, I., Sandoval, W. A., Smith, B.,Steinmuller, F., & Leone, T. J., (2001 ). BGuILE:Stategic and conceptual scaffolds for scientificinquiry in biology classrooms. In S. M. Carver,& D. Klahr (Eds.). Cognition and Instruction:Twenty Five Years of Progress. Mahwah, NJ:Erlbaum.

Roser, M., Fugelsang, J., Dunbar, K., Corballis, P.,& Gazzaniga, M. (Submitted). Causality per-ception and inference in the split-brain.

Schunn, C. D., & Klahr, D. (1995). A 4-spacemodel of scientific discovery. In the Proceedingsof the 1 7th Annual Conference of the CognitiveScience Society. Pittsburgh, PA.

Schunn, C. D., & Klahr, D. (1996). The prob-lem of problem spaces: When and how to gobeyond a 2-space model of scientific discov-ery. Part of symposium on building a theory ofproblem solving and scientific discovery: Howbig is N in N-space search? In the Proceedingsof the 1 8th Annual Conference of the CognitiveScience Society. San Diego, CA.

Seger, C., Poldrack, R., Prabhakaran, V., Zhao,M., Glover, G., & Gabrieli, J. (2000). Hemi-spheric asymmetries and individual differencesin visual concept learning as measured by func-tional MRI. Neuropsychologia, 38, 1 3 16–1 324 .

Shrager, J, & Langley, P. (1990). ComputationalModels of Scientific Discovery and Theory For-mation. San Mateo, CA: Morgan Kaufmann.

Simon, H. A. (1977). Models of Discovery. Dor-drecht, The Netherlands: D. Reidel Publishing

Simon, H. A., & Lea, G. (1974). Problem solvingand rule induction. In H. Simon (Ed.), Modelsof Thought (pp. 329 – 346). New Haven, CT:Yale University Press.



Smith, E. E., Shafir, E., & Osherson, D. (1993).Similarity, plausibility, and judgments of prob-ability. Cognition. Special Issue: Reasoning andDecision Making, 49, 67–96.

Thagard, P. (1992). Conceptual Revolutions. Cam-bridge, MA: MIT Press.

Thagard, P. (1999). How Scientists ExplainDisease. Princeton, NJ: Princeton UniversityPress.

Thagard, P., & Croft, D. (1999). Scientific dis-covery and technological innovation: Ulcers,dinosaur extinction, and the programminglanguage Java. In L. Magnani, N. Nerses-sian, & P. Thagard (Eds.). Model-based Rea-soning in Scientific Discovery. New York, NY:Plenum.

Tweney, R. D. (1989). A framework for the cog-nitive psychology of science. In B. Gholson, A.Houts, R. A. Neimeyer, & W. Shadish (Eds.),Psychology of Science and Metascience. Cam-bridge, MA: Cambridge University Press.

Tweney, R. D., Doherty, M. E., & Mynatt, C. R.(1981 ). On Scientific Thinking. New York, NY:Columbia University Press.

Valdes-Perez, R. E. (1994). Conjecturing hiddenentities via simplicity and conservation laws:Machine discovery in chemistry. Artificial In-telligence, 65(2), 247–280.

Wason, P. C. (1968). Reasoning about a rule.Quarterly Journal of Experimental Psychology,20, 273–281 .

Wertheimer, M. (1945). Productive Thinking.New York: NY Harper.

Wharton, C., & Grafman, J. (1998). Reasoningand the human brain. Trends in Cognitive Sci-ence. 2 , 54–59.

Wharton, C. M., Grafman, J., Flitman, S. S.,Hansen, E. K., Brauner, J., Marks, A., etal. (2000). Toward neuroanatomical modelsof analogy: A positron emission tomographystudy of analogical mapping. Cognitive Psychol-ogy, 40, 1 73–197.


726

Documents

CHAPTER 29 Scientific Thinking and Reasoningdunbarlab/pubpdfs... · 2010-05-17 · CHAPTER 29 Scientific Thinking and Reasoning Kevin Dunbar Jonathan Fugelsang What is Scientific