Kepler's accounts of his discoveries were usually in the form of a narrative, following the twists and turns of his thought processes, pausing for a burst of rapture when he thought he had made a discovery. This form of writing made it very difficult for Newton and others to find the three laws of planetary motion in Kepler's work.
In contrast, Newton transformed a narrative account of his experiments on optics into a more logical, inductive account in order to publish it in the Transactions of the Royal Society in 1672. In 1666, he organized fifty experiments with a prism in a kind of exploratory format, with one experiment suggesting another. Newton's first draft of an article on this topic, submitted to the Transactions , rigorously followed BACON's maxim: "What the sciences stand in need of is a form of induction which shall analyze experience and take it to pieces, and by a due process of exclusion and rejection lead to an inevitable conclusion" (Newton quoted in Bazerman, 1988, p. 91). Newton wrote his article as if he had followed this method while he was conducting his experiments, carefully articulating, testing and rejecting hypotheses until he arrived at the only possible explanation.
In fact, had he written a Keplerian narrative of the research, the order in which the experiments were conducted and the reasons for moving from one to another would have been different. The inductive format allowed Newton to organize the experiments into a logical chain visible only in hindsight. Bazerman is careful to point out that this reorganization might not be a deliberate strategy; Newton's memory of experiments done six years earlier may have been gradually transformed to correspond with how, in hindsight, he felt the research ought to have been done, a common cognitive phenomenon (Neisser, 1982). But it is likely that the act of writing at least catalyzed this transformation.
Newton's article ended with an invention: the reflecting telescope, which he claimed to derive from his discovery that different colors were refracted differentially when they passed from one medium to another. Actually, such a derivation was unnecessary, since Cassegrain independently made the same invention. Furthermore, at least part of Newton's derivation was in error, though his invention was sound. Here we see the beginning of one of the classic myths: the best inventions are derived from scientific discoveries.
Bazerman goes on to describe the letters Newton wrote, strengthening his argument in response to criticisms. This led to a new style in the first book of his Opticks, which Bazerman calls a 'juggernaut of persuasion'. Newton's experiments now play a supporting role in an axiomatic framework. The audience's questions and doubts are anticipated and dealt with, so that no alternate interpretation seems possible. As Bazerman (Bazerman, 1988, p. 124) says, "in Book I of the Opticks, Newton powerfully grabs hold of our reason and experiences until we have seen exactly what he wants us to have seen, in both the concrete and cognitive senses of the word."
Although Bazerman is more concerned with the persuasive elements of Newton's discourse, it is clear that these revisions led Newton to a more coherent, theoretical understanding of his own discovery. Writing can also spark discovery. One of the best examples comes from Larry Holmes' work on Antoine Lavoisier and Hans Krebs. Holmes argued that "it is from finely detailed case studies of the investigations of highly creative scientists that we are most likely to reach eventually a clearer understanding of the general nature of creative imagination in science" (Holmes, 1985, xvii). He conducted a detailed study of the Antoine Lavoisier's work in (what we would now call) organic chemistry over a twenty-years period. One of the major insights to emerge from this work is that the act of writing is itself part of the discovery process.
The extent to which Lavoisier developed his thought while writing his memoirs suggests a function for scientific papers that is not often emphasized. Scientific papers are characterized in many different ways: as reports of completed research, as announcements of discoveries, as vehicles for knowledge claims, as the end products of a process of "inscription," as the prime manifestation of the "context of justification," and as the necessary prerequisite for recognition as a practicing scientist. It has become commonplace to point out that as historical accounts of the discoveries they report, published scientific papers are misleading. For the actual pathway of thought and experiment they substitute the best combination of argument and evidence that the author can muster to justify the conclusions he has already reached. When, however, we have been able through laboratory records to approximate more closely the real historical course, we can perceive the relation between that course and its representation in the published paper in a more positive light. Although a scientific paper is everything that is implied in the above labels, it is, or at least for Lavoisier it was, far more. He was not merely contriving idealized or distorted versions of investigations, of which true versions already existed. He was transforming open-ended clusters of ideas and operations into organized, bounded investigative units. Not until he had chosen what to include and to exclude, clarified, linked together the parts, and rationalized what he had done did a coherent, completed investigation exist. Sometimes, as we have seen, he very nearly created an investigation on paper by bringing together experiments that had formerly been part of other investigations. Producing his scientific papers was, in short, not a matter of reporting accurately or inaccurately on something he had previously done, but an integral part of the creative process (Holmes, 1985, p. 488).
For example, Holmes traced changes in Lavoisier's thinking across several drafts of a manuscript he eventually presented to the Academy of Sciences in May of 1777. In the early drafts, he hypothesized that the portion of the air absorbed in respiration (what we now call Oxygen) gives a red color to the blood, just as this same portion of air gave a red color to metals with which it combined (producing what we now call rust). Laviosier wrote happily, "I believe that the theory of respiration has now been established" (Lavoisier quoted in Holmes, 1987, p. 223).
By the next draft, he was already tempering this conclusion and in subsequent drafts he scribbled and worked over an alternative hypothesis: that this new kind of air (which he called dephlogisticated air and we now call Oxygen) was converted to fixed air in the lungs. It was this hypothesis that became one of his most important discoveries. Holmes is careful to note that "We cannot always tell whether a thought that led him to modify a passage, recast an argument, or develop an alternative interpretation occurred while he was still engaged in writing what he subsequently altered, or immediately afterward, or after some interval during which he occupied himself with something else; but the timing is, I believe, less significant than the fact that the new developments were consequences of the effort to express ideas and marshal supporting information on paper" (Holmes, 1987, p. 225).
Holmes also conducted a detailed study of Hans Krebs' discovery of how urea is synthesized from ammonia and carbon dioxide, a closed circle of reactions referred to as the ornithine cycle. When Krebs began the experiments that led to this discovery, he was not proceeding according to the hypothetico-deductive method. After finishing medical training in 1925, he worked as a research assistant for the distinguished biochemist Otto Warburg, who "had developed methods for measuring, with sensitive manometers, the rates of respiration of thin slices of tissue places in a fluid medium" (Holmes, 1989, p. 60). In 1931, Krebs began his own research program and looked for problems he could solve with his new tools. Within nine months, he had discovered the ornithine cycle, a process that has been emulated by a computer program called KEKADA (Kulkarni, 1988). While BACON focused on discovery processes occur after data has already been gathered, KEKADA attempted to simulate the process by which Krebs generated new data. The program could not actually conduct an experiment, but it could propose one, and the experimenter could provide the result.
KEKADA was programmed with a wide range of heuristics. Some of the heuristics were ones Krebs himself mentioned, including "a standard biochemical strategy: if a given compound exerts some particular action, check whether derivatives of that compound have similar actions" (Kulkarni, 1988, p. 146). Most were inferred from Holmes' detailed accounts of his processes. Kulkarni & Simon arranged these heuristics in a hierarchy, from those that were general and could be used across a wide variety of problem domains and those that were restricted to biochemistry. They also classified them into nine types, including 'problem choosers', 'experiment-proposers', expectation-setters' and 'hypothesis-modifiers'. The two classifications of heuristic were potentially independent, but in fact, the problem choosers identified by Kulkarni and Simon were all general and the hypothesis-modifiers were mostly domain-specific.
The 'standard biochemical strategy' noted by Krebs was classified as domain-specific, and a modified version of it was included in a hypothesis-generating heuristic:
If a surprising outcome occurs involving A as one of the reactants, then hypothesize that there is a class of substances containing A (or its derivatives) that will produce the same outcome (Kulkarni, 1988, p. 156).
An example of a general or weak heuristic from the problem generator category is:
If the outcome of an experiment violates expectations for it, then make the study of this puzzling phenomenon a task and add it to the agenda (Kulkarni, 1988, p. 153).
This problem generator can trigger the hypothesis generator noted above. KEKADA was also equipped with background knowledge about a variety of substances and their expected reactions. On each cycle, the conditions of each of KEKADA's heuristics were matched against working memory. When a heuristic was selected and ran, it altered the contents of working memory, and another cycle began. Kulkarni & Simon also distinguished between those heuristics that determined the next hypothesis and those that determined the next experiment. Both of the heuristics noted as examples above searched the hypothesis space. These experiment proposers include a variety of specific heuristics to follow up on the class of substances containing A, depending on KEKADA's specific hypotheses about how the reaction works. For example, if A and B react to form C , then the experiment proposer suggests experiments on A and B separately and in combination. The results of these experiments would be supplied by the programmer.
One of the most interesting features of KEKADA is its capacity to follow-up on surprises generated by violations of the expectations it started with. The initial experiment with Ornithine produced a surprising amount of urea; Krebs dropped everything else to pursue this surprise. If BACON had been equipped with this capability, it might have been possible to have it start with circular orbits, then propose additional observations when its expectations were violated.
KEKADA clearly models more aspects of discovery than BACON, and illustrates the potential for heuristic-based analyses of scientific discovery. However, the program does not simulate two key aspects of Krebs' discovery: his acquisition and use of a tissue-slicing technique that became his 'secret weapon' (Kulkarni, 1988) and the role of writing. While discussing his 1932 paper reporting the discovery of the ornithine cycle, Krebs remarked that,
I spent a lot of time on writing, but usually while the work was still going on. And I find in general only when I try to write it up, then do I find the gaps. I cannot complete a piece of work and then sit down and write the paper (Holmes, 1987, p. 226).
Let us see how our four generalizations about discovery fare after this brief consideration of the cases of Lavoisier and Krebs.
1. Discovery depends on finding problem significant enough to be labeled an important achievement.
Lavoisier and Krebs both began with problems that were considered significant by major practitioners in their fields.
2. Discovery depends on transforming that problem into a form that suggests a promising path to solution.
Lavoisier had his own unique way of framing the problem in terms of "conceptual structures that were novel, deep and persistent, in the context of the state of the fields he entered" (Holmes, 1989, p. 63). Krebs was less conceptual and more empirically opportunistic--he followed surprising results. In the course of explaining them, he did have to formulate and test novel hypotheses. But the Krebs case suggests that a scientist who possesses good methodological techniques and problem-solving heuristics may be able to create opportunities for data-driven problem transformations.
Interestingly, both Lavoisier and Krebs at points held contradictory views of an important problem. At one stage, Lavoisier entertained both the idea that respiration worked (a) by absorbing one of two parts of air or (b) by removing 'fire matter' (heat) from the air. When exploring this new ornithine effect, Krebs had to grapple with the fact that ornithine seemed to be both catalyst and product of the reaction. Similarly, Kepler seemed to maintain for a time both the view that orbits had to be perfect circles and that they could not be circular. Holmes argues that, "in moving from an existing conceptual framework to a new one, scientists often cannot make a single leap from one coherent mental framework to another. They may have to endure, for extended periods of time, deep fissures within their mental worlds" (Holmes, 1989, p. 53).
3. Discovery depends on finding good data.
Because Laviosier and Krebs were both experimenters, they had to manufacture or invent the data they used. Lavoisier had to choose his experiments more carefully than Krebs, and showed a greater tendency to dismiss anomalous results as errors. Much of this difference may be accounted for simply by the fact that Krebs worked in an age of bigger science, where laboratories had more sophisticated equipment and there was a greater opportunity to carry out multiple experiments. Therefore, Krebs was able to create new data more rapidly than Lavoisier, and may also have felt more secure about eliminating sources of error (Holmes, 1989).
This generalization ought to be modified, then:
3. Discovery depends on finding or inventing good data.
Invention as used here does not mean creating fake data; rather, it reminds us that experimental scientists have to manipulate nature, and that manipulation depends on technologies and techniques that need to be invented.
4. Discovery depends on a combination of flexibility and stubbornness.
Lavoisier modified his theories based on experimental data, but "Lavoisier did not give in easily when his results appeared not to fit his expectations. He regularly guessed at possible sources of error and made whatever corrections he thought reasonable to bring the results more closely in line with his theoretical needs" (Holmes, 1989, p. 58).
Krebs displayed great flexibility in following surprising results. It is less clear where stubbornness plays a role in his work. Perhaps his case suggests a modification of our generalization:
4. Discovery depends on a combination of flexibility and stubbornness, depending both on the individual scientist's cognitive style and on the nature of the problem.
Style and problem interact, here, because scientists often choose problems that suit their styles. As a result of Holmes' work on both Lavoisier and Krebs, we will have to add a fifth generalization:
5. The act of writing is part of the discovery process.
For Lavoisier and Krebs, writing was critical to formulating their discoveries. Even Kepler's narratives doubtless helped him clarify his thinking.
James Gleick, in his biography of Richard Feynman, reports an exchange between a historian, Charles Weiner, and Feynman concerning Feynman's scientific notes. Feynman claimed that he "actually did the work on the paper." Weiner countered that the work must have been done in his head, with the record appearing on the paper. Feynman responded, "No, it's not a record, not really. It'sworking. You have to work on paper..." (Gleick, 1992, p. 409).
We don't know to what extent re-shaping his arguments for publication affected Feynman's understanding of his discoveries. Gleick notes that "he wrote in astonishing volume as he worked--long trains of thought, almost suitable to serve immediately as lecture notes." Furthermore, these lecture notes were often published. It would be interesting to know more about Feynman's process of revision as he turned notes into lectures.
At any rate, the Feynman example provides further support for the idea that writing plays a central role in discovery. As with the previous generalization about stubbornness, individual scientists differ in their writing style--for some, major conceptual changes can be seen in the notes, others in draft manuscripts, still others in both.
This page was last edited: Wednesday, July 14, 1999