Scientific Discovery and Technological Innovation:

Ulcers, Dinosaur Extinction, and the Programming Language Java

Paul Thagard and David Croft

Philosophy Department

University of Waterloo

Waterloo, Ontario, N2L 3G1


Whereas scientists formulate laws and theories to account for observations, inventors create new technology to accomplish practical goals. Scientific discovery and technological innovation are among the most important accomplishments of the creative human mind. The aim of this paper is to compare how scientists produce discoveries with how inventors produce new technology. After briefly reviewing an account of the recent discovery of the bacterial theory of ulcers, we show that a similar account applies to the discovery that dinosaurs became extinct because of an asteroid collision. Both these discoveries involved a combination of serendipity, questioning and search. We then describe how these three processes also contributed to a very important recent technological innovation, the development of the programming language Java. The paper concludes with a more general assessment of the similarities and differences between cognitive processes involved in discovery and invention.




Until recently, peptic ulcers were thought by medical practitioners to be caused by excess acidity, and were treated using antacids. The popular view that ulcers are caused by psychological stress was never substantiated by medical research. In the past decade, there has been a dramatic shift in medical thinking about ulcers, prompted by the discovery by two Australian physicians, J. Robin Warren and Barry Marshall, that most peptic ulcers are caused by infection by a bacterium, Helicobacter pylori. Thagard (1999) provides an account of the psychological, social, and physical processes involved in the discovery and acceptance of the bacterial theory of ulcers. Part of the account was a discussion of how serendipity, questioning, and search each contributed to the discoveries of Warren and Marshall.


Research on the cognitive processes involved in discovery has tended to ignore the substantial role of serendipity in scientific research. Research is serendipitous when scientists make discoveries that they were not seeking. Because the creation of laws and theories requires substantial focussed cognitive effort, their discovery is never purely serendipitous. But things and phenomena are often surprisingly encountered as part of the physical process of interacting with the world, as when Roentgen discovered X-rays and Fleming found penicillium mold capable of killing bacteria. In 1979, J. Robin Warren, a pathologist in Perth, Australia, microscopically noticed spiral bacteria in gastric biopsies. He was not looking for bacteria in the stomach ­ he just happened to find them in the course of his normal duties as a pathologist. After much investigation and conceptual change, these bacteria were later named Helicobacter pylori and found to be responsible for most peptic ulcers. Because it was never the intention of Dr. Warren to search for stomach bacteria, discovery of H. pylori was clearly a matter of serendipity. Serendipity also contributed to the culturing of the bacteria after many failures, when technicians accidentally kept the bacteria incubating for five days rather than two.

In 1981, Dr. Barry Marshall began working with Dr. Warren to determine whether the bacteria found by Warren had any medical significance. They generated numerous important questions, including:

Are the bacterial correlated with disease?

How can the bacteria be eliminated?

Would elimination of the bacteria affect disease?


Questioning is an important part of the early stages of scientific discovery when problems to be solved are being recognized and constructed. Once a problem is well-formed, it may be possible to use much more constrained searches to generate answers to questions. After Marshall and Warren had raised the question of whether the bacteria might be responsible for disease, it was relatively straightforward to design an experiment to determine whether the bacteria correlate with common gastrointestinal diseases such as gastritis, gastric ulcer, and duodenal ulcer. The correlations found between bacterial presence and ulcers were surprising, but they were not serendipitous because the researchers were looking to see whether they occur. The search for such correlations could be done systematically once questioning had posed a problem to be solved. Similarly, once Warren and Marshall questioned how the bacteria can be eliminated, they could do a systematic search involving a known array of antibiotics to determine which ones might be effective at eliminating the disease. Subsequent experiments determined that a combination of antibiotics was capable of eliminating the bacteria and curing peptic ulcers.


A much less condensed version of this account of discovery can be found in Thagard (1999). Serendipity, questioning, and search each contributed to the discoveries that produced the bacterial theory of peptic ulcers. For convenience, we shall refer to this account of discovery as the SQS model, for serendipity-questioning-search. The task at hand is to determine whether the SQS model applies to other cases of scientific discovery and technological innovation.


Dinosaur Extinction


Walter Alvarez (1998) has written a vivid personal account of one of the most important recent discoveries in geology and paleontology. In the mid-1970s, while investigating the geological origins of the Apennine mountains in Italy, he discovered a layer of clay at the boundary between the Cretaceous and the Tertiary sediments in rocks near Gubbio. The KT boundary marks the time when dinosaurs and many other organisms became extinct, so Alvarez and his collaborators began investigating the clay for clues as to what might have happened. They found that the clay contains an unusually high amount of iridium, more typical of asteroids and comets. This led to the hypothesis that an asteroid might have hit the earth 65 million years ago and caused the mass extinction, a hypothesis that has received considerable empirical support. (For a discussion of the explanatory coherence of the collision hypothesis and its competitors, see Thagard, 1991.)


Alvarez's detailed description of the chain of discoveries leading to the new theory of dinosaur extinction shows that it fits the SQS model developed for the ulcer case. Serendipity figured in several of the empirical discoveries. Initially, Alvarez's interest was in deciphering the origin of the Apennine mountains, and he began collaborating with Bill Lowrie, a geophysicist specializing in paleomagnetism. They discovered surprising magnetic reversals in rocks near Gubbio, and were fortunate that the rocks were full of forams (marine protozoans) that enabled them to date the reversals. Discovery of these reversals was serendipitous in that they were not being sought by the researchers. According to Alvarez (1998, p. 38): "In a roundabout, unexpected, very lucky way, we had gotten the break that all young researchers dream of!"

The discovery of the centimeter-thick layer of clay at the KT boundary was also serendipitous. At Gubbio and at each new outcrop in the Apennines, Alvarez and Lowrie found the thin layer of fossil-free clay between the last limestone bed with Cretaceous forams and the first limestone bed containing only new Tertiary forams. In order to determine the duration of the extinction, Alvarez wanted to find out how long it had taken to deposit the clay layer. He consulted his father, the Nobel prize winning physicist Luis Alvarez, concerning how to measure the sedimentation rate of the clay at the KT boundary. After the failure of an initial attempt to measure sedimentation via the abundance of the isotope beryllium-10, Luis Alvarez suggested that they look at the amount of iridium in the clay. Iridium is much more common in meteorites than in the earth's crust, and they reasoned that a relatively large amount of iridium would indicate a slow sedimentation rate that had allowed meteorite dust to accumulate. Then came another big surprise. The researchers had expected almost no iridium if the clay had been deposited rapidly, and about 0.1 parts per billion if it had been deposited slowly, but they discovered that the clay samples actually contained around 9 parts per billion. Although the researchers were definitely looking to find out how much iridium was in the clay, the discovery of an extraordinarily large amount of it, suggestive of an asteroid collision, was unexpected. Thus serendipity struck several times in the process of inquiry, including the discovery of magnetic reversals, the finding of the clay layer, and the detection of large amounts of iridium.


Of course, serendipity is not just a matter of blindly stumbling on important phenomena, but occurs in the context of highly insightful questioning. Alvarez had initially set himself the task of determining how the Apennines had originated, and generated new questions as the research evolved, including:


Why was there a layer of clay at the KT boundary?

What happened to the Cretaceous forams?

What happened to the dinosaurs?

What was the sedimentation rate of the KT clay?

What extraterrestrial event could have caused both extinction and iridium in clay?

How could a giant impact cause extinction?


Alvarez (1998, p. 42) recounts his awareness that he had come across a major area of scientific investigation:


I remember very clearly walking around the grounds at Lamont one day shortly after Al Fischer's talk there and realizing fully that this was a world-class scientific problem. Much of the work we do as scientists involves filling in the details about matters that are basically understood already, or applying standard techniques to new specific cases. But occasionally there is a question that offers an opportunity for a really major discovery. Choosing what problems and what kid of problems to work on is a critical strategic decision for a scientist. The question of the KT extinction looked like one that could lead in totally new directions, and by the time I finished my walk, I had decided that I would try to solve it.


Much of Alvarez's success, like that of Marshall and Warren in the ulcers case, was the result of generating acute questions that lead to new hypotheses and empirical research.


The cognitive sciences, including psychology, artificial intelligence, and philosophy, currently lack a good theory of how questions, especially fruitful ones, are generated. In the ulcer case, questions arose as the result of surprise (at finding the novel bacteria), curiosity (about whether they are medically significant), need (to help people with stomach problems), and other subsidiary questions generated to help answer the original questions. In Alvarez' thinking, the main sources of questioning appear to be surprise (at finding the clay and at its high Iridium content), curiosity (about the causes of extinction), and the posing of subsidiary questions aimed at providing answers to ones already generated. We need a general cognitive theory of how such sources lead people to ask particular questions, but no attempt to generate such a theory will be made here. Note that a theory of question generation will need not only to consider linguistic and reasoning processes, but will also have to take into account emotional aspects of cognition, since surprise and curiosity are in part emotional states.


Although question generation is little understood, cognitive theory can tell us a lot more about the processes by which questions are answered. A why-question is answered abductively, by generating a hypothesis that explains the puzzling phenomena that generated the question. Analogies are often useful in this abductive process, when explanations of similar phenomena are used to suggest potential hypotheses for the puzzling phenomena (Holyoak and Thagard, 1995). Alvarez (1998, p. 75) does not remember when the hypothesis of an asteroid impact first came up, but he does recall being very interested in impact craters on the Moon and the planets. It is possible, therefore, that the impact hypothesis was generated analogically:


Asteroids have hit the moon and left large craters.

So, similarly, asteroids might have hit the earth and produced the iridium layer.


Unfortunately, no one could explain how an asteroid impact could cause world-wide extinction, until Luis Alvarez remembered reading about the 1883 explosion of the Indonesian volcano, Krakatoa, which had blown so much dust into the air that weather all over the world was affected. Analogically, Alvarez conjectured that the impact of a huge asteroid could send so much dust into the air that it would get dark all around the world and make the whole food chain collapse, producing extinction of the dinosaurs and many other organisms. Hence it appears that analogical reasoning helped to provide answers to some of the important questions that Alvarez was pursuing.


Other problems generated by Walter and Luis Alvarez were so well-defined that their solution can well be described in terms of search through a space of possibilities:


What is the sequence of magnetic reversals above and below the KT boundary?

What element could be used to measure the sedimentation rate of the KT boundary clay?

Where is the impact site?


In each of these cases, there was a determinate set of possible answers, and researchers could proceed by searching for solutions. In contrast, the process of generating questions in the first place is much more open-ended and is not well characterized in terms of search. The space of possible questions is as large as the space of possible sentences, and little is known about how to constrain search of this space to generate questions that are interesting and potentially fruitful.


In sum, the discovery of the asteroid impact theory of dinosaur extinction seems, like the ulcers case, to have involved serendipity, questioning, and search. Now we can consider whether the SQS model also applies to technological innovation.




Since computer programming was invented in the 1940s, many hundreds of programming languages have been developed. Most have turned out to be of little practical use, but a few languages such as FORTRAN and C have had widespread application. In the 1990s, the most successful new programming language has been Java, constructed by James Gosling in 1991 as part of a project at Sun Microsystems. Its spectacular success is in part the result of its suitability for Internet use. Java applets are programs that can be easily accessed using World Wide Web browsers such as Netscape. When a user opens a web page containing a Java applet, the program is downloaded onto the user's own computer and runs regardless of the kind of computer used. Java was enormously innovative in being platform independent, producing programs that can be run on computers with different CPU chips and operating systems. But Java was not designed with the Web in mind, for the first Web browser Mosaic was only developed in 1993, two years after the creation of Java. Java's development provides an interesting story of serendipity, analogy, and creative problem solving.


In 1990, Patrick Naughton, a top programmer at Sun Microsystems told Sun's chief Scott McNealy that he was quitting to join NeXT Computer where he could work on more interesting projects (Bank, 1995). Convinced by Naughton's contention that Sun was becoming insufficiently innovative, McNealy told Naughton he could have a million-dollar budget to put together a small team of outstanding programmers and engineers that would work without corporate interference from Sun. With carte blanche to pursue new projects, Naughton recruited Gosling and a few other top people, and in a hot tub in Lake Tahoe in 1991 they decided to build a prototype of a small device that could control everyday consumer appliances.


To control this device, they originally decided to program the device in C++, a popular computer language that combined the language C, originally build for applications using the operating system Unix, with ideas about object oriented programming borrowed from another language, Simula. However, for use in consumer appliances, the program needed to be more reliable and simpler than was possible with C++, so Gosling decided to develop a new computer language. The result was initially called "Oak" after a tree outside his window, but was later renamed "Java" for marketing reasons.


Oak was basically a simplified, more reliable adaptation of C++, but its major innovation was in the way it could be used by different kinds of computers. Normally, a program written in a particular programming language is compiled into machine language on a specific kind of computer. For example, a C++ compiler running on a Macintosh will produce very different machine code than will be produced by a C++ compiler running on a Windows computer. Because Oak was meant to be used on a wide variety of different devices, Gosling designed it to be compiled, not into machine language, but into bytecodes, an intermediate language that can then be interpreted for use on particular machines. Any computer that has a bytecode interpreter can run a compiled Java program using its own machine language. The resulting process may be slower than running a program written in C++, but much more flexible in that it does not require a different compiler for every computer platform.

By August, 1992, Naughton, Gosling and their seven-person team had produced a prototype personal digital assistant with a small, touch-operated screen that could control TVs and VCRs, but attempts to sell the device for use in interactive television and computer games failed. Then in June, 1993, the first Mosaic browser was released and the World Wide Web began to take off. Bill Joy, Sun Microsystems' co-founder, realized that Oak could naturally be adapted for use on the Internet. By 1995, Gosling had produced a Web-suitable version of Oak, now renamed Java, and Naughton had written HotJava, an interpreter for Web browsers. Since then, many thousands of Java applications have been produced and are available on the World Wide Web. For example, Thagard has made available on his Web site a Java version of his program ECHO, which evaluates hypotheses based on their explanatory coherence ( Previously, researchers wanting to run ECHO had to be sent special versions for their own particular computer, but the Java version of ECHO is usable by anyone with an up-to-date Web browser.


Let us now consider whether the SQS model of scientific discovery applies to the invention of Java. At first glance, serendipity was not involved: Gosling designed the language to solve the problem of programming a variety of consumer appliances. But the application to the Internet was clearly serendipitous, because none of the Sun Microsystems team had any idea of eventual Web applications when they were developing their prototype product. By luck, a programming language designed for one purpose turned out to be incredibly useful for another purpose that arose only later. We can distinguish at least three kinds of serendipity:


Finding something not sought for, as in the discovery of H. pylori and the clay layer at the KT boundary.

Finding something sought for, but by an unplanned method, as in the accidental culturing of H. pylori through longer incubation. (Roberts, 1989, calls this pseudoserendipity.)

Finding something sought for that turns out to useful for a purpose unconnected with the reason for which it was sought, as in the creation of Java.


Another example of the last sort of serendipity is the invention of Post-It Notes, which came about because a of much delayed realization that a weak glue could actually be useful.


Much more important than serendipity to the development of Java was the generation of interesting and fruitful questions. Naughton's team started with very general questions that they gradually refined:


What will be the next wave of computing?

How can computers be used more effectively in consumer electronics?

How can a sophisticated personal digital assistant (PDA) be built for consumer appliances?

What programming language could be used for the new PDA?

What features should a new programming language have?

How could programs be made to run on a wide range of electronic devices?

What could the new PDA be profitably use for?

How could the newly developed programming language be used on the Web?


Producing the PDA and the new programming language were prodigious technological feats, but initiation of these projects required the generation of questions that provided a focus for ongoing work.


Once questions were formed, then more constrained processes characterizable as search could be used to generate answers. The question of what programming language to use at first seems amenable to answer by search, since there were only a small number of reasonably appropriate languages to consider. But Gosling's decision that C++ was inadequate provoked a less constrained effort to produce a new programming language. Here analogy was a crucial ingredient, since Gosling based much of Oak on C++, although he also incorporated features of other programming languages that he liked. Use of multiple analogies is common in the development of programming languages, as we described in the creation of C++ from C and Simula.


An analogy of a very different sort seems to have brought Gosling the key insights about how a programming language could be used to control many kinds of devices. At a Doobie Brothers concert in 1991, Gosling reacted to the wiring, speakers, and lights: "I kept seeing imaginary packets flowing down the wires making everything happen. I'd been thinking a lot about making behavior flow through networks in a fairly narrow way. During the concert, I broke through on a pile of technical issues." (quoted by Bank, 1995). The flow of sound and light at the concert suggested to him ideas about the flow of information in computer networks. A much more prosaic analogy may have contributed to the application of the crucial bytecode concept. Gosling (1995) wrote: "The solution we settled on was to compile to a byte coded machine independent instruction set that bears a certain resemblance to things like the UCSD Pascal P-codes." More constrained search processes must have been involved in solution of more low-level problems concerning the details of the construction of Java.


As the next section will discuss, there are differences as well as similarities in the cognitive processes involved in scientific discovery and technological innovation. But at a higher level, it is clear that technological innovation in the case of the Java programming language involved all three ingredients of the SQS model presented earlier: serendipity, questioning, and search. Let us now do a more fine-grained cognitive comparison.


Cognitive Processes in Discovery and Innovation


Scientists and inventors ask different kinds of questions. Because scientists are largely concerned with identifying and explaining phenomena, they generate questions such as:


Why did X happen? What is Y? How could W cause Z?


In contrast, inventors have more practical goals that lead them to ask questions of the form:


How can X be accomplished? What can Y be used to do? How can W be used to do Z?


Despite the differences in the form of the questions asked by scientists from the form of the questions asked by inventors, there is no reason to believe that the cognitive processes underlying questioning in the two contexts are fundamentally different. Scientists encounter a puzzling X and try to explain it; inventors identify a desirable X and try to produce it. As mentioned above, there is no currently available theory of question generation, but roughly the same cognitive theory should be able handle both scientific and technological questions. Both kinds of question are produced in response to surprise, curiosity, need, and other questions, and both kinds involve emotions such as puzzlement and excitement.


In scientific discovery, why-questions are answered abductively, by generating hypotheses that provide explanations. As identified by C. S. Peirce and subsequent students of scientific reasoning, the basic pattern is:


Why X? Y would explain X. So maybe Y.


By definition, abduction concerns the generation of explanatory hypotheses, so abduction is not central to invention, which primarily involves the generation of answers to questions that are practical rather than explanatory. But there is a reasoning process in technological innovation that has a very similar structure to abductive inference (cf. Dasgupta, 1996):


How to do X? Y might do X. So maybe try Y.


For both invention and discovery, the difficult task is to generate the answer Y that provides a possible solution for the problem X. Abductive generation of hypotheses can be performed using rules such as "If Y then X", by causal schemas that connect Y and X, by visual representations of causal connections (Shelley, 1996), and by analogical inference described in the next paragraph. Invention, generation of inventive solutions can similarly be performed using rules, causal schemas, visual representations such as diagrams, and analogical inference. Abduction may also be a subsidiary process in technological innovation, when innovation depends on generating explanations of physical processes relevant to the new technology.


Analogical inference in discovery and innovation has a difference focus, but a common structure. Analogical discovery proceeds roughly as follows:


Why X? X is like Y, which is explained by Z. So maybe X is explained by something like Z.


Analogical invention is analogous:


How to do X? X is like Y, which is accomplished by Z. So maybe X can be accomplished by something like Z.


Although the purposes of analogical discovery and analogical invention are different, they both involve solving a target problem by finding a source problem that can be mapped to the target problem in order to suggest a solution to it. Analogies have often contributed to technological innovation:

Laennec invented the stethoscope by analogy to a child's game involving listening to the end of piece of wood (Thagard, 1999);

Alexander Graham Bell modeled the telephone in part on the human ear (Gorman, 1992);

Thomas Edison used his earlier work on telegraphy to suggest solutions to problems involved in designing electric lights (Israel, 1998, p. 168).


An important part of scientific discovery that this paper has not yet addressed is the formation of new concepts such as Helicobacter pylori. Whereas new scientific concepts are generated to refer to entities and processes that are discovered or postulated as part of explanatory hypotheses, new technological concepts are formed to describe newly created entities and processes. In the Java case, the most important new concept was applet , introduced to refer to programs that could be sent over the Internet as Java bytecodes to be used by Web browsers. In science, theoretical concepts are largely generated by combining existing concepts, and the same seems to be true for technological innovation. Applet is roughly the result of combining "program application" and "Internet transfer". Bytecode was not a completely new idea generated for Java, but was rather modified from use in earlier programming languages such as Smalltalk; the key extension was that bytecodes could be sent over the Internet to run on many different kinds of computers. New technologies have generated numerous new concepts: automobile, VCR, computer chip, graphical user interface, and so on. Producing new concepts can be a crucial part of generating novel questions, as when Marshall and Warren named the newly discovered bacteria and used the new name to ask questions about it. In both scientific discovery and technological innovation, creative thinkers build on their familiarity with the knowledge available in a domain to produce new concepts, problems, and solutions.


It is important to recognize that technological innovation is not a purely cognitive process operating in the mind of individuals. Even though Gosling was the major programmer for Java, he worked extremely closely with a team of programmers and engineers at Sun Microsystems, and other people were responsible for key ideas such as adapting Java for the Internet. Scientific discovery is similarly a social as well as a cognitive process (Thagard, 1997, 1999). Barry Marshall and Robin Warren worked with many collaborators on the testing of the bacterial theory of ulcers, and Walter Alvarez worked with his father and others on the development and testing of the collision theory of dinosaur extinction. Like scientific research, technological innovation is a physical as well as a cognitive and social process, involving interactions with the world and multiple refinements as devices are improved. In science, a theory is often refined after experiments based on it display problems. Even more frequently, new technology requires many rounds of alteration and adjustment before it can accomplish the practical goals that it was designed to meet. Java went through numerous changes as it was improved and adapted for Internet and other uses.




Two cases of scientific discovery and one of technological innovation are hardly sufficient to generalize conclusively about the relation between discovery and invention. But the ulcers, dinosaur, and Java cases are rich examples of the creative mind at work, and our analysis of them has suggested some major similarities and minor differences in the cognitive and other processes involved in discovery and innovation. Discovery and innovation both involve serendipity, when physical and social processes introduce unexpected elements into the cognitive process of problem solving. Both involve questioning, although the questions generated differ in whether they are directed at explanatory or practical goals. Both often involve the use of analogy, with fundamentally similar processes of analogical inference. Both often involve the generation of new concepts, although technological concepts usually apply to newly created devices rather than to discovered or postulated entities. Although abductive inference to explanatory hypotheses is much more central to scientific discovery than to technological innovation, inference to possible solutions to technological problems seems to involve very similar representations and processes. We can therefore conclude that scientific discovery and technological innovation are cognitively very much alike. Scientists and inventors are among our most creative thinkers, but there is little fundamental difference in the mental origins of their creativity.



Alvarez, W. (1998). T. Rex and the Crater of Doom. New York: Vintage.

Bank, D. (1995). The Java saga. Wired, 3(12).

Dasgupta, S. (1996). Technology and Creativity. New York: Oxford University Press.

Gorman, M. E. (1992). Simulating Science. Bloomington, IN: Indiana University Press.

Gosling, J. (1995). Oak intermediate bytecodes. available at

Gosling, J. (n. d.). A brief history of the green project, available at

Holyoak, K. J., & Thagard, P. (1995). Mental Leaps: Analogy in Creative Thought. Cambridge, MA: MIT Press/Bradford Books.

Israel, P. (1998). Edison: A Life of Invention. New York: Wiley.

Roberts, R. M. (1989). Serendipity: Accidental Discoveries in Science. New York: Wiley.

Shelley, C. P. (1996). Visual abductive reasoning in archaeology. Philosophy of Science, 63, 278-301.

Thagard, P. (1991). The dinosaur debate: Explanatory coherence and the problem of competing hypotheses, in Philosophy and AI: Essays at the Interface, J. Pollock & R. Cummins (Eds.), (pp. 279-300). Cambridge, Mass.: MIT Press/Bradford Books.

Thagard, P. (1997). Collaborative knowledge. Noûs, 31, 242-261.

Thagard, P. (1999). How Scientists Explain Disease. Princeton, NJ: Princeton University Press.


Back to Science and Disease articles.