In Proceedings of the Nineteenth Annual Conference of the Cognitive Science Society.Michael Shafto and Pat Langley (Eds.). Mahwah, New Jersey: Lawrence Erlbaum, 1997, pp. 161-166

Recent Work in Computational Scientific Discovery

Lindley Darden (darden@umiacs.umd.edu)
Committee on the History and Philosophy of Science, Department of Philosophy
University of Maryland, College Park, MD 20742 USA
Abstract
   This paper reviews work in computational scientific
discovery. After a brief discussion of its history, the
focus will be on work since 1990. The second half of
the paper discusses the author’s use of three methods for
studying reasoning strategies in scientific change:
historical-philosophical vs. live-in-the-lab vs.
computational, pointing out advantages and
disadvantages of the computational method.

Introduction
There are a number of approaches to the study of reasoning
in scientific discovery. In addition to computational
approaches, work continues in cognitive science (e.g.,
Schunn & Dunbar, 1996), in laboratory studies (e.g., Darden
& Cook, 1994; Dunbar, 1995) and in philosophy of science
(e.g., Bechtel & Richardson, 1993; Darden, 1991; Kleiner,
1993; Nersessian, 1992; Nickles, 1994; Schaffner, 1993;
Spirtes, Glymour & Scheines, 1993). Unfortunately, of the
over 200 papers and abstracts submitted for the Philosophy
of Science Association meeting in 1996, none were on the
topic of reasoning in scientific discovery (Darden, Ed., 1996;
1997). Most philosophers of science do not view discovery
as a central topic in the field, despite continuing work by
those of us called “friends of discovery” (Nickles, Ed. 1980).
It is encouraging that the Cognitive Science Society is
sponsoring this Symposium on Scientific Discovery.
   This paper will briefly review the history of
computational scientific discovery that uses methods from
artificial intelligence. (Non-cognitive, non-AI computational
work is outside the scope of this paper.) The first part of the
paper will concentrate on the work since 1990 (Shrager &
Langley, Eds.). The extensive reference list provides a guide
for further reading. The second half of the paper will
compare three methods used in my own work on reasoning
strategies in scientific change. Finally, I will point out
advantages and disadvantages of the computational approach
from my perspective as a philosopher of science.

Pioneering Work
The study of computational scientific discovery emerged
from the view that science is a problem solving activity,
that heuristics for problem solving can be applied to the
study of scientific discovery in either historical or
contemporary cases, and that methods in artificial
intelligence provide techniques for building computational
systems. Pioneers in this work are Bruce Buchanan (e.g.,
1982) and Herbert Simon (e.g., 1977). Buchanan was trained
as a philosopher of science at a time when the profession
was dominated by Popper’s (1965) view that there is no
logic of discovery. Buchanan stated the new research
 program: “The traditional problem of finding an effective
method for formulating true hypotheses that best explain
phenomena has been transformed into finding heuristic
methods that generate plausible explanations. The problem
of giving rules for producing true scientific statements has
been replaced by the problem of finding efficient heuristic
rules for culling the reasonable candidates for an explanation
from an appropriate set of possible candidates” [and finding
methods for constructing the candidates] (Buchanan 1985,
110-111). Discovery as heuristic search in a search space
enabled AI methods to be applied to discovery tasks.
   The first expert system, DENDRAL, was a scientific
discovery system. It formed hypotheses about chemical
compounds, given mass-spectrographic data (Lindsay,
Buchanan, Feigenbaum, & Lederberg, 1980;1993). This was
followed by Meta-DENDRAL, which discovered new rules
in mass spectrographic analysis, so as to by-pass the
problem of getting rules from experts (Buchanan &
Feigenbaum, 1978). Although its original algorithm was a
computational realization of Lederberg’s systematic scan
strategy (Lederberg, 1965), DENDRAL was built to carry
out a contemporary, difficult scientific task rather than as a
model of human cognition.
   A more historical-cognitive approach was the aim of the
work on BACON, which rediscovered various scientific laws
by finding patterns in numerical data (Langley, Simon,
Bradshaw & Zytkow, 1987). Simon’s early work on finding
patterns in sequences (Simon & Kotovsky, 1963) was
extended in BACON to heuristic search for patterns in
numerical data. The most creative of BACON’s abilities was
the decomposition of relational data to conjecture intrinsic
properties in one or more of the objects engaging in the
relations. This step went beyond curve-fitting and was based
on the metaphysical assumption that an entity’s relational
properties are caused by its intrinsic properties. In addition
to the data-driven tasks modeled in BACON, the group also
investigated theory-driven discovery in STAHL. One
wonders to what extent these programs model actual
cognitive processes of historical scientists, as opposed to
finding strategies which are sufficient to reproduce the
historical results. As with most simulations, they provide
“how possibly” accounts. Using studies of notebook
evidence, the KEKADA system (Kulkari & Simon, 1988)
modeled reasoning patterns in some discoveries of the
biochemist Hans Krebs and focused on responses to
surprising experimental results, helping to dispel the
mystery of serendipity in discovery.
   A seminal conference on computational methods for
scientific discovery, whose proceedings were published in
1990 (Shrager & Langley, Eds.) is a useful source for the
state to the field at that time.

161

Recent Work
Some of the pioneers in scientific discovery, e.g., Buchanan,
Simon, and Zytkow, push ahead with their research
programs. Others who contributed to the 1990 volume are
still working on discovery. The American Association for
Artificial Intelligence sponsored a Spring Symposium on
Systematic Methods of Scientific Discovery in March,
1995. A special issue of Artificial Intelligenceon
computational discovery is about to appear, although fewer
papers were received than the editors wished (Simon, Valdes-
Perez & Sleeman, forthcoming). Data-mining in scientific
databases is an active area of research, as are other
computational approaches applied to individual sciences,
e.g., intelligent systems in molecular biology. It is
becoming more difficult to locate computational discovery
work because much of it is published in scientific journals--
a good sign that the methods of producing results of interest
to practicing scientists.
   Buchanan (e.g., Lee et al., 1996) continues work on rule
induction applied to various scientific databases. Simon is
studying the difficult problems of constructing diagrammatic
representations (Larkin & Simon, 1987; Qin & Simon,
1995) and of modeling relations between diagrammatic and
verbal reasoning (Tabachneck-Schijf, Leonardo, & Simon,
1996). Zytkow continues to work on various aspects of
discovery, including analyzing the components needed for an
autonomous discovery agent (e.g., Zytkow, 1995/96) and
knowledge discovery in databases (e.g., Zytkow &
Zembowicz, 1996).
   Much of the current work in computational discovery is
occurring within applications to particular sciences.
According to Peter Karp, the whole field of bioinformatics is
doing computational scientific discovery but there is a
gradient from computational discoveries that are not based
on AI methods, to computational discoveries that are based
on AI methods, to methods with a “cognitive flavor.” Not
much of the bioinformatics work falls into the last category.
However, Karp (et al., 1996) applied reasoning by analogy
to predict metabolic pathways in the bacterium, H.
influenzae,based on the extensive knowledge base that he
and Monica Riley, a bacterial geneticist, have developed for
E. coli.
   Larry Hunter, a frequent editor of publications in AI and
molecular biology (e.g., Hunter 1993), recently informed me
that there is a clear success is the application of AI
technology to molecular biology: hidden Markov models
(HMMs) for molecular sequence analysis. They are being
applied to automatically build models of families of
nucleotide and amino acid sequences. These models are
useful as extremely sensitive classifiers of novel sequences,
and also generate multiple sequence alignments of large
numbers of sequences in a computationally efficient way.
Tools based on this approach are now in wide use in the
biological community. A review article is Eddy (1996).
Also, AI-based qualitative reasoning technologies have
produced several good applications in reasoning about
metabolism. Perhaps somewhat surprising is that the work
in intelligent systems in molecular biology, for the most
part, does not employ discovery methods discussed at the
Shrager and Langley (Eds. 1990) conference.
    The extensive protein sequence database has provided a
challenge for those seeking to find computational methods
to predict how the linear amino acids will fold into the
secondary and tertiary structures in proteins. The Human
Genome Project, which is rapidly producing millions of
bases of sequence information about both human and model
organism genomes, presents a challenge for computational
approaches. Good programs are needed for discovering genes,
both coding regions and regulatory regions, in these linear
sequences. Current programs are not good at finding introns,
intervening sequences between the coding regions of genes.
Since the genetic system has some means of detecting
introns, one can expect computational systems to be able to
discover the signal(s). Knowledge discovery in scientific
databases (e.g., Fayyad, Haussler & Stolorz, 1996) promises
to be an important area in coming years.
   Raul Valdes-Perez’s (1994) work in chemistry shows the
power of computational systems in doing a systematic
search of a hypothesis space, given certain constraints.
MECHEM is able to find reaction pathways that chemists
have missed.
   Buchanan’s work on rule discovery in scientific databases
and Valdes-Perez’s work on systematically conjecturing
chemical reaction pathways illustrate the power of design AI
systems that aim, not at realistically modeling human
cognitive capacities, but using computational methods to
circumvent human limitations. Humans are not good at
searching massive databases and manipulating sets of rules
with many features to make predictions. Cognitive science
research has shown that humans have a tendency to focus
too rapidly on one hypothesis before doing a systematic
search of a hypothesis space. Discovery programs that are
more systematic and more thorough than humans are an aid
to scientists.

Computational Discovery: Pros and Cons
My own work on reasoning in scientific change focuses
on an cyclic process: discovery, assessment, revision. Given
a good revision procedure, one’s discovery methods can be
weaker. Strategies for these processes include: strategies for
producing new ideas, e.g., analogies, abstraction
instantiation, interfield relations; strategies for theory
assessment, e.g., prediction-testing, relations to theories in
other fields; and strategies for anomaly resolution (Darden
1991, Ch. 15). After extensive historical study of the
development of Mendelian genetics, I proposed hypothetical
strategies of the three types. The historical evidence was
inadequate to show that they are descriptive cognitive
strategies actually used by geneticists. Instead, they are
hypothetical strategies that couldhave been used in the
historical development of the theory of the gene to produce
the changes that didoccur (Darden, 1991). One needs to
show that these strategies are effectiveproblem-solving
strategies, instances of useful “compiled hindsight” (Darden,
1987), applicable to additional cases, worthy of being used
by contemporary scientists or to build AI discovery systems.
   I visited in Joshua Lederberg’s Laboratory for Molecular
Genetics and Informatics and participated in episodes of
anomaly resolution that exemplified some of the revision
strategies I had proposed (Darden & Cook 1994). One

162
difficulty with the live-in-the lab approach is that little may
happen while you are there; fortunately, I was able to
observe some anomaly resolution strategies in use.
Although I have attempted to implement some of the
strategies in AI programs in order to demonstrate their
efficacy (e.g., Darden & Rada, 1988; Kettler & Darden 1993;
Darden, 1997), I have returned to historical-philosophical
work, testing whether strategies from the Mendelian case
apply to molecular biology (Darden, 1995).
   Computational discovery work has advantages and
disadvantages. Finding an adequate knowledge representation
for a scientific case is difficult. Early work attempted to
represent the relations between genes and chromosomes in
part-whole hierarchies and to implement reasoning via
inheritance and upward propagation of properties (Darden &
Rada, 1988). A much more fruitful method for knowledge
representation in genetics was the functional representation
(Josephsons, Eds., 1994) for genetic processes (Darden
1997). Furthermore, when one is designing a computational
system to rediscover a historical hypothesis, one must
navigate between designing a system that trivially
reproduces exactly what one is seeking versus designing a
system that is unable to accomplish the task at all. Analogy
systems often suffer these problems: either the analog is
represented in such a way that the system easily finds it or
there are so many analogs that the task becomes impossible
(for attempts to navigate between these problems, see
Kettler & Darden, 1993; Holyoak & Thagard, 1995).
   An advantage of computational methods is the precision
and completeness that is required to build a working system.
The philosopher-historian may neglect aspects that the
programmer must specify in detail if the system is to run. A
computational approach forces one to reexamine aspects that
may be otherwise neglected. However, this advantage is
purchased at the price of much time and effort to implement
even small parts of a historical case. Various aspects of
human discovery, such as the use of pictorial models (e.g.,
the beads on a string model for genes on chromosomes),
provide substantial difficulties when designing an
implementation. On the plus side, once one has invested the
effort in building a running system, then there is the fun of
running experiments, doing “what-if” analyses, testing
alternative strategies.
   The approach in our TRANSGENE system (Darden,
Moberg, Thadani & Josephson, 1992; Darden, 1997) was
also used by Karp (1990) in his GENSIM and HYPGEN
systems and points to a fruitful way to design a
computational discovery system. A qualitative simulator of
biological (or other) processes is built and used to make
predictions. Data is supplied to test the predictions and
another component of the system compares the prediction
with data, detects anomalies, and uses diagnosis/redesign
strategies to localize the fault in the simulator and redesign a
module to remove the anomaly. Perhaps this architecture
may be of use in building future AI systems or perhaps
more traditional simulation models might be coupled with a
revision system to do diagnosis/redesign for anomaly
resolution and model improvement.
   It will be exciting to see what computational scientific
discovery produces in the coming years.

Acknowledgments
The TRANSGENE work was supported by the General
Research Board of the University of Maryland and the
National Science Foundation Grant No. RII-9003142. Any
opinions, findings and conclusions or recommendations
expressed in this material are those of the author and do not
necessarily reflect those of the National Science Foundation.
The TRANSGENE system was designed in collaboration
with John and Susan Josephson and Dale Moberg; TR.3 was
implemented by Sunil Thadani. This paper was written
while I enjoyed the hospitality of the Center for Philosophy
of Science at the University of Pittsburgh. Very helpful
were discussions with, and reprints received from, Bruce
Buchanan and Herb Simon. Rapid email responses from
Larry Hunter, Peter Karp, and Pat Langley were appreciated.
Sets of reprints from Kevin Dunbar, Nancy Nersessian, Tom
Nickles, and Jan Zytkow aided me in learning about their
recent work. I enjoyed the demo of MECHEM by Raul
Valdes-Perez and I profited from his web page:
www.cs.cmu.edu/~sci-disc

References
Alberdi, E. & Sleeman, D. (1997, forthcoming). “ReTAX:
   A Step in The Automation of Taxonomic Revision,”
   Artificial Intelligence91(2).
Baker, L. M. & Dunbar, K. (1996). “Constraints on the
   Experimental Design Process in Real-World Science,” in
   G.W. Cottrell (Ed.), Proceedings of the Eighteenth Annual
   Conference of the Cognitive Science Society,pp. 154-
   159. Mahwah, NJ: Lawrence Erlbaum Associates.
Bechtel, W. & Richardson, R.C. (1993). Discovering
   Complexity: Decomposition and Localization as Strategies
   in Scientific Research.Princeton, N. J.: Princeton
   University.
Buchanan, B.G. (1982). “Mechanizing the Search for
   Explanatory Hypotheses,” in P. Asquith & T. Nickles
   (Eds.), PSA 1982V. 2, pp. 129-146. Philosophy of
   Science Association, East Lansing, Michigan.
Buchanan, B.G. (1985). “Steps Toward Mechanizing
   Discovery,” in K. Schaffner (Ed.), Logic of Discovery and
   Diagnosis in Medicine.Berkeley: University of California
   Press.
Buchanan, B.G. (1994). “The Role of Experimentation
   in Artificial Intelligence,” Phil. Trans. R. Soc. Lond.A
   349:153-166.
Buchanan, B.G. & Feigenbaum, E. A. (1978). “DENDRAL
   and Meta-DENDRAL: Their Applications Dimension,”
   Artificial Intelligence11:5-24.
Cohen, D, M., Kulikowski, C., & Berman, H. (1995).
   “DEXTER: A System that Experiments with Choices of
   Training Data Using Expert Knowledge in the Domain of
   DNA Hydration,” Machine Learning21:81-101.
Corruble, V. & Ganascia, J.-G. (1997, forthcoming).
   “Induction and The Discovery of the Causes of Scurvy: A
   Computational Reconstruction,” Artificial Intelligence
   91(2).
Dama, M. & Dunbar, K. (1996). “Distributed Reasoning:
   An Analysis of Where Social and Cognitive Worlds Fuse,”

163
   in G. W. Cottrell (Ed.), Proceedings of the Eighteenth
   Annual Conference of the Cognitive Science Society,pp.
   166-170. Mahwah, NJ: Lawrence Erlbaum Associates.
Darden, L. (1976). “Reasoning in Scientific Change:
   Charles Darwin, Hugo de Vries, and the Discovery of
   Segregation,” Studies in History and Philosophy of
   Science7:127-169.
Darden, L. (1980). “Theory Construction in Genetics,” in T.
   Nickles (Ed.), Scientific Discovery: Case Studies.
   Dordrecht: Reidel.
Darden, L. (1982). “Artificial Intelligence and Philosophy of
   Science: Reasoning by Analogy in Theory Construction,”
   in T. Nickles & P. Asquith (Eds.), PSA 1982,V. 2, pp.
   147-165. East Lansing, Michigan: Philosophy of Science
   Association.
Darden, L. (1985). “Hugo de Vries’s Lecture Plates and the
   Discovery of Segregation,” Annals of Science42:233-242.
Darden, L. (1987). “Viewing the History of Science as
   Compiled Hindsight,” AI Magazine8(2):33-41.
Darden, L. (1990). “Diagnosing and Fixing Faults in
   Theories,” in J. Shrager & P. Langley (Eds.),
   Computational Models of Scientific Discovery and Theory
   Formation.San Mateo, California: Morgan Kaufmann.
Darden, L. (1991). Theory Change in Science: Strategies
   from Mendelian Genetics.New York: Oxford University
   Press.
Darden, L. (1992). “Strategies for Anomaly Resolution,” in
   R. Giere (Ed.), Cognitive Models of Science.
   Minneapolis: University of Minnesota Press.
Darden, L. (1995). “Exemplars, Abstractions, and
   Anomalies: Representations and Theory Change in
   Mendelian and Molecular Genetics,” in J. G. Lennox & G.
   Wolters (Eds.), Concepts, Theories, and Rationality in the
   Biological Sciences.Pittsburgh, PA: University of
   Pittsburgh Press.
Darden, L. (1997). “Anomaly-Driven Theory Redesign:
   Computational Philosophy of Science Experiments,” in T.
   W. Bynum & J. Moor (Eds.), Digital Phoenix: How
   Computers are Changing Philosophy.Oxford: Blackwell.
Darden, L. (Ed.) (1996). PSA 1996,Proceedings of the 1996
   Biennial Meeting of the Philosophy of Science
   Association, Part I: Contributed Papers, Philosophy of
   Science,Supplement to Volume 63, September 1996.
Darden, L. (Ed.) (1997). PSA 1996,Proceedings of the 1996
   Biennial Meeting of the Philosophy of Science
   Association, Part II: Symposium Papers, Philosophy of
   Science,Supplement to Volume 64, forthcoming
   September 1997.
Darden, L. & Rada, R. (1988). “Hypothesis Formation
   Using Part-Whole Interrelations,” in David Helman (Ed.),
   Analogical Reasoning.Dordrecht: Reidel.
Darden, L. & Rada, R. (1988). “Hypothesis Formation Via
   Interrelations,” in Armand Prieditis (Ed.), Analogica.Los
   Altos, California: Morgan Kaufmann.
Darden, L., Moberg, D., Nagarajan, S. & Josephson, J.
   (1991). Anomaly Driven Redesign of a Scientific Theory:
   The TRANSGENE.2 Experiments(Tech. Rep. 91-LD-
   TRANSGENE). Columbus, OH: Laboratory for Artificial
   Intelligence Research, Ohio State University.
 Darden, L., Moberg, D., Thadani, S. & Josephson, J.
   (1992). A Computational Approach to Scientific Theory
   Revision: The TRANSGENE Experiments(Tech. Rep.
   92-LD-TRANSGENE). Columbus, OH: Laboratory for
   Artificial Intelligence Research, Ohio State University.
Darden, L. & Cook, M. (1994). “Reasoning Strategies in
   Molecular Biology: Abstractions, Scans and Anomalies,”
   in D. Hull, M. Forbes, & R. M. Burian (Eds.), PSA
   1994,v. 2, pp. 179-191. East Lansing, Michigan:
   Philosophy of Science Association.
Dunbar, K. (1993). “Concept Discovery in a Scientific
   Domain,” Cognitive Science17:397-434.
Dunbar, K. (1995). “How Scientists Really Reason:
   Scientific Reasoning in Real-World Laboratories,” in R. J.
   Sternberg & J. E. Davidson (Eds.), The Nature of Insight.
   Cambridge, MA: MIT Press.
Dunbar, K. (1996). “Beyond the Myth of The Lone
   Scientist: Distributed Reasoning in Science and Scientific
   Discovery,” Cognitive Studies: Bulletin of the Japanese
   Cognitive Science Society3:24-26.
Dunbar, K. (forthcoming). “How Scientists Think: On-line
   Creativity and Conceptual Change in Science,” in T. B.
   Ward, S. M. Smith & J. Vaid (Eds.), Conceptual
   Structures and Processes: Emergence, Discovery and
   Change.Washington, D. C.: American Psychological
   Association Press.
Eddy, S. R. (1996). “Hidden Markov Models,” from
   ftp://genome.wustl.edu/pub/eddy/hmm-review.ps.Z
Fayyad, U., Djorgovski, S. G., & Weir, N. (1996). “From
   Digitized Images to Online Catalogs,” AI Magazine17
   (2): 51-66.
Fayyad, U., Haussler, D. & Stolorz, Z. (1996). “KDD for
   Science Data Analysis: Issues and Examples,” In
   Proceedings of the Second International Conference on
   Knowledge Discovery and Data Mining (KDD-96),pp. 50-
   56. Menlo Park, CA: AAAI Press.
Fayyad, U., Piatetsky-Shapiro, G., Smyth, P. (1996).
   “From Data Mining to Knowledge Discovery in
   Databases,” AI Magazine17 (3): 37-54.
Griffith, T. W., Nersessian, N. & Goel, A. (1996). The
   Role of Generic Models in Conceptual Change(Tech.
   Rep. CIT-GOGSCI-96/02). Atlanta, GA: College of
   Computing, Georgia Institute of Technology.
Holyoak, K. J. and Thagard, P. (1995). Mental Leaps:
   Analogy in Creative Thought.Cambridge, Mass.: MIT
   Press.
Hunter, L. (Ed.) (1993). Artificial Intelligence and Molecular
   Biology.Cambridge, MA: MIT Press.
de Jong, H. & Rip, A. (1997, forthcoming). “The Computer
   Revolution in Science: Steps towards the Realization of
   Computer-Supported Discovery Environments,” Artificial
   Intelligence91(2).
Josephson, J.R. & Josephson, S.G. (Eds.) (1994).
   Abductive Inference: Computation, Philosophy,
   Technology.New York: Cambridge University Press.
Karp, P. (1989). Hypothesis Formation and Qualitative
   Reasoning in Molecular Biology.Ph.D. Dissertation,
   Stanford University, Stanford, California.
Karp, P. (1990). “Hypothesis Formation as Design,” in J.
   Shrager and P. Langley (Eds.), Computational Models of

164
   Scientific Discovery and Theory Formation.San Mateo,
   California: Morgan Kaufmann.
Karp, P. (1993). “A Qualitative Biochemistry and its
   Application to the Regulation of the Tryptophan Operon,”
   in L. Hunter (Ed.), Artificial Intelligence and Molecular
   Biology.Cambridge, MA: AAAI Press and MIT Press.
Karp, P., Ouzouni, C. & Paley, S. M. (1996). “HinCyc: A
   Knowledge Base of the Complete Genome and Metabolic
   Pathways of H. influenzae,”in D.J. States, P. Agarwal, T.
   Gaasterland, L. Hunter & R. Smith (Eds.), ISMB-96:
   Proceedings of the Fourth International Conference on
   Intelligent Systems for Molecular Biology,pp. 116-124.
   Menlo Park, CA: AAAI Press.
Kettler, B. & Darden, L. (1993). “Protein Sequencing
   Experiment Planning Using Analogy,” in L. Hunter, D.
   Searls, & J. Shavlik (Eds.), ISMB-93, Proceedings of the
   First International Conference on Intelligent Systems for
   Molecular Biology,pp. 216-224. Menlo Park, CA: AAAI
   Press.
Kleiner, S. A. (1993). The Logic of Discovery: A Theory of
   the Rationality of Scientific Research.Dordrecht: Kluwer.
Kulkarni, D. and Simon, H.A. (1988). “The Processes of
   Scientific Discovery: The Strategy of Experimentation,”
   Cognitive Science12:139-175.
Langley, P., Simon, H.A., Bradshaw, G.L. & Zytkow,
   J.M. (1987). Scientific Discovery: Computational
   Explorations of the Creative Process.Cambridge,
   Massachusetts: MIT Press.
Larkin, J. H. & Simon, H. A. (1987). “Why a Diagram is
   (Sometimes) Worth Ten Thousand Words,” Cognitive
   Science11:65-99.
Lederberg, J. (1965). “Signs of Life: Criterion-System of
   Exobiology,” Nature207:9-13.
Lee, Y., Buchanan, B. G., Klopman, G., Dimayuga, M. &
   Rosenkranz, H. S. (1996). “The Potential of Organ
   Specific Toxicity for Predicting Rodent Carcinogenicity,”
   Mutation Research358:37-62.
Lindsay, R. K., Buchanan, B. G., Feigenbaum, E. A. &
   Lederberg, J. (1980). Applications of Artificial Intelligence
   for Organic Chemistry: The DENDRAL Project.New
   York: McGraw Hill.
Lindsay, R. K., Buchanan, B. G., Feigenbaum, E. A. &
   Lederberg, J. (1993). “DENDRAL: A Case Study of the
   First Expert System for Scientific Hypothesis Formation,”
   Artificial Intelligence61:209-261.
Nersessian, N. J. (1992). “How Do Scientists Think?
   Capturing the Dynamics of Conceptual Change in
   Science,” in R. N. Giere (Ed.), Cognitive Models of
   Science.Minneapolis: University of Minnesota Press.
Nersessian, N. J. (1995). “Opening the Black Box:
   Cognitive Science and History of Science,” OSIRIS
   10:194-211.
Nickles, T. (Ed.) (1980). Scientific Discovery, Logic and
   Rationality.Dordrecht: Reidel.
Nickles, T. (1994). “Enlightenment Versus Romantic
   Models of Creativity in Science--and Beyond,” Creativity
   Research Journal7:277-314.
Popper, Karl (1965). The Logic of Scientific Discovery.
   New York: Harper Torchbooks.
 Provost, F. J. & Buchanan, B. G. (1995). “Inductive Policy:
   The Pragmatics of Bias Selection,” Machine Learning
   20:35-61.
Qin, Y. & Simon, H.A. (1995). “Imagery and Mental
   Models,” in J. Glasgow, H. H. Narayanan, & B.
   Chandrasekaran (Eds.), Diagrammatic Reasoning:
   Computational and Cognitive Perspectives.Menlo Park,
   CA: AAAI/MIT Press.
Richman, H. B., Staszewski, J. J. & Simon, H. A. (1995).
   “Simulation of Expert Memory using EPAM IV,” The
   Psychological Review102:305-330.
Schaffner, K. (1993). Discovery and Explanation in Biology
   and Medicine.Chicago: University of Chicago Press.
Schunn, C. D. & Dunbar, K. (1996). “Priming, Analogy,
   and Awareness in Complex Reasoning,” Memory &
   Cognition24:271-284.
Shrager, J. & Langley, P. (Eds.) (1990). Computational
   Models of Scientific Discovery and Theory Formation.
   San Mateo, California: Morgan Kaufmann.
Simon, H.A. (1977). Models of Discovery.Dordrecht:
   Reidel.
Simon, H.A. and Kotovsky, K. (1963). “Human
   Acquisition of Concepts for Sequential Patterns,”
   Psychological Review70:534-546.
Simon, H.A., Valdes-Perez, R.E. & Sleeman, D.H. (1997,
   forthcoming). Editorial: “Scientific Discovery and
   Simplicity of Method,” Artificial Intelligence91(2).
Spirtes, P., Glymour, C. & Scheines, R. (1993).
   Causation, Prediction, and Search.New York: Springer
   Verlag.
Swanson, D.R. & Smalheiser, D.H. (1997, forthcoming).
   “An Interactive System for Finding Complementary
   Literatures: A Stimulus to Scientific Discovery,” Artificial
   Intelligence91(2).
Tabachneck-Schijf, H. J. M., Leonardo, A. M., Simon,
   H.A. (1996). CaMeRa: A Computational Model of
   Multiple Presentations(Tech. Rep. CIP 530). Pittsburgh,
   PA: Carnegie Mellon University.
Valdes-Perez, R.E. (1992). “Algorithm to Generate Reaction
   Pathways for Computer-Assisted Elucidation,” Journal of
   Computational Chemistry13:1079-1088.
Valdes-Perez, R.E. (1992). “A Necessary Condition for
   Catalysis in Reaction Pathways,” Journal of Physical
   Chemistry96:2394-2396.
Valdes-Perez, R.E. (1993). “Algorithm to Test the
   Structural Plausibility of a Proposed Elementary
   Reaction,” Journal of Computational Chemistry14:1454-
   1459.
Valdes-Perez, R.E. (1994). “Algorithm to Infer the
   Structures of Molecular Formulas within a Reaction
   Pathway,” Journal of Computational Chemistry15:1266-
   1277.
Valdes-Perez, R.E. (1994). “Human/computer Interactive
   Elucidation of Reaction Mechanisms: Application to
   Catalyzed Hydrogenolysis of Ethane,” Catalysis Letters
   28:79-87.
Valdes-Perez, R.E. (1994). “Heuristics for Systematic
   Elucidation of Reaction Pathways,” Journal of Chemical
   Information and Computer Sciences34:976-983.

165
Valdes-Perez, R.E. (1994). “Conjecturing Hidden Entities
   via Simplicity and Conservation Laws: Machine
   Discovery in Chemistry,” Artificial Intelligence65:247-
   280.
Valdes-Perez, R.E. (1995). “Generic Tasks of Scientific
   Discovery,” in R. Valdes-Perez (Ed.) Working Notes:
   AAAI Spring Symposium Series: Symposium:
   Systematic Methods of Scientific Discovery.
Valdes-Perez, R.E. (1995). “Some Recent Human/Computer
   Discoveries in Science and What Accounts for Them,” AI
   Magazine16(3):37-44.
Valdes-Perez, R.E. (1995). “Machine Discovery in
   Chemistry: New Results,” Artificial Intelligence74:191-
   201.
Valdes-Perez, R.E. (1995). “Machine Discovery Praxis,”
   Foundations of Science1:219-224.
Valdes-Perez, R.E. (1996). “A New Theorem in Particle
   Physics Enabled by Machine Discovery,” Artificial
   Intelligence,82:331-339.
Valdes-Perez, R.E. (1996). “Computer Science Research on
   Scientific Discovery,” Knowledge Engineering Review
   11:57-66.
Valdes-Perez, R.E. & Minden, J.S. (1995). “Drosophila
   Melanogaster Syncytial Nuclear Divisions are Patterned:
   Time-Lapse Images, Hypothesis, and Computational
   Evidence,” Journal of Theoretical Biology175:525-532.
Valdes-Perez, R.E. & Zytkow, J.M. (1996). “Systematic
   Generation of Constituent Models of Particle Families,”
   Physical ReviewE 54:2102-2110.
Zembowicz, R. & Zytkow, J.M. (1996). “From
   Contingency Tables to Various Forms of Knowledge in
   Databases,” in U. Fayyad, G. Piatetsky-Shapiro, P.
   Smyth, & R. Uthurusany (Eds.), Advances in Knowledge
   Discovery and Data Mining,Menlo Park, CA: AAAI
   Press.
Zytkow, J.M. (1993). “Introduction: Cognitive Autonomy
   in Machine Discovery,” Machine Learning: Special Issue
   on Machine Discovery12:7-16.
Zytkow, J.M. (1995/96). “Creating a Discoverer:
   Autonomous Knowledge Seeking Agent,” Foundations of
   Science2:253-283.
Zytkow, J.M. (1996). “Automated Discovery of Empirical
   Laws,” Fundamenta Informaticae27:299-318.
Zytkow, J.M. (Ed.) (1996). Machine Discovery.Kluwer
   Academic Publishers: Dordrecht.
Zytkow, J.M. & Lewenstam, A. (1990). “Analytical
   Chemistry; The Science of Many Models,” Fresenius
   Journal of Analytical Chemistry338:225-233.
Zytkow J.M., Zhu, J. & Hussam, A. (1990). “Automated
   Discovery in a Chemistry Laboratory,” in Proceedings of
   the Eighth National Conference on Artificial Intelligence,
   pp. 889-894. Menlo Park, CA: AAAI Press.
Zytkow, J.M. & Zembowicz, R. (1993). “Database
   Exploration in Search of Regularities,” Journal of
   Intelligent Information Systems,2:39-81.
Zytkow, J.M. & Zembowicz, R. (1996). “Automated
   Pattern Mining with a Scale Dimension,” in Proceedings
   of the Second International Conference on Knowledge
   Discovery and Data Mining,pp. 889-894. Menlo Park,
   CA: AAAI Press.

166