Academia.eduAcademia.edu

Outline

Title
Abstract
FAQs
Figures
References
All Topics
Computer Science
Software

The automation of science

Profile image of wayne aubreywayne aubrey

2009, Science

Abstract

We have demonstrated the discovery of physical laws, from scratch, directly from experimentally captured data with the use of a computational search. We used the presented approach to detect nonlinear energy conservation laws, Newtonian force laws, geometric invariants, and system manifolds in various synthetic and physically implemented systems without prior knowledge about physics, kinematics, or geometry. The concise analytical expressions that we found are amenable to human interpretation and help to reveal the physics underlying the observed phenomenon. Many applications exist for this approach, in fields ranging from systems biology to cosmology, where theoretical gaps exist despite abundance in data.

FAQs

sparkles

AI

What are the core functionalities of robot scientist Adam?add

Robot scientist Adam autonomously generates and tests hypotheses, executing high-throughput experiments in microbial genomics, measuring over 6.6 million data points.

How does Adam formalize scientific experiments?add

Adam uses a customized ontology called LABORS, which organizes over 10,000 research units in a structured, logic-based format to enhance reproducibility.

What methodological innovations differentiate Adam from traditional research methods?add

Adam implements laboratory robotics for hypothesis-driven experimentation, allowing automated cycles without human intervention, significantly expanding experimental throughput.

What role does DIR1 play in systemic acquired resistance (SAR)?add

DIR1 is essential for SAR and exudate-induced resistance, serving as a putative signal carrier in lipid transfer protein pathways.

How do Adam's findings challenge existing knowledge about orphan enzymes?add

Adam identified genes encoding orphan enzymes backed by statistical significance (P < 0.05), improving understanding of yeast metabolic pathways.

Figures (6)
train and defined-growth-medium experiments each day (from a selection of thousands of yeast strains), with each experiment lasting up to 5 days. The Jesign enables measurement of ODsos5nm for each experiment at least once avery 30 min (more often if running at less than full capacity), allowing ac- curate growth curves to be recorded (typically we take over a hundred mea- surements a day per well), plus associated metadata. See the supporting online material for pictures and a video of Adam in action. Use of robot scientist enables all aspects ofa scientific investigation to be formalized in logic. For the core organization of this formalization, we used the ontology of scientific experiments: EXPO (//, 12). This ontology formalizes generic knowledge about experiments. For Adam, we developed LABORS, a customized version of EXPO, expressed in the description logic lan- guage OWL-DL (/7). Application of LABORS produces experimental descriptions in the logic- reading frames (ORFs), ~800 metabolites] (5), expressed in the logic programming language Prolog; (ii) a general bioinformatic database of genes and proteins involved in metabolism; (iii) software to abduce hypotheses about the genes encoding the orphan enzymes, done by using a combination of standard bioinformatic software and databases; (iv) software to deduce experi- ments that test the observational consequences of hypotheses (based on the model); (v) software to plan and design the experiments, which are based on the use of deletion mutants and the addition of selected metabolites to a defined growth medium; (vi) laboratory automation software to physically execute the experimental plan and to record the data and metadata in a relational database; (vii) software to analyze the data and metadata (gen- erate growth curves and extract parameters); and (vili) software to relate the analyzed data to the hypotheses; for example, statistical methods are required to decide on significance. Once this in- frastructure is in place, no human intellectual inter- vention is necessary to execute cycles of simple hypothesis-led experimentation. [For more details of the software, and its application to a related functional genomics problem, see (/6) and figs. S1 and 82]. To further test Adam's conclusions, we ex- amined the scientific literature on the 20 genes investigated (Table 1) (76). This revealed the ex- istence of strong empirical evidence for the cor- rectness of six of the hypotheses; that is, the enzymes were not actually orphans (Table 1).
train and defined-growth-medium experiments each day (from a selection of thousands of yeast strains), with each experiment lasting up to 5 days. The Jesign enables measurement of ODsos5nm for each experiment at least once avery 30 min (more often if running at less than full capacity), allowing ac- curate growth curves to be recorded (typically we take over a hundred mea- surements a day per well), plus associated metadata. See the supporting online material for pictures and a video of Adam in action. Use of robot scientist enables all aspects ofa scientific investigation to be formalized in logic. For the core organization of this formalization, we used the ontology of scientific experiments: EXPO (//, 12). This ontology formalizes generic knowledge about experiments. For Adam, we developed LABORS, a customized version of EXPO, expressed in the description logic lan- guage OWL-DL (/7). Application of LABORS produces experimental descriptions in the logic- reading frames (ORFs), ~800 metabolites] (5), expressed in the logic programming language Prolog; (ii) a general bioinformatic database of genes and proteins involved in metabolism; (iii) software to abduce hypotheses about the genes encoding the orphan enzymes, done by using a combination of standard bioinformatic software and databases; (iv) software to deduce experi- ments that test the observational consequences of hypotheses (based on the model); (v) software to plan and design the experiments, which are based on the use of deletion mutants and the addition of selected metabolites to a defined growth medium; (vi) laboratory automation software to physically execute the experimental plan and to record the data and metadata in a relational database; (vii) software to analyze the data and metadata (gen- erate growth curves and extract parameters); and (vili) software to relate the analyzed data to the hypotheses; for example, statistical methods are required to decide on significance. Once this in- frastructure is in place, no human intellectual inter- vention is necessary to execute cycles of simple hypothesis-led experimentation. [For more details of the software, and its application to a related functional genomics problem, see (/6) and figs. S1 and 82]. To further test Adam's conclusions, we ex- amined the scientific literature on the 20 genes investigated (Table 1) (76). This revealed the ex- istence of strong empirical evidence for the cor- rectness of six of the hypotheses; that is, the enzymes were not actually orphans (Table 1).
of metabolites tested. Existing annotation is the summary from the Saccharomyces Genome Database of the annotation of the ORF. Dry is the summary of whether the annotated function is the same as predicted by Adam. If a gene already has an associated function, we do not consider this to be contradictory to Adam's conclusions unless this function is capable of explaining the observed growth phenotype, for example, BCY1. ida indicates inferred from direct assay and iss, inferred from sequence or structural similarity (5). Wet is the result of our manual enzyme assays. See (16) for details. Table 1. The orphan enzymes and Adam's hypotheses. The hypothesized genes are those which Adam abduced encoded an orphan enzyme. Prob. is Adam's Monte Carlo estimate of the probability of obtaining the observed discrimination accuracy or better with a random labeling of replicates. The discrimination is between the differences in growth curves observed with the addition of specified metabolites to the wild type and the deletant. Acc. is the highest accuracy for a metabolite species in discriminating between the growth curves observed with the addition of specified metabolites to the wild type and the deletant. No. is the number
of metabolites tested. Existing annotation is the summary from the Saccharomyces Genome Database of the annotation of the ORF. Dry is the summary of whether the annotated function is the same as predicted by Adam. If a gene already has an associated function, we do not consider this to be contradictory to Adam's conclusions unless this function is capable of explaining the observed growth phenotype, for example, BCY1. ida indicates inferred from direct assay and iss, inferred from sequence or structural similarity (5). Wet is the result of our manual enzyme assays. See (16) for details. Table 1. The orphan enzymes and Adam's hypotheses. The hypothesized genes are those which Adam abduced encoded an orphan enzyme. Prob. is Adam's Monte Carlo estimate of the probability of obtaining the observed discrimination accuracy or better with a random labeling of replicates. The discrimination is between the differences in growth curves observed with the addition of specified metabolites to the wild type and the deletant. Acc. is the highest accuracy for a metabolite species in discriminating between the growth curves observed with the addition of specified metabolites to the wild type and the deletant. No. is the number
FIg. 2. ASsay results Tor CAZUA aCctiv- ity. The proteins encoded by YGL202W, YJLO60W, YER152C, and YDL168W were expressed from OpenBiosystems (www. openbiosystems.com) yeast ORF clones and purified. Activity was tested in an assay of NADPH (reduced form of nico- tinamide adenine dinucleotide phosphate) production based on (22). t-c.-aminoadipic acid and 2-oxoglutarate were provided as substrates and pyridoxal phosphate as cofactor. Glutamate production was assayed by using commercially available yeast glutamate dehydrogenase, which uses NADP as cofactor and deaminates Absorbance at 340nm -¥- control YDL168w 1 -P> no protein added “Ar YGL202w -E YER152C — YJLosow 0 5 10 15 20 Time (min) glutamate, producing ammonia and NADPH and regenerating 2-oxoglutarate (16). Also consister 2A20A activity is experimental evidence indicating a higher activity with -c-aminoadipic acid over alanine or aspartate (26). A major motivation for the formalization of experimental knowledge is the expectation that such knowledge is more easily reused to answer other scientific questions. To test this, we investi- gated whether we could reuse Adam’s functional genomic research (/6). An example question investigated was the relative growth rates (Umax) in rich and defined media of the deletion strains compared with those of the wild type. What was observed, in both media, was a skewed dis- tribution, with a few deletants having a much lower Umax than that of the wild type, but most having a slightly higher Unax. These observations question the common assumption that wild-type S. cerevisiae is optimized for Umax and provide quantitative test data for yeast systems biology models (/9).
FIg. 2. ASsay results Tor CAZUA aCctiv- ity. The proteins encoded by YGL202W, YJLO60W, YER152C, and YDL168W were expressed from OpenBiosystems (www. openbiosystems.com) yeast ORF clones and purified. Activity was tested in an assay of NADPH (reduced form of nico- tinamide adenine dinucleotide phosphate) production based on (22). t-c.-aminoadipic acid and 2-oxoglutarate were provided as substrates and pyridoxal phosphate as cofactor. Glutamate production was assayed by using commercially available yeast glutamate dehydrogenase, which uses NADP as cofactor and deaminates Absorbance at 340nm -¥- control YDL168w 1 -P> no protein added “Ar YGL202w -E YER152C — YJLosow 0 5 10 15 20 Time (min) glutamate, producing ammonia and NADPH and regenerating 2-oxoglutarate (16). Also consister 2A20A activity is experimental evidence indicating a higher activity with -c-aminoadipic acid over alanine or aspartate (26). A major motivation for the formalization of experimental knowledge is the expectation that such knowledge is more easily reused to answer other scientific questions. To test this, we investi- gated whether we could reuse Adam’s functional genomic research (/6). An example question investigated was the relative growth rates (Umax) in rich and defined media of the deletion strains compared with those of the wild type. What was observed, in both media, was a skewed dis- tribution, with a few deletants having a much lower Umax than that of the wild type, but most having a slightly higher Unax. These observations question the common assumption that wild-type S. cerevisiae is optimized for Umax and provide quantitative test data for yeast systems biology models (/9).
It could be argued that the scientific knowl- edge “discovered” by Adam is implicit in the formulation of the problem and is therefore not novel. This argument that computers cannot originate anything is known as Lady Lovelace’s objection (20): “The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform” (her italics). We accept that the knowledge automatically generated by Adam is of a modest kind. However, this knowledge is not trivial, and in the case of the genes encoding 2A20A, it sheds light on, and perhaps solves, a 50-year-old puzzle (2/).
It could be argued that the scientific knowl- edge “discovered” by Adam is implicit in the formulation of the problem and is therefore not novel. This argument that computers cannot originate anything is known as Lady Lovelace’s objection (20): “The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform” (her italics). We accept that the knowledge automatically generated by Adam is of a modest kind. However, this knowledge is not trivial, and in the case of the genes encoding 2A20A, it sheds light on, and perhaps solves, a 50-year-old puzzle (2/).
hole plant immunity, called systemic \ ,\ / acquired resistance (SAR), often de- velops after localized foliar infections by diverse pathogens. In this process, leaves dis- tal to the localized infection become primed to activate a stronger defense response upon sec- ondary infection (/). Leaves infected with SAR- inducing bacteria produce vascular sap, called petiole exudate, which confers disease resistance to previously unexposed (naive) plants (2, 3). This indicates that a mobile systemic signal(s) is involved in SAR (4). Although the hormone jasmonic acid Fig. 1. Azelaic acid specifically confers resistance to Pseudomonas syringae. (A) Azelaic acid-induced resistance is concentration-dependent. Plants were sprayed with 1, 10, 100, and 1000 uM azelaic acid in 5 mM MES (pH 5.6) or 5 mM MES (pH 5.6) alone 2 days before infection with P. syringae pv. maculicola strain PmaDG3 (OD¢o9 = 0.0001). (B) Induced resistance is time-dependent. Plants sprayed with 5 mM MES or 1 mM azelaic acid for the time periods indicated were subsequently inoculated with PmaDG3. (C) 5 mM MES or 1 mM azelaic acid was injected into local leaves. Two days later, either local or systemic leaves were infected with PmaDG3. (D) Dicarboxylic acids (1 mM) of different carbon-chain lengths were applied to Arabidopsis. M, 5 mM Mes; Cg, suberic acid; Co, azelaic acid; C19, sebacic acid. (E) Mobility of deuterium-labeled azelaic acid [HOOC(CD,),COOH] injected into WT leaves. Azelaic acid amounts were determined in petiole exudates (left) and distal leaves (right) after local injection with 1 mM azelaic acid. *P < 0.05; **P < 0. 01; ¢ test. Error bars indicate SE.
hole plant immunity, called systemic \ ,\ / acquired resistance (SAR), often de- velops after localized foliar infections by diverse pathogens. In this process, leaves dis- tal to the localized infection become primed to activate a stronger defense response upon sec- ondary infection (/). Leaves infected with SAR- inducing bacteria produce vascular sap, called petiole exudate, which confers disease resistance to previously unexposed (naive) plants (2, 3). This indicates that a mobile systemic signal(s) is involved in SAR (4). Although the hormone jasmonic acid Fig. 1. Azelaic acid specifically confers resistance to Pseudomonas syringae. (A) Azelaic acid-induced resistance is concentration-dependent. Plants were sprayed with 1, 10, 100, and 1000 uM azelaic acid in 5 mM MES (pH 5.6) or 5 mM MES (pH 5.6) alone 2 days before infection with P. syringae pv. maculicola strain PmaDG3 (OD¢o9 = 0.0001). (B) Induced resistance is time-dependent. Plants sprayed with 5 mM MES or 1 mM azelaic acid for the time periods indicated were subsequently inoculated with PmaDG3. (C) 5 mM MES or 1 mM azelaic acid was injected into local leaves. Two days later, either local or systemic leaves were infected with PmaDG3. (D) Dicarboxylic acids (1 mM) of different carbon-chain lengths were applied to Arabidopsis. M, 5 mM Mes; Cg, suberic acid; Co, azelaic acid; C19, sebacic acid. (E) Mobility of deuterium-labeled azelaic acid [HOOC(CD,),COOH] injected into WT leaves. Azelaic acid amounts were determined in petiole exudates (left) and distal leaves (right) after local injection with 1 mM azelaic acid. *P < 0.05; **P < 0. 01; ¢ test. Error bars indicate SE.
*Present address: Boyce Thompson Institute for Plant Research, Tower Road, Ithaca, NY 14853-1801, USA. tTo whom correspondence should be addressed. E-mail: jgreenbe@midway.uchicago.edu
*Present address: Boyce Thompson Institute for Plant Research, Tower Road, Ithaca, NY 14853-1801, USA. tTo whom correspondence should be addressed. E-mail: jgreenbe@midway.uchicago.edu

Related papers

AI Descartes: Combining Data and Theory for Derivable Scientific Discovery
Lior Horesh

2021

Scientists have long aimed to discover meaningful formulae which accurately describe experimental data. One common approach is to manually create mathematical models of natural phenomena using domain knowledge, then fit these models to data. In contrast, machine-learning algorithms automate the construction of accurate data-driven models while consuming large amounts of data. Ensuring that such models are consistent with existing knowledge is an open problem. We develop a method for combining logical reasoning with symbolic regression, enabling principled derivations of models of natural phenomena. We demonstrate these concepts for Kepler’s third law of planetary motion, Einstein’s relativistic time-dilation law, and Langmuir’s theory of adsorption, automatically connecting experimental data with background theory in each case. We show that laws can be discovered from few data points when using formal logical reasoning to distinguish the correct formula from a set of plausible formu...

View PDFchevron_right
The Future of Fundamental Science Led by Generative Closed-Loop Artificial Intelligence
Larisa Soldatova

arXiv (Cornell University), 2023

Recent machine learning and AI advances disrupt scientific practice, technological innovation, product development, and society. As a rule, success in classification, pattern recognition, and gaming occurs whenever there are clear performance evaluation criteria and access to extensive training data sets. Yet, AI has contributed less to fundamental science, such as discovering new principled explanatory models and equations. To set the stage for a fundamental AI4Science, we explore a perspective for an AI-driven, automated, generative, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space. Generative AI, in general, and Large Language Models (LLMs), in particular, serve here to translate and break down high-level human or machine conjectures into smaller computable modules inserted in the automated loop. Discovering fundamental explanatory models requires causality analysis while enabling unbiased efficient search across the space of putative causal explanations. In addition, integrating AI-driven automation into the practice of science would mitigate current problems, including the replication of findings, systematic production of data, and ultimately democratisation of the scientific process. These advances promise to unleash AI's potential for searching and discovering the fundamental structure of our world beyond what human scientists have achieved or can achieve. Such a vision would push the boundaries of new fundamental science beyond automatizing current workflows and unleash new possibilities to solve some of humanity's most significant challenges.

View PDFchevron_right
Machine Science: What's Missing
Kevin Elliott

Science, 2010

THE PERSPECTIVE BY J. EVANS AND A. Rzhetsky ("Machine science," 23 July, p. 399) implies that the next Einstein could be a computer. During the past centuries, the epistemological debates on the scientifi c trends and revolutionary aspects in the scientifi c breakthroughs of every century showed the centrality of the investigators and their impressive capabilities (and sometimes their luck). The use of computer-aided simulations and analysis has been an important part of modern science; technological developments have led to shorter computational times and the opportunity for scientists to develop an increas ed number of simulations, mathematical models, and scientifi c calculi. However, despite the effectiveness of computer tools, the centrality of the researchers has not changed in science. Scientists defi ne the computation of models, the analysis of data, and the validation of scientifi c hypotheses, as well as the guidelines for ad hoc software and large-scale computations. Technology is a useful tool for the scientists, but it cannot solve open problems such as the Riemann Conjecture. "Machine science" could more accurately be considered "machine-aided science." Science will not fundamentally change until a machine can produce results or solve open problems without human direction. The Perspective describes our current scientifi c methodology, not the beginning of a new scientifi c era.

View PDFchevron_right
Toward Building Science Discovery Machines
Abdullah Khalili

arXiv (Cornell University), 2021

The dream of building machines that can do science has inspired scientists for decades. Remarkable advances have been made recently; however, we are still far from achieving this goal. In this paper, we focus on the scientific discovery process where a high level of reasoning and remarkable problem-solving ability are required. We review machine learning techniques used in scientific discovery with their limitations. We survey and discuss the main principles driving the scientific discovery process. These principles are used in different fields and by different scientists to solve problems and discover new knowledge. We provide many examples of the use of these principles in different fields such as physics, mathematics, and biology. We also review AI systems that attempt to implement some of these principles. We argue that building science discovery machines should be guided by these principles as an alternative to the dominant approach of current AI systems that focuses on narrow objectives. Building machines that fully incorporate these principles in an automated way might open the doors for many advancements.

View PDFchevron_right
An Autonomous, Emergent Model of Fundamental Physics: Understanding the Universe by Designing It
Randall O'Reilly

2006

Abstract: A good way to really understand something is to try to build it yourself from scratch. This paper takes this approach to understanding the physical universe. We seek to build an autonomous model that produces known physics in a completely self-contained manner, as a way of understanding how nature works, automatically and without any human or other form of intervention, to produce the known universe.

View PDFchevron_right
Towards Robot Scientists for autonomous scientific discovery
wayne aubrey

2010

We review the main components of autonomous scientific discovery, and how they lead to the concept of a Robot Scientist. This is a system which uses techniques from artificial intelligence to automate all aspects of the scientific discovery process: it generates hypotheses from a computer model of the domain, designs experiments to test these hypotheses, runs the physical experiments using robotic systems, analyses and interprets the resulting data, and repeats the cycle. We describe our two prototype Robot Scientists: Adam and Eve. Adam has recently proven the potential of such systems by identifying twelve genes responsible for catalysing specific reactions in the metabolic pathways of the yeast Saccharomyces cerevisiae. This work has been formally recorded in great detail using logic. We argue that the reporting of science needs to become fully formalised and that Robot Scientists can help achieve this. This will make scientific information more reproducible and reusable, and promote the integration of computers in scientific reasoning. We believe the greater automation of both the physical and intellectual aspects of scientific investigations to be essential to the future of science. Greater automation improves the accuracy and reliability of experiments, increases the pace of discovery and, in common with conventional laboratory automation, removes tedious and repetitive tasks from the human scientist.

View PDFchevron_right
Automating science: Philosophical and social dimensions
Chris Wood, Vlad Schüler-Costa

Clark Glymour argued in 2004 that "despite a lack of public fanfare, there is mounting evidence that we are in the midst of ... a revolution - premised on the automation of scientific discovery". This paper highlights some of the philosophical and sociological dimensions that have been found empirically in work conducted with robot scientists - that is, with autonomous robotic systems for scientific discovery. Robot scientists do not supply definite answers to the discussed questions, but rather provide "proofs of concept" for various ideas. For example, it is not that robot scientists solve the realist/antirealist philosophical debate, but that when working with robot scientists one has to make a philosophical choice - in this case, to assume a realist view of science. There are still few systems for autonomous scientific discovery in existence, and it is too early to generalize and propose new theories. However, being "in the midst of ... a revolution" it is important for the research community to re-examine views pertinent to scientific discovery. This paper highlights how experience with robot scientists could inform discussions in other disciplines, from philosophy of science to computer creativity research.

View PDFchevron_right
The automation of scientific research
Ruben Garcia Pedraza

The Automation of Scientific Research, 2025

This work marks the beginning of a bold and urgent proposal: the automation of scientific research through Global Artificial Intelligence (GAI). As the first volume in a larger series, it introduces and develops the emerging concept of GAI—a system designed to think, learn, and evolve beyond the limitations of individual human researchers. Rooted in the theory of Impossible Probability, these foundational papers lay the groundwork for a new era of knowledge production, where intelligence is not only simulated but truly global, collaborative, and self-improving.

View PDFchevron_right
Unifying Science Through Computation: Reflections on Computability and Physics
John V Tucker

Special Sciences and the Unity of Science, 2011

Many activities of a contemporary working scientist involve the idea of the unity of science. There are countless examples where the ideas and methods of one subject find application in another. There are subjects that comfortably straddle the border between disciplines. There are problems that can only be tackled by multidisciplinary approaches. Science is a loose federation of diverse intellectual, experimental and material communities and cultures. However, these cultures are strong. In this paper we reflect upon an area of research that is attracting the attention of computer scientists, mathematicians, physicists and philosophers: the relationship between theories of computation and physical systems. There are intriguing questions about the computability of physics, and the physical foundations of computability, that can set the agenda for a new subject, and that will not go away. Research is in an early phase of its development, but has considerable potential and ambition. First, we will argue that concepts of computability theory have a natural place in physical descriptions. We will look at incomputability and (i) the idea that computers "exist" in Nature, (ii) models of physical systems and notions of prediction, and (iii) hypercomputation. We will reflect upon computability and physics as an example of work crossing the frontiers of two disciplines, introducing new questions and ways of argument in physics, and enabling a reappraisal of computers and computation. We will also notice the social phenomenon of suspicion and resistance, as the theories are unbalanced by their encounter with one another. 8 The reader can have a look at the Eddington's Cosmic Equation, that was a source of explanation of how the Universe came into existence. 9 Though, in fact, for some time, physicists were tempted to define the law 1 r 2.0•••025. Ridiculous, isn't it? But it worked for a few years, when physicists lost their faith for reasons that will become clear soon.

View PDFchevron_right
(Non)-neutrality of science and algorithms: Machine Learning between fundamental physics and society
Aniello Lampo

ArXiv, 2020

The impact of Machine Learning (ML) algorithms in the age of big data and platform capitalism has not spared scientific research in academia. In this work, we will analyse the use of ML in fundamental physics and its relationship to other cases that directly affect society. We will deal with different aspects of the issue, from a bibliometric analysis of the publications, to a detailed discussion of the literature, to an overview on the productive and working context inside and outside academia. The analysis will be conducted on the basis of three key elements: the non-neutrality of science, understood as its intrinsic relationship with history and society; the non-neutrality of the algorithms, in the sense of the presence of elements that depend on the choices of the programmer, which cannot be eliminated whatever the technological progress is; the problematic nature of a paradigm shift in favour of a data-driven science (and society). The deconstruction of the presumed universalit...

View PDFchevron_right

References (49)

  1. References and Notes
  2. P. W. Anderson, Science 177, 393 (1972).
  3. E. Noether, Nachr. d. König Gesellsch. d. Wiss. zu Göttingen, Math-Phys. Klasse 235 (1918).
  4. J. Hanc, S. Tuleja, M. Hancova, Am. J. Phys. 72, 428 (2004).
  5. D. Clery, D. Voss, Science 308, 809 (2005).
  6. A. Szalay, J. Gray, Nature 440, 413 (2006).
  7. R. E. Valdés-Pérez, Commun. Assoc. Comput. Mach. 42, 37 (1999).
  8. R. D. King et al., Nature 427, 247 (2004).
  9. P. Langley, Cogn. Sci. 5, 31 (1981).
  10. R. M. Jones, P. Langley, Comput. Intell. 21, 480 (2005).
  11. J. R. Koza, Genetic Programming: On the Programming of Computers by Means of Natural Selection. (MIT Press, Cambridge, MA, 1992).
  12. S. Forrest, Science 261, 872 (1993).
  13. J. Duffy, J. Engle-Warnick, Evolutionary Computation in Economics and Finance 100, 61 (2002).
  14. F. Cyril, B. Alberto, in 2007 IEEE Congress on Evolutionary Computation, S. Dipti, W. Lipo, Eds. (IEEE Press, Singapore, 2007), pp. 23-30.
  15. B. Elena, B. Andrei, L. Henri, in Seventh International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC '05) (IEEE Press, 2005), pp. 321-324.
  16. J. Bongard, H. Lipson, Proc. Natl. Acad. Sci. U.S.A. 104, 9943 (2007).
  17. S. Nee, N. Colegrave, S. A. West, A. Grafen, Science 309, 1236 (2005).
  18. P. Jäckel, T. Mullin, Proc. R. Soc. London Ser. A 454, 3257 (1998).
  19. T. Shinbrot, C. Grebogi, J. Wisdom, J. A. Yorke, Am. J. Phys. 60, 491 (1992).
  20. Y. Liang, B. Feeny, Nonlinear Dyn. 52, 181 (2008).
  21. M. Mor, A. Wolf, O. Gottlieb, in Proceedings of the 21st ASME Biennial Conference on Mechanical Vibration and Noise (ASME Press, Las Vegas, NV, 2007), pp. 1-8.
  22. P. Gregory, R. Denis, F. Cyril, in Evolution Artificielle, 6th International Conference, vol. 2936, L. Pierre, C. Pierre, F. Cyril, L. Evelyne, S. Marc, Eds. (Springer, Marseilles, France, 2003), pp. 267-277.
  23. E. D. De Jong, J. B. Pollack, in Genetic Programming and Evolvable Machines, vol. 4 (Springer, Berlin, 2003), pp. 211-233.
  24. S. H. Strogatz, Nature 410, 268 (2001).
  25. P. A. Marquet, Nature 418, 723 (2002).
  26. This research was supported in part by Integrative Graduate Education and Research Traineeship program in nonlinear systems, a U.S. NSF graduate research fellowship, and NSF Creative-IT grant 0757478 and CAREER grant 0547376. We thank M. Kurman for editorial consultation and substantive editing of the manuscript. References and Notes
  27. Nature 440, 383 (2006).
  28. T. Hey, A. Trefethen, in Grid Computing: Making the Global Infrastructure a Reality, 36 809 (Wiley, New York, 2003).
  29. S. Toulmin, in Encyclopaedia Britannica Deluxe Edition 2004 CD (Encyclopaedia Britannica UK, London, 2003).
  30. T. Berners-Lee, J. Hendler, O. Lassila, Sci. Am. 284, 34 (May 2001).
  31. M. Ashburner et al., Nat. Genet. 25, 25 (2000).
  32. R. D. King et al., Nature 427, 247 (2004).
  33. B. G. Buchanan, G. L. Sutherland, E. A. Feigenbaum, in Machine Intelligence, B. Meltzer, D. Michie, Eds. (Edinburgh Univ. Press, Edinburgh, UK, 1969), vol. 5, pp. 209-254.
  34. P. Langley, H. A. Simon, G. L. Bradshaw, J. M. Zytkow, Scientific Discovery: Computational Explorations of the Creative Process (MIT Press, Cambridge, MA, 1987).
  35. J. M. Zytkow, J. Zhu, A. Hussam, in Proceedings of the Eighth National Conference on Artificial Intelligence (AAAI-90), 889 (AAAI Press, Menlo Park, CA, 1990).
  36. L. Hood, J. R. Heath, M. E. Phelps, B. Lin, Science 306, 640 (2004).
  37. L. N. Soldatova, R. D. King, J. R. Soc. Interface 3, 795 (2006).
  38. L. N. Soldatova, A. Clare, A. Sparkes, R. D. King, Bioinformatics 22, e464 (2006).
  39. R. D. King et al., The Robot Scientist Project, www.aber. ac.uk/compsci/Research/bio/robotsci/ (2008).
  40. J. Warringer, E. Ericson, L. Fernandez, O. Nerman, A. Blomberg, Proc. Natl. Acad. Sci. U.S.A. 100, 15724 (2003).
  41. K. E. Whelan, R. D. King, BMC Bioinformatics 9, 97 (2008).
  42. Materials and methods are available as supporting material on Science Online.
  43. I. Horrocks, P. F. Patel-Schneider, F. van Harmelen, Web Semantics 1, 7 (2003).
  44. J. D. Ullman, Principles of Database and Knowledge-Base Systems, Vol. I: Classical Database Systems (Computer Science, New York, 1988).
  45. M. J. Herrgard et al., Nat. Biotechnol. 26, 1155 (2008).
  46. A. Turing, Mind 236, 433 (1950).
  47. T. M. Zabriskie, M. D. Jackson, Nat. Prod. Rep. 17, 85 (2000).
  48. M. Young, J. Exp. Bot. 24, 1172 (1973).
  49. This work was funded by grants from the Biotechnology and Biological Sciences Research Council to R.D.K., S.G.O., and M.Y.; by a SRIF 2 award to R.D.K.; by fellowships from the Royal Commission for the Great

Related papers

Automated discovery of scientific concepts: Replicating three recent discoveries in mechanics
yaron hakuk

Advanced Engineering Informatics, 2020

Throughout history, humanity cherished new scientific discoveries; many people devoted their lives to find them in laborious manual processes. Since the emergence of artificial intelligence, about 60 years ago, the possibility of automated scientific discovery has become conceivable and has attracted significant and growing interest. Many studies of automated scientific discovery have been presented over the years. This article presents a new automated discovery approach that crosses disciplines and transfers knowledge between them. The approach requires rich and formal background knowledge to find concepts, methods, or laws not known in one discipline by using their counterpart in other disciplines, specifically disciplinary knowledge that is represented in the Interdisciplinary Engineering Knowledge Genome. Three recent discoveries in mechanics were replicated automatically through software execution, demonstrating the validity of the approach. In future studies, the goal is to discover new knowledge in mechanics and/or electronics, as well as venture into other disciplines including outside engineering.

View PDFchevron_right
Automating Sciences: Philosophical and Social Dimensions
Chris Mellingwood

IEEE Technology and Society Magazine, 2018

Clark Glymour argued in 2004 that "despite a lack of public fanfare, there is mounting evidence that we are in the midst of … a revolution-premised on the automation of scientific discovery" [1]. This paper highlights some of the philosophical and sociological dimensions that have been found empirically in work conducted with robot scientists-that is, with autonomous robotic systems for scientific discovery. Robot scientists do not bring definite answers to the discussed questions, but rather provide "proofs of concept" for various ideas. For example, it is not that robot scientists solve the realist/antirealist philosophical debate, but that when working with robot scientists one has to make a philosophical choice-in this case, to assume a realist view of science. There are still few systems for autonomous scientific discovery in existence, and it is too early to generalize and propose new theories. However, being "in the midst of … a revolution" it is important for the research community to reexamine views pertinent to scientific discovery. This paper highlights how experience with robot scientists could inform discussions in other disciplines, from philosophy of science to computer creativity research. Scientific Discovery and Robot Scientists The branch of Artificial Intelligence (AI) devoted to developing algorithms for acquiring scientific knowledge is known as "scientific discovery." The pioneering work in scientific discovery was the development of learning algorithms for analysis of mass-spectrometric data [2]. In the subsequent 50 years, much has been achieved and there are now convincing examples in which computer programs have made explicit contributions to scientific knowledge (e.g., [3], [4], [5], [6]). However, the general impact of such programs on science has been limited. This is slowly changing as the expansion of automation in science is making it increasingly possible to couple scientific discovery software to laboratory instrumentation [6], [7], [8], [9]. Science is an excellent testbed for the development of AI discovery systems: Scientific problems are abstract, but also involve the real-world level knowledge.

View PDFchevron_right
ARTIFICIAL INTELLIGENCE FOR PHYSICISTS: ADVANCING PHYSICAL UNDERSTANDING AND DISCOVERY THROUGH MACHINE LEARNING AND DATA SCIENCE
JEFERSON CHAGAS

This paper explores the transformative potential of artificial intelligence (AI) in the field of physics, emphasizing the application of machine learning (ML) and data science techniques to enhance physical understanding, modeling, and discovery. AI-driven approaches enable physicists to leverage vast computational resources and data-driven insights, opening new avenues for research and pushing the boundaries of scientific knowledge. The paper provides an overview of key AI concepts and methods relevant to physics, including symbolic regression, deep learning, reinforcement learning, and automated experimentation. It also discusses the challenges and opportunities associated with the integration of AI into physics research workflows, such as data quality, interpretability, and collaboration between physicists and AI experts. The paper presents case studies and practical examples demonstrating the successful application of AI in various subfields of physics, such as particle physics, astrophysics, and quantum computing. It concludes with a vision for the future of AI-driven physics, emphasizing the need for interdisciplinary education, open science, and responsible innovation.

View PDFchevron_right
Data-based Discovery of Governing Equations
Waad Subber

2021

Most common mechanistic models are traditionally presented in mathematical forms to explain a given physical phenomenon. Machine learning algorithms, on the other hand, provide a mechanism to map the input data to output without explicitly describing the underlying physical process that generated the data. We propose a Data-based Physics Discovery (DPD) framework for automatic discovery of governing equations from observed data. Without a prior definition of the model structure, first a free-form of the equation is discovered, and then calibrated and validated against the available data. In addition to the observed data, the DPD framework can utilize available prior physical models, and domain expert feedback. When prior models are available, the DPD framework can discover an additive or multiplicative correction term represented symbolically. The correction term can be a function of the existing input variable to the prior model, or a newly introduced variable. In case a prior mode...

View PDFchevron_right
Automated Science
Paweł Kawalec

Any serious attempt to give an account of the cognitive aspect of science – as contrasted with e.g. its economic, social or cultural aspects – cannot ignore the automation revolution. In the conception presented in this paper the results of computer science are taken seriously and integrated with earlier ideas concerning what constitutes scientific inquiry. The central idea is that of reliable inquiry. On the reliabilist criteria presented here, if a method is logically warranted to get to the right answer – given data and background knowledge available – and to stick to it afterwards, then it is reliable. This is a normative theory, but the standard of reliability is adjusted to the domain of the inquiry at hand. This paper presents the central tenets of the reliabilist conception of science and briefly outlines the main results underlying it. The philosophical task to deliver an adequate understanding of science is taken to be continuous with scientific research itself (naturalism), a major part of which is concerned with delivering causal explanations (causality) and can only be carried out with limited resources (computability). An outline of open problems and directions for future research concludes the paper.

View PDFchevron_right

Related topics

  • Robotics
  • Programming Languages
  • Artificial Intelligence
  • Software
  • Generating Function
    • 580 California St., Suite 400
    San Francisco, CA, 94104
    © 2025 Academia. All rights reserved