automating science

Upload: robgo

Post on 30-May-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/14/2019 Automating Science

    1/2

    stand to suffer further losses; in Mauritius, the

    largest nonthreatened native frugivore is the

    gray white-eye, a bird weighing a mere 0.009

    kg (see the supporting online material).

    An extended megafauna concept has the

    potential to promote synergy between other-

    wise disparate research and conservation foci,

    and to facilitate broader syntheses of eco-

    system-level effects of extinctions of the

    largest vertebrates and the resulting ecological

    shrinkage. It is high time to more fully under-

    stand and ameliorate the recent and ongoing

    losses of all the hugest, and fiercest, and

    strangest forms (13).

    References and Notes1. R. Dirzo, A. Miranda, inPlant-Animal Interactions:

    Evolutionary Ecology in Tropical and Temperate Regions,P. W. Price, T. M. Lewinshon, G. W. Fernandes, W. W.Benson, Eds. (Wiley, New York, 1991), pp. 273287.

    2. P. L. Koch, A. D. Barnosky,Ann. Rev. Ecol. Evol. Syst. 37,215 (2006).

    3. A. D. Barnosky et al.,Science 306, 70 (2004).4. D. H. Janzen, P. S. Martin,Science 215, 19 (1982).5. P. R. Guimaraes Jr., et al., PLoS ONE3, e1745 (2008).6. P. S. Martin, R. G. Klein, Quaternary Extinctions: A

    Prehistoric Revolution (Univ. of Arizona Press, Tucson,1995).

    7. R. N. Owen-Smith,Megaherbivores: The Influence of VeryLarge Body Size on Ecology(Cambridge Univ. Press, NewYork, 1988).

    8. H. J. Meehanet al.,J. Biogeogr. 29, 695 (2002).9. D. M. Hansen et al., PLoS ONE3, e2111 (2008).

    10. R. G. Colevatti et al.,Mol. Ecol. 12, 105 (2003).11. S. J. Wright et al., Biotropica 39, 289 (2007).12. W. Hallwachs, inFrugivores and Seed Dispersal, A. Estrad

    T. H. Fleming, Eds. (Kluwer, Boston, 1986), pp. 28530413. A. R. Wallace, The Geographical Distribution of Animal

    (Harper, New York, 1876).14. Supported by the Velux Foundation (D.M.H.), Fundao

    de Amparo Pesquisa do Estado de So Paulo (M.G.),and Conselho Nacional de Desenvolvimento Cientfico Tecnolgico (M.G.).

    Supporting Online Materialwww.sciencemag.org/cgi/content/full/324/5923/42/DC1Fig. S1Table S1References

    10.1126/science.11723

    www.sciencemag.org SCIENCE VOL 324 3 APRIL 2009

    CREDIT:ROBERTK.

    LINDSAY

    PERSPECT

    The idea of automating as-

    pects of scientific activity

    dates back to the roots of

    computer science, if not to Francis

    Bacon. Some of the earliest pro-

    grams automated the processes of

    creating ballistic tables, cracking

    cyphers, collecting laboratory

    data, etc., by carrying out a set of

    instructions from start to finish.

    Starting with DENDRAL in the

    1960s (1), artif icial intelligence

    programs such as Prospector (2

    ),Bacon (3), and Fahrenheit (4)

    automated some of the planning,

    analysis, and discovery portions

    of the scientific enterprise. How-

    ever, most of these programs were

    still designed to run a calculation

    to completion, produce an answer,

    and then stop. They did not fully

    close the loop in the sense of examining the

    results of their actions, deciding what to try

    next, potentially cycling forever.

    Two reports on pages 85 and 81 of this

    issue push the boundaries of automatic scien-

    tific experimentation and discovery. Kinget al. (5) describe a robotic system for running

    biological experiments, evaluating their

    results, and deciding what experiments to try

    next. Schmidt and Lipson (6) describe their

    work on discovering compact equations that

    characterize complex nonlinear dynamical

    systems, derived from visual observation of

    such systems. As these reports show, it is pos-

    sible for one computer program to step

    through the activities needed to conduct a con-

    tinuously looping procedure that starts with a

    question, carries out experiments to answerthe question, evaluates the results, and refor-

    mulates new questions.

    The main goals of automation in science

    have been to increase productivity by increas-

    ing efficiency (e.g., with rapid throughput), to

    improve quality (e.g., by reducing error), and

    to cope with scale, allowing scientific treat-

    ment of topics that were previously impossi-

    ble to address. Tycho Brahe spent a lifetime

    recording observations that allowed Johannes

    Kepler to formulate Keplers laws of planetary

    motion; today, computer-controll

    data collection is commonplac

    and necessary for both exper

    mental and observational scienc

    Automating many activities b

    yond data collection offers eve

    more benefits.

    In the near term, a useful met

    phor is to consider computers

    intelligent assistants. Some assi

    tants gather data and attend

    such tasks as noise filtering, da

    smoothing, outlier rejection, andata storage. Other assistants a

    specialists at statistical analys

    still others at bench work. Th

    metaphor has driven many r

    search projects over the past se

    eral decades and has led to man

    of the most successful applic

    tions of computers.

    An early articulation of this metaph

    is Joshua Lederbergs effort at Stanfor

    University School of Medicine to develop a

    automated biomedical laboratory to examin

    the soil of Mars for traces of life, as part

    the 1975 Viking mission deployed by thU.S. National Aeronautics and Space A

    ministration. The robot assistant Lederbe

    designed, with engineer Elliott Levintha

    consisted of a conveyor belt that scooped u

    samples of Martian soil and deposited the

    within a computer-controlled mass spectrom

    eter. Each soil sample was bombarded wi

    electrons, producing a fragmentation patte

    that sorted the charged particles (ions) accor

    ing to their mass. This pattern was transmitte

    to Earth, where scientists could analyze it f

    Computers with intelligence can design and

    run experiments, but learning from the result

    to generate subsequent experiments requires

    even more intelligence.Automating ScienceDavid Waltz1 and Bruce G. Buchanan2

    COMPUTER SCIENCE

    1Center for Computational Learning Systems, ColumbiaUniversity, New York, NY 10115, USA. 2Computer ScienceDepartment, University of Pittsburgh, Pittsburgh, PA15260, USA. E-mail: [email protected]; [email protected]

    Semiautomated. Scientists at Stanfords Instrumentation Research Laboratory(circa 1970) linked a gas chromatograph and high-resolution mass spectrome-ter to computers to automate studies of biological fluids, meteorites, and othermaterials. Stanfords DENDRAL Project experimented with automated interpreta-tion of the data and experiment planning to specify nuclear magnetic resonanceor infrared data that would resolve ambiguities in the mass spectral data.

    Published by AAAS

  • 8/14/2019 Automating Science

    2/23 APRIL 2009 VOL 324 SCIENCE www.sciencemag.org4

    PERSPECTIVES

    evidence of organic compounds and microbial

    life. In addition, part of Lederbergs vision for

    this instrumentation was also to close the loop

    by performing the analysis onboard the space-

    craft to inform a next round of experiments

    without waiting for Earth-based instructions.

    This was, in part, the motivation for the DEN-

    DRAL project at Stanford in which an intelli-

    gent assistant hypothesized the molecular

    structure of organic molecules on the basis of

    mass spectrometry data (see the figure).

    Intelligent assistants are currently numer-

    ous and well integrated into the activities of

    science and industry. In the longer term, how-

    ever, new kinds of computer programs are

    needed to cope with the sheer volume of data

    that can be collected automatically (79) and

    with the volume of relevant information avail-

    able in the literature.

    Closing the loop from experiment design

    and data collection to hypothesis formation

    and revision, and from there to new experi-

    ments, will be one important way to cope withthe volume of data. A new wave of programs

    will test the efficacy of using computers in

    closed-loop fashion and will explore the ques-

    tions of which activities can be automated, and

    which ones we would even want to automate.

    Even for the relatively straightforward task of

    data collection, there are myriad questions to

    answer before streaming data from a labora-

    tory instrument into a computer, including

    why particular data are being collected, which

    variables should be measured, and which

    instrument will measure them. If no such

    instrument exists, can it be designed and built?

    Beyond coping with the volume of data,

    however, computers need to be called into serv-

    ice to cope with the volume of information and

    background knowledge relevant to any scien-

    tific question. Search engines and automated

    libraries will return more articles in response to

    a query than anyone has time to read. (For

    example, Google returns about 200,000 hits for

    the phrase laboratory automation and 10 mil-

    lion hits for the pair of terms science +

    automation.) Programs that have the intelli-

    gence to read and interpret the online informa-

    tion for us will contribute to the next level of

    closing the loop. This is already an active area

    of computer science research (10).

    For any such program to select the most

    cost-effective and informative hypotheses,

    prune hypotheses that cannot be realized

    experimentally, avoid repeating unsuccessful

    experiments that have already been tried byothers, etc., it must include a rich model of

    the entire process of the loop, as well as

    knowledge of the specific scientific area

    being automated. This will increasingly

    involve a substantial modeling effort, as is

    already required for planning and interpret-

    ing experiments in systems biology or

    weather and climate.

    For the foreseeable future, the prospect of

    using automated systems as assistants holds

    vast promise as these assistants are becomin

    not only faster but much broader in their cap

    bilitiesmore knowledgeable, more creativ

    and more self-reflective. Human-machin

    partnering systems that match the tasks

    what each partner does best can potential

    increase the rate of scientific progress drama

    ically, in the process revolutionizing the pra

    tice of science and changing what scientis

    need to know.

    References and Notes

    1. R. K. Lindsay, B. G. Buchanan, E. A. Feigenbaum, J.

    Lederberg,Applications of Artificial Intelligence for

    Organic Chemistry: The DENDRAL Project(McGraw-Hill

    New York, 1980).

    2. R. O. Duda, J. Gaschnig, P. E. Hart, in Expert Systems in

    the Microelectronic Age, D. Michie, Ed. (Edinburgh Uni

    Press, Edinburgh, 1979), pp. 153167.

    3. P. Langley, Cognit. Sci. 5, 31 (1981).

    4. J. M. Zytkow, J. Zhu, A. Hussam, in Proceedings of the 8

    National Conference on Artificial Intelligence (AAAI Pre

    Menlo Park, CA, 1990), pp. 889894 (www.aaai.org/

    Papers/AAAI/1990/AAAI90-133.pdf).

    5. R. D. King et al.,Science 324, 85 (2009).

    6. M. Schmidt, H. Lipson,Science324

    , 81 (2009).7. C. Anderson, Wired16.07, June 2008 (www.wired.com/

    science/discoveries/magazine/16-07/pb_theory).

    8. A. Srinivasan, S. H. Muggleton, M. J. E. Sternberg, R. D

    King,Artif. Intell. 85, 277 (1996).

    9. S. Muggleton, Nature 440, 409 (2006).

    10. Machine Reading: Papers from the 2007 AAAI Spring

    Symposium (www.aaai.org/Library/Symposia/Spring/

    ss07-06.php).

    11. Supported in part by NSF grant 0738341 (B.G.B.) and

    NIH grant 5 U54 CA 121852-03 C11, and by

    Consolidated Edison Corp. grant CU08-8987 (D.W.).

    10.1126/science.11727

    Projections of global energy needs for

    sustainable development suggest a

    nearly 50% increase by 2030 (a mere

    21 years hence) (1). This increase can be met

    satisfactorily by only one kind of alternative

    energythe Sun. One approach to convert

    solar energy into a fuel is to use it to split

    water into H2 and O2. A number of strategiesfor the visible lightdriven splitting of water

    are being pursued with varying levels of suc-

    cess. On page 74 of this issue, Kohl et al. (2)

    describe a very different reaction system for

    water splitting that uses light but also has

    thermally driven steps. The basis of this

    approach, in which key steps involve ligand

    modification, shows that reactions that har-

    vest solar energy can be found among the

    unlikeliest of compounds.

    Research in solar energy conversion fol-

    lows three principal strategies. The first is the

    direct conversion of light into electrical

    energy, as in the photovoltaic (PV) devices

    that are currently being produced and installedaround the world. Challenges here include

    increasing the efficiency and durability of

    such devices while reducing their cost to make

    them competitive with cheaper but environ-

    mentally problematic coal-fired power plants.

    The Energy Information Administration of

    the U.S. Department of Energy projects that

    global use of coal for electricity will grow rel-

    ative to other sources in the next 21 years (1).

    Considerable research on dye-sensitized solar

    cells (3, 4) has made them an interesting alter-

    native to traditional silicon-based PVs, wi

    demonstration units being deployed. Effor

    also continue for new PV devices based o

    thin-film designs that use either amorphou

    silicon, cadmium telluride, copper indiu

    gallium selenide, or organic charge-transf

    compounds on flexible supports that can b

    manufactured on a massive scale (5, 6).The second strategy is to use natures ph

    tosynthetic apparatus to produce biofue

    from plants or waste agricultural by-product

    Some of these approaches, such as corn-t

    ethanol, are marginal in terms of econom

    and climate-change benefits (7). Other bi

    mass sources, such as switchgrasses and agr

    cultural by-products, may be economical

    more viable and have a sufficiently high pos

    tive net energy balance. To be feasible, meth

    ods must be developed for the facile catalyt

    A metal complex splits water into hydrogen an

    oxygen through an unusual series of steps.Rethinking Water SplittingRichard Eisenberg

    CHEMISTRY

    Department of Chemistry, University of Rochester,Rochester, NY 14627, USA. E-mail: [email protected]

    Published by AAAS