Open Research License

Proposal by Victoria Stodden.

Source

Article: ENABLING REPRODUCIBLE RESEARCH: OPEN LICENSING FOR SCIENTIFIC INNOVATION (draft version). By Victoria Stodden.

URL = http://www.stanford.edu/~vcs/papers/Licensing08292008.pdf

Text

THE RATIONALE FOR THE OPEN RESEARCH LICENSE: THE ALIGNMENT OF INCENTIVES

"Open standards and open access are insufficient to promote the free discovery and development of science since the success of a scientist is measured by citations and the amount of subsequent work he or she engenders. This reward system can create short-sighted incentives to both move quickly to working on the next scientific publication and not release the full research compendium in the belief that other scientists will “steal” work by building upon it without attribution. I suggest an attribution license is required that will perpetuate virally through all derivative works, thereby ensuring attribution for all parts of the research compendium.

Secondly, scientists need to have a guide to make the release of their complete research product as easy and as useful to others as possible. An appropriate license will do this both by making it possible for researchers to release everything under one umbrella license and publicizing the concept of doing so. A tailored license would bring the discussion beyond mere open source to Richard Stallman’s concept of free software and free research.

Thirdly, the current copyright system closes scientific research in such a way that is counter to the scientific ethos of reproducibility.

A COMPILATION LICENSE IS REQUIRED

Copyright law in the U.S. does not permit the copyright of “raw facts” but original products derived from those facts can be and are, in fact, assigned automatically whenever a creative work is produced. In this automatic assignment, comes the prevention of copying and using the work in another creative or scientific endeavor. In the case of scientific research a tension is created since the scientific ethos is to reproduce previous results and build on them to generate further scientific understanding. The default copyright can be limited if the authors take steps to limit those rights by using an alternative license for their work such as the GNU General Public License (“Copyleft”) or the Creative Commons license.

A. Selection and Arrangement of Data

In Feist Publications, Inc. v. Rural Telephone Service, the Court found that white pages telephone directories are not copyrightable; copyrightable works must have creative originality:

. . . the copyright in a factual compilation is thin.

Notwithstanding a valid copyright, a subsequent compiler remains free to use the facts contained in another’s publication to aid in preparing a competing work, so long as the competing work does not feature the same selection and arrangement.

Currently the Court holds databases protectable. A license that applies to the “selection and arrangement” of a database, in a virally attributive way, can encourage scientists to release the datasets they have compiled by providing a legal framework for copyrightability. This permits the application of a license to foster reproducible research.

Most computational research work takes place in a university setting and many universities claim some ownership rights over the research product. In a November 1, 2007 discussion with Katharine Ku, Director of the Office of Technology Licensing (OTL) at Stanford University, the concern was not in copyright and focused on primarily on patents. The OTL did not perceive any conflict between the Open Research License I am proposing and their interests as a university.

THE OPEN RESEARCH LICENSE

The Open Research License is a compilation of existing licenses: the Creative Commons BY attaches license to the media components of the compendium, the BSD license to code components, and if the scientist chooses to release his or her data to the public domain, attaching the Science Commons Database Protocol to the data.

The CC BY license is designed for media: to “share your creations with others and use music, movies, images, and text online that’s been marked with a Creative Commons license.” If used alone, it is misapplied to the academic research compendium since it does not adequately cover code and, in fact, using the CC BY license for code is actively discouraged by Creative Commons. The BSD license evolved from the development of Berkeley Unix code and is a standard license for open code. Using the BSD license alone for scientific compendia leaves the documentation, figures, final paper and other forms of scholarship, the experimental design, GUIs interfacing with the algorithms, pseudocode, and dataset build methodologies for example, without an adequate license. But all of these works could be released appropriately under the CC BY license that ensures consistent viral attribution for the entire compendia.

This selection of licenses allows for viral attribution and, by avoiding the Share Alike aspect common to many licenses, ensures each scientist is attributed for only the work he or she has created. If Share Alike were not excluded from this license, each the entire derivative work (or new scientific discovery) would carry the ORL license, including the upstream work’s attribution. In order to promote scientific research, it is sensible to allow the downstream researcher the choice of whether he or she would like to attach the ORL to his or her work (although the ORL remains attached to any upstream work he or she may have incorporated). Specifically, there must be no bar to building upon previous scientific research. A corollary benefit to the ORL’s relaxation of the Share Alike component is that it becomes easier for startups to employ the research as part of their technology without having all their (possibly) proprietary work come under the ORL.

As a simple umbrella license the ORL is easier to use than the alternative. Without the ORL, each time he or she releases scholarship, the scientist would have to fashion together a combination of licenses from an entire spectrum of choices. Since the ORL uses common existing licenses, there are no compatibility or interoperability issues with existing licensing schemes.

POTENTIAL PROBLEMS IN DELINEATING COMPONENTS OF THE COMPENDIUM

Making a distinction as to which components of the research compendium belong under which license might be blurry: for example algorithm descriptions and pseudocode are frequently included in computational research. Arguably, each could be considered either code (there is no requirement that code must be functional to be covered by the BSD License for example, just that it be “source”) or media (pseudocode is also text that traditionally could be covered under a CC license). Finally, there is no adequate licensing structure that intentionally applies to the structures that house the data used. The data itself is not copyrightable but often a phenomenal amount of work goes into preparation of the dataset for research and there is no reason why this should not be attributed to the scientist and explained openly to future researchers who would like to use these data. Precisely how the data were generated or gathered, any processing done to the data to clean or verify it, and the current layout of the data are all vital pieces of information for a scientist to reproduce or understand the final result. These aspects could be emphasized as important and captured by the ORL. This dovetails neatly with the aspirations of Claerbout’s really Reproducible Research.

THE COSTS AND BENEFITS OF THE OPEN SCIENCE RESEARCH LICENSE

The NSF goal that publicly funded research be publicly available achieves important objectives: accountability and oversight in the use of government funds; promotion of scientific knowledge through both 1) direct conveyance and 2) facilitation of the opportunity to verify and improve answers to scientific questions; and the “sunshine principle” (knowledge of future public release creates incentives for better work). A license that can protect and promote these goals by aligning the scientific researcher’s interests by providing for attribution, could not only forward our scientific knowledge but dramatically improve participation by scientists in collaborative research, encourage citizen-scientists to actively engage in research, and institutionalize the web as the mode for release of scientific discovery.

Attribution of work is a cornerstone of scientific discovery and currently a tension exists for scientists between the public release of research, thereby risking loss of attribution, and limited but attributed journal publication. This can be resolved by releasing scientific research under an appropriately tailored license. The ORL would encourage academic researchers to release their work completely, permitting verification of the current findings, facilitating further scientific results in the particular area of research, and preserving attribution for research work. Such a license would also have the corollary effect of producing better science: a researcher who anticipates release of all his or her work to the public is apt to do a much more careful job.

The ORL will provide a mechanism for scientists to license the meta-knowledge associated with the creation and perfecting of their data. Prior to the ORL, this would not fall under any license. The ORL will also provide metadata that can be used to associate the entire research product license status in a machine-readable way as a single product, which would be inherently more difficult if different components were under different licenses.

The ORL holds the promise of encouraging better tools for research dissemination and investigation. The license will provide cultural pressure that encourages reproducible research, and perhaps encourages journals to education, research and access to information” in updating international copyright norms to respond to challenges arising from advances in information and communications technologies, including global digital networks.1” WIPO Copyright Treaty,

When the entire research compendium is released to the public, this can obviate the ability of the researchers to covertly begin a commercial venture based on the research results. This concern is contrary to scientific principles and the funding mandate of the NSF in the sense that science is a public good - work licensed under the ORL can be commercially used, it just cannot be built upon secretly. As one researcher has pointed out, an advantage to open code and clarity of experimental method is publicity of the new work.

Another concern is the inherent confidentiality of some data. Some data, for example personal medical records, sensitive national security data, or proprietary industry data should not be publicly released. This can be counteracted by sanitizing the data as much as possible so that any personal or sensitive information is not released. In fact the National Academy of Sciences advocates the release of as much data as possible, even if there is a risk terrorist organizations may use it to damage United States interests.

Their evaluation is that the value of the scientific output outweighs the risk of information falling into dangerous hands. The NAS also would like to promote international scientific cooperation and is concerned undue restrictions on data would hamper this process. It may also be the case that some data may require built-in security and integrity checks that must be kept confidential for the experiment to operate. This creates the corollary concern that not all the data methodology can be released. This may not be a true cost of this license since it is clear such data would not have been released in any event. The ORL may encourage innovative ways to allow some reproducibility, such as providing an online system for other researchers to choose algorithm parameters or specific sections of data and simply be returned processed results.

Algorithms may rely on proprietary libraries. Hopefully these libraries will be brought under the rubric of the ORL and opened to the wider research community. If not, the ORL may discourage the use of potentially fruitful proprietary libraries.38 Use of the ORL may involve a rethinking of university copyright and patenting policies. There may be conflicting third party obligations or conflicts with previously patented work used in the current research.

The ORL may encourage a change in the valuation of scientific work away from pure research results toward algorithm modification for useful purposes. For example, industrial applications may become a vital part of research on the web and non-researchers may be able to use the scientific research more readily than under traditional publication methods. Opening scientific research to the public has the benefit of providing the opportunity will exist for anyone with a web connection to get involved, even releasing their own derivative works under the ORL. This throws open the peer review process to anyone so motivated.

Since the ORL facilitates the communication of research and ensures attribution, it avoids two of the stumbling blocks to very large scale collaboration. The internet naturally suggests such collaboration and the ORL, by making entire research product coherently and consistently available and ensuring attribution, encourages this use of the internet’s potential. The ORL may facilitate internet-based data sharing research models. Such a machine readable license will enable researchers to search for ORL licensed work more easily.

A researching scientist may have done more experimentation than is practical for a traditional research paper. Releasing the full research product allows for the reporting and attribution of more results and experimental configurations than would ordinarily be publishable.

As alluded to in the introduction, by ensuring open easy access to others’ research, the ORL will stand as a bulwark against plagiarism and falsification of scientific results. If even the potential exists for peers to verify all your methodologies, the incentive to cheat is greatly reduced. For exactly the same reason that attribution is an important feature of the ORL, a scientist’s reputation is his or her career and the threat of being known as scientifically dishonest is exceedingly strong.

The role of third parties will be clear and consistent under the ORL, and this may not be if scientists do not have a clear licensing structure for computational work. This is especially important as the university is a common setting for computational research, and universities nearly always peer review process in a similar fashion as the patent review process."