Computational Engineering, Empirical Computer Science, Distributed Systems

Repetition of Research (Unintended)

| Comments

Most of us are familiar with the classical narrative of ongoing scientific progress: that each new discovery builds upon previous ones, creating an ever upward-rising edifice of human knowledge shaped something like an inverted pyramid. There’s also an idea that, in the semi-distant past of a few hundred years ago one person could know all the scientific knowledge of his (or her) time, but today the vast and ever-expanding amount of information available means that scientists must be much more specialized, and that as time passes they will become ever more specialized.

There is some truth to these ideas, but there are problems as well: when new knowledge is created, how is it added to the edifice? How do we make sure that future scholars will know about it and properly reference it in their own works? If a scientist must be incredibly specialized to advance knowledge, then what does he (or she) do when just starting out? How does one choose a field of research? And what happens when the funding for research into that area dries up? Contrary to what we learned in grade school, a scientist cannot choose to simply study, say, some obscure species of Peruvian moth and spend the next 40 years of summers in South America learning everything there is to know about it without also spending some time justifying that decision to colleagues and funding bodies.

New scientific knowledge is published, generally in refereed conference proceedings and/or journals, but sometimes also online in the form of research reports or blog posts. The good thing about refereed publications is that they have been read by the author’s (or authors’) peers, who have rendered a judgment that the work is of sufficient quality and novelty to merit adding it to the official scientific literature. The bad thing is that this process is slow and prone to errors–referees are human beings, and they work on a voluntary basis, and they may only be partial experts on the topic of the article in question. The fact of the matter is that it is extremely likely that a lot of science is repeated, in fact that it makes sense that a lot of work gets repeated since it actually takes less work to re-research some answers in the lab (or, for mathematics, on the chalkboard) than it does to keep track of every single research output produced by every researcher, everywhere, for all time.

Let that sink in for a minute: There are costs associated with storing, indexing, and searching vasts amounts of data. For some problems, for some researchers, at some time, those costs may actually exceed the cost of simply re-discovering the findings anew. An additional wrinkle is that the people making the decision to embark on a new research project by definition do not know how difficult it will be to properly research the problem, and are strongly motivated NOT to discover that their proposed solution has already been discovered by someone else (think: graduate students looking for interesting thesis projects, or untenured junior faculty hoping to pad out their research portfolio, or anyone submitting a grant proposal), so long as the omission is not so completely obvious as to be embarrassing.

Consideration of the second point provides some more startling insights: yes, researchers do become highly specialized in their knowledge, but this specialization is NOT defined by an objective partitioning of the universe of knowledge into discrete, manageable, non-overlapping subsets, but rather by what journals or conferences they follow, who their colleagues are, and other factors like their personal algorithms for sorting and selecting papers from the constant flood of new information. The primary limitation on scientists is how many papers they are physically and mentally capable of reading and assimilating within a given timespan, and this value is likely a constant that varies only somewhat from individual to individual. Further limiting the potential for specialization is the fact that every paper ranges over a wide variety of semi-related topics, in order to increase its chance of being deemed sufficiently “interesting” and “novel” to warrant publication by a randomly selected collection of “peers”. It is not only possible, but likely, that there are many different groups of researchers who study similar problems, but never know about each other because they follow different conferences, and use differing terminology. Further, they will use different basic assumptions when designing and evaluating experiments, having different standards for what constitutes “good” or “bad” solutions to related problems. For an example of this, I’d like to highlight the almost complete disconnect between groups that study scheduling algorithms for distributed systems, which take a very “mathematical” and proof-oriented approach, falling back on studies of heuristics for very abstract problem formulations when there are no reasonable guaranteed algorithms, and groups that study “autonomic computing”, which take a much more biologically inspired approach to what is essentially the same problem, starting from a more common-sense (but less mathematically tractable) approach to defining success or failure, relying first on heuristics, but often falling back on biologically inspired approaches like ant-colony optimization or genetic algorithms.

So, a lot of research gets repeated, and for some people in some situations, that might not be an entirely bad thing. Still, the situation seems something less than optimal. Hence, there have been movements later toward not only greater sharing of research data, both before, and after publication, but also better sharing of data through the use of standardized, extensible formats like XML or JSON, and semantic markup of metadata. By using global- and discipline-standardized ontologies, or grammars that define relationships between syntactic primitives, it is possible to develop general purpose repositories that nonetheless allow for complicated queries and discipline-specific reasoning about results. By using these standards and technologies and applying well-though-out policies for research data management, it should be possible to lower the bar for indexing and searching the existing body of scientific knowledge (which to those on the inside looks much less like an edifice and more like a maelstrom).

I plan on covering both the development of data management policies and why they are a good idea, and enabling technologies for the storage and semantic markup of data in future posts.