Or, why I am grateful for the opportunity to explore serendipity.
Grant agencies that fund scientific research play the long game. They expect a return on investment in the form of value created by researchers for the society at large well beyond the committed funds, at some nebulous point in the future. In exploring the consequences of my own graduate school years, I reflect on how shockingly cheap it can be to profoundly change a particular status quo.
In the year 2000, my prospective PhD advisor Rafael Romero supported me in my application for a PhD fellowship from the Spanish government. It was granted. The fellowship, consisting of about 850 euros per month for 4 years, made me financially independent at age 22. In Barcelona this monthly stipend was enough to keep me fed and housed, with some small change to spare. But I didn't need much more than that, given that I was a member of a lab in the well-equipped University of Barcelona. Being in the lab became my life.
I would like to think that, whoever established the fellowship program at the Spanish Ministry of Education and Science was aware of what they had set in motion . I most certainly didn't. In short, I was free to engage in research full time, aided by an advisor that encouraged exploration. The project wasn't flashy, or timely, and I am not sure that I had thought it through. What's more, in retrospect it is clear that I didn't know what I was doing. Yet it was funded.
That decision, by unknown bureaucrats sorting through evaluations of fellowship proposals and applicant CVs, did not only change my life, but that of many around me. Likely that includes you, and your own computer, which perhaps unbeknownst to you it is running software that I wrote. For free.
Because during my graduate studies, I had time to learn freely, well outside my academic field, and I learned to program. Then, I started applying my programming skills to fulfill the image processing needs that arose in my developmental biology studies. Then I started creating tools that didn't exist, and I shared them freely. These included a number of plugins for ImageJ.
One such ImageJ plugin, developed during a visit to the Hartenstein lab at UCLA, was the "A 3D Editing" plugin for drawing spline-based contours or tubes to model three-dimensional anatomy.
Crucially, the plugin documentation page had a link that read: "hire me". It's still there today, in a page that I can no longer edit. That link, and that plugin, are responsible for what happened next: I got an email from Rodney J. Douglas, from the Institute of Neuroinformatics in Zurich. Professor Douglas wanted to discuss about the possibility of writing a bigger version of that tiny plugin, to support the then nascent field of large-scale connectomics. Was I flattered.
Fast forward a few months, Rodney and I started the TrakEM2 software: a plugin that grew bigger than ImageJ itself. So big, in fact, that its multiple library dependencies made it hard to deploy, and the necessity to interact with its data structures prompted me to facilitate dynamic scripting for real-time interaction with them. The first led to the creation of Fiji, thanks to the insight and great skill of Johannes Schindelin who put my ImageJ folder under version control with git, in 2007, and became Fiji version zero. The second was the creation of the scripting interpreter in ImageJ, which later led to the Script Editor partly via a Google Summer of Code. So Fiji was born, and with it, a strong community that delivered to the world astounding tools like OpenSPIM by the Pavel Tomancak's lab, multi-view registration and fusion, and the Big Data Viewer. 
Would you have predicted that funding a mostly clueless PhD student to research planarian embryonic development would have led to the creation of software that enabled thousands of labs worldwide to perform complex image processing and analysis? Neither would I. The Fiji paper accumulated thousands of citations in 4 years, and only a fraction of researchers using Fiji actually report that they do so in their papers. The Fiji software clearly filled a void, and spared an entire generation of biologists from the staleness and lack of extensibility of commercial tools, and their extraordinarily expensive software licenses.
What's more: would a top-down, dedicated effort have ever organically nucleated a community of researchers that would have delivered, and continue to deliver and support, software that empowers thousands of laboratories world wide? Probably not, and given that failures are seldom reported, it is hard to learn about what could have been.
What I find extraordinary, with the benefit of hindsight, is how inexpensive the whole operation was. I do not know how many PhD fellowships were granted in the year 2000, but the impact of just this one--mine, that I happen to know of--has been enormous. Even if none of the other awardees contributed anything, which is not the case, it would have been worthwhile.
In summary: fund broadly, fund randomly, and also fund selectively if you must, after the other two. Examples abound indicating that it is better to fund many small grants than a few large ones: creativity is best sampled broadly and nurtured frugally. As the story of the green fluorescent protein GFP shows , predicting returns on investment in research is impossible. As every biologist knows, diversity is key to generating and exploiting opportunity. Foster it.
To conclude by expressing my heartfelt gratefulness to the Spanish tax payers and to those who had the vision to implement the PhD fellowship program during the 1990s. May you bring it back in full force.
P.S.: "Fund people not projects", as J. P. A. Ioannidis eloquently proposed.
- Sadly, the following year (2001) the number of awarded PhD fellowships went down a lot, and nowadays the FPU program ("Formación de Profesorado Universitario") barely exists if at all. Even the Spanish Ministry of Education and Science does not exist anymore!
- I take pride in having taught some extraordinary computer scientists like Stephan Saalfeld and Stephan Preibisch what an ImagePlus is. Then in turn they taught me 100x more.
- GFP was discovered in the 1960s from studies in jellyfish (far removed from what the NIH would fund nowadays), but its applicability to developmental biology was not realized until 1992, and for reporting neural activity in conjunction with calmodulin (GCaMP) not until 2004. Modern neuroscience would be impossible without GFP and GCaMP or their conceptual equivalents.