The Science of Open Source

The OpenScience Project is interesting. As its About page explains:

The OpenScience project is dedicated to writing and releasing free and Open Source scientific software. We are a group of scientists, mathematicians and engineers who want to encourage a collaborative environment in which science can be pursued by anyone who is inspired to discover something new about the natural world.

But beyond this canonical openness to all, there is another, very important reason why scientific software should be open source. With proprietary software, you simply have to take on trust that the output has been derived correctly from the inputs. But this black-box approach is really anathema to science, which is about examining and checking every assumption along the way from input to output. In some sense, proprietary scientific software is an oxymoron.

The project supports open source scientific software in two ways. It has a useful list of such programs, broken down by category (and it's striking how bioinformatics towers over them all); in addition, those behind the site also write applications themselves.

What caught my eye in particular was a posting asking an important question: "How can people make money from open source scientific software?" There have been two more postings so far, exploring various ways in which free applications can be used as the basis of a commercial offering: Sell Hardware and Sell Services. I don't know what the last one will say - it's looking at dual licensing as a way to resolve the dilemma - but the other two have not been able to offer much hope, and overall, I'm not optimistic.

The problem goes to the root of why open source works: it requires lots of users doing roughly the same thing, so that a single piece of free code can satisfy their needs and feed off their comments to get better (if you want the full half-hour argument, read Rebel Code).

That's why the most successful open source projects deliver core computing infrastructure: operating system, Web server, email server, DNS server, databases etc. The same is true on the client-side: the big winners have been Firefox, OpenOffice.org, The GIMP, Audacity etc. - each serving a very big end-user group. Niche projects do exist, but they don't have the vigour of the larger ones, and they certainly can't create an ecosystem big enough to allow companies to make money (as they do with GNU/Linux, Apache, Sendmail, MySQL etc.)

Against this background, I just can't see much hope for commercial scientific open source software. But I think there is an alternative. Because this open software is inherently better for science - thanks to its transparency - it could be argued that funding bodies should make it as much of a priority as more traditional areas.

The big benefit of this approach is that it is cumulative: once the software has been funded to a certain level by one body, there is no reason why another should't pick up the baton and pay for further development. This would allow costs to be shared, along with the code.

Of course, this approach would take a major change of mindset in certain quarters; but since open source and the other opens are already doing that elsewhere, there's no reason why they shouldn't achieve it in this domain too.