Imagine a theoretical paper. The author developed some theory and wrote the equation down on the paper and published it. Up to this point, there is no difference from the activity of novel writers (or whatever). But one of the strength of science is that once it is created (or discovered), the scientific result becomes self-standing (and sometimes looks trivial). The point is that anyone can get the same answer as far as one follows the correct (logical or mathematical) path, and anyone can test the correctness. Reproducibility is the heart of science.
Consider numerical studies. Someone showed some simulation results by his/her own developed program. As we know, numerical simulations are sometimes very sensitive and small changes of implementations causes unexpected difference. How can we feel the same reproducibility we can feel on theoretical studies on the numerical stuff? Open source activities fit the needs for the scientific researches (at least) in terms of numerical methods. We should note that this is not only on the checking processes on the results, but also for the next step of the paper. If we can share the working code, the progress would be much faster in any research field. That's what science are evolving in the past (mainly on theoretical and experimental fields).
This is what I am thinking and why I am releasing my codes. This is why I took the GNU General Public License (see Chapter A), not other licenses.