A recent tweet by Greg Wilson hits the nail on the head:
VM images are just PDFs you can run. If we use ‘em for reproducibility today, ppl will have to hack tools to mine their content tomorrow.
In fact, both virtual machines and PDFs were designed for a very specific purpose, but that purpose is not scientific publishing. PDFs were designed for capturing the exact contents of printed paper pages in a computer file. There are very good reasons for doing this, of course, and that’s why PDFs became popular. Virtual machines were designed to make a complete software stack, from operating system to application software, transportable from one physical computer to another. There are very good reasons for doing this as well, which is why virtual machines became popular.
Both PDFs and virtual machines were adopted for scientific publishing out of convenience: they were well-established technologies that seemed to correspond to an immediate need. PDFs replaced printed paper as the publishing medium for science in what was a minimal transition to electronic publishing: only the medium was replaced. But electronic publishing would allow much better: raw data and computational methods could be machine-readable and thus reusable by others. Virtual machines provide a low-effort way of archiving and publishing computations in a minimal way, requiring no change to the software and file formats we use today. But again a big chance is missed to do better, and make data and code accessible and reusable.
ActivePapers was specifically designed to address these issues, and it is perhaps the first file format designed with the needs of scientific publishing as its top priority. An ActivePaper combines datasets and the code that works on it, in a way that permits reusing one or the other (or both, of course) in a fully tracable way. For doing science not only reproducibly, but also openly, that’s very important.