We need a technology that can be abandoned and still be readable in future times.
Any technological solution is bound to fail because maintaining it requires repeated investment in either maintaining what will become an obsolete storage format in the future, or repeatedly re-writing it as new media are invented.
It's all very well suggesting that technology from people such as "Carnegie Mellon University and IBM Research" might be worth using, but this assumes a certain amount of continuity to maintain the physical storage that requires organisations to survive. You cannot rely on government or industry to still be around in the future, and the 'Cloud' (whatever is meant by that) needs to be maintained as well.
You end up with stupid chicken-and-egg situations if the description of the programs and machines necessary to read the media is only stored on the media itself.
I respect Vint Cerf. He's very influential. But he's not, in the grand scheme of things, an engineer (his degrees are in Mathematics, and he's managed various teams and companies mainly on data communication). Nowadays, he's good at the grand scheme thinking, not the detail.
He was being interviewed on Radio 4 this morning, and I got the feeling that he was either dumbing down what he was saying for a non-technical audience, or that he did not fully understand various fundamentals on machine architecture and what would be necessary to maintain in order to run a program from a current generation of machines. I would hope that it was the former, but I was not convinced. When taking about the systems, he talked about taking a snapshot of the software "with a description of the machine it runs on", glossing over that the description would have to be incredibly detailed to capture all the nuances of machine architecture to allow a working machine to be reconstructed from that description.
I would suspect strongly that it would already be nigh on impossible already to reconstruct systems from people like DG, Prime or Tandem (amongst others) unless working physical instances exist.
Trying to capture all of the operating characteristics of a complex modern processor like Power 8 or a Haswell and the associated support chipsets to allow it to be reimplemented in the future on architectures unimaginable at the moment would be a herculean task!
Much better would be to ban the use of all proprietary closed file formats, and keep the definition of the open file formats in enough detail to reconstruct the data stored in those formats.
But this does not alter the fact that there needs to be readable media maintained in perpetuity.