1. Assume a magical black box which, when the "press" button is pressed, prints out the *exact* result of an election. Also assume that it is the definitive Black Box--while the result is [by definition] invariably correct, no one knows how it works. It's a hypothetical, accept everything above as a given. What do you think the odds of such a thing becoming the accepted method of national elections?
Well, that is almost a perfect description of the current Brazilian voting machine. What it prints is the exact result of the election on that voting section, by definition... of the Supreme Electoral Tribunal, who has the last word on the matter, as absurd as it may be. It was instituted in 1996 with great fanfare, before the people could be made aware of the huge risk (malicious software stealing, say 10% of the votes on every one of the 400'000 voting sections). The people is slowly getting wiser, but it may be another decade before the TSE is forced to replace it.
3. Recounts. There is little to suggest that a recount in conventional elections would produce results more accurate than the initial tallying. But it makes people feel better. Those tangible slips of paper, as ridiculous and flawed as they are, are used even when a purely digital apparatus recording choices directly from a keyboard to electronic storage (like a hugely-redundant RAID or something) would be cheaper, more convenient, and [provably] more reliable. Go figure, but that's how it is.
They are not ridiculous at all. With a purely digital system, a well-placed insider can steal votes on a national scale, without leaving any evidence. With a paper-only system, ballot boxes can be stuffed or tampered locally while no one is looking. With both, fraud becomes much, much harder: the fraudster would have to hack the software before the election, and then physically tamper with every ballot box after the election, replacing the paper ballots to exactly match the totals chosen by the software. Thus the combination of the two systems is considerably more secure than each system could be by itself.
Ideally, every voting section should count the paper ballots immediately after the election closes. If the totals don't match, there should be an investigation, but neither should be assumed to be more correct than the other. If the differnce is considerd significant, the votes in that section should be invalidated and everybody who voted there should vote again. If that happens at more than a few sections, then a global digital fraud should be suspected and the entire election should be redone.
