- You enter into the voting booth with your ballot and cross off your preferred candidate or party. No trust needed.
- You fold it together and put it into a ballot box. No trust needed.
- The ballot box is sealed and driven to a counting station. This is done under supervision of all stakeholders, meaning that cheating is extremely hard because your political adversary is present.
- The ballot boxes are unsealed and the votes are counted. Note that the votes are counted by all stake holders (typically one or more people from each party) making it hard to cheat.
- The final vote is passed on.
None of the above steps require trust in any one person or entity, and the probability of cheating (if the procedure is done correctly) is quite low. If there is some anomaly the votes are saved so they can be counted again.
And again if "probability is low" is the bar, then we can surely keep _exploring_ Internet voting systems without engendering rejection as academicly 'impossible' for the whole concept.
It's not a hard problem, it's an _impossible_ problem.
If there's any security hole between the BIOS, CPU, firmware, OS, app, a state-sponsored attacker will get it.
The US got into air-gapped systems. You think China (or Russia, Iran, ISIS, Europe, Anonymous, you name it) won't break in?
And what do you do if you discover that there was external vote manipulation? Hold re-elections?
And how easy is it for someone to delete, say, a couple thousand votes in a strategic state? It's hard to physically do it without evidence and a conspiracy, but if a hacker (or a simple bug) gets in, all bets are off.
That's why the paper-trail is so important