Assuming that we can achieve exactly equal security to paper voting for a moment, what have we achieved other than adding another few layers of middlemen and cost?
Arguably you could only count the paper ballots in case of litigation, otherwise you'd just use the result of the machine. Sounds like a decent compromise.
That being said here in France polling stations manage to count the ballots in a couple of hours at most after closing, so I'm not sure if it's really worth it.
The US moved to electronic voting in part due to the Florida vote counting fiascos in 2000. Electronic voting machines prevent users from making undervotes and overvotes (picking no candidates, picking more than one candidate). The machines also prevent the hanging chad issue in 2000 where the voters intention was unclear. Electronic voting machines also enable blind voters to cast votes without assistance giving them privacy and assurance that their vote wasn't tampered with. Paper voting can also have it's own security weaknesses like ballot stuffing.
> what have we achieved other than adding another few layers of middlemen and cost?
and uncertainty. In principle the idea of hitting a button and having the non-networked machine spit out a paper trail that can be electronically counted sounds like a good compromise. But this leaves 1 or potentially 2 sources of uncertainty that are not necessary: (1) that what the machine prints actually corresponds to what you wanted and (2) that the counting machine actually counts what it says on the paper.
(1) is not such a risk if you can verify it visually before submitting. There do exist I believe some ideas in cryptography to allow you carry a record of your vote that allows you to verify it without revealing the choice, these could have interesting application here, but I don't know the pros and cons.
(2) is even a risk in the non-machine case, although it can be mitigated by having multiple independent parties do the count. But it can't easily be done by machine and have the same level of certainty.
There's a fundamental issue in voting, which is that ballot markings may not have a consistent meaning from voter to voter. Most analysis of voting methods ignore this and it's impacts, which are difficult to quantify, but it's pretty clear the effect is maximized in two cases:
(1) systems which limit rankings to a fixed number (the most extreme case for real ballot methods being two) of ranks (approval and FPTP both are two-ranks methods), and
(2) system which use numerical ranking systems that draw fiber distinctions than mere ordinal ranking (range/score voting being the main example.)
The problem is minimized in ranked-ballots methods, though there is room for debate over whether forced or unforced rankings are better in this regard.
For this reason, I would reject range voting for most public elections independent of practical difficulty (there might be exceptional cases where a consistent meaning can be attached to range ballots, but it's not the case in normal public elections.)
OTOH, ranked ballots Condorcet methods which need to compare pairwise results are probably more tractable with e-voting (or, rather, e-tallying.)
Believe me, its advocates have heard every criticism you can imagine, and the counter-argument is robust. I recommend you check out the book "Gaming the Vote".
That being said here in France polling stations manage to count the ballots in a couple of hours at most after closing, so I'm not sure if it's really worth it.