I'd like to call your attention to this week's electile dysfunction news, which is about a mini-Minnesota situation in Fairfax County, Virginia. I think it's instructive because it illustrates how some problems with "paperless" voting are actually quite similar to a more old-fashioned form of voting, "paper only" voting, and a mooted new-fangled kind of voting, Internet voting.
The basic situation is reported somewhat blandly by the Washington Post in "Ballot Machine Malfunctions, Fairfax Race Left in Limbo" but the juicy details are covered in Jeremy Epstein's blog. My summary is this: in a close local race, a voting machine malfunctioned so that its vote totals were incorrect, and the votes in question were enough to tip the race either way. In the end, the winner was chosen more or less like a coin toss.
So what it is the basic problem here? There was only one copy of the vote data. With the touch screen electronic ballot devices used in Fairfax County, there is only the electronic representation of the ballot -- essentially a single copy, because although you can make digital copy of a digital ballot, the correctness of the copy is suspect if the computer involved is suspect. And in this case, the computer clearly malfunctioned, so we're not really helped by the fact that it contained two or more identical instances of flawed vote data. So when there is effectively just one copy, and that is incorrect, there is essentially no way to re-check the vote totals to determine who really won.
But before you pile on with the many vocal detractors of DREs (so called "paperless" balloting devices), please also consider that the same problem applies to a more old-fashioned form of voting, where there are only paper ballots (or PBHC voting, paper ballot, hand count). The situation could occur like this: a precinct's ballot box turns out to contain fewer (or more) ballots than the number of people who voted; hence we have no confidence that the ballot box is legitimate, containing all and only the paper ballots that were cast in the precinct that day. There is no recourse because these are the only ballots we have. Ouch again. The same applies to lever machines in a way even more analogous to DREs. There are no paper ballots, but only the vote totals recorded on the odometers on the back of the lever machines. If the odometers aren't consistent with the number of ballots cast, there is no way to re-count.
The common theme here is that there is only one instance of each ballot (in the case of PBHC) or vote (in the case of DREs and lever machines). That's why we're seeing increasing adoption of the hybrid scheme of paper ballots that are electronically tabulated in the precinct, and also saved in ballot boxes. In that case, we have 3 data points: the voter totals from the poll books, the paper ballots, and the electronic ballots recorded by the tabulator. If any one is inconsistent with the other two, we have some recourse if the other two agree with each other. That's a big improvement.
Lastly, there is a cautionary note here for proponents of Internet voting schemes in which people cast their votes on a computer that is similar to a paperless DRE (and may in fact be a citizen's home PC) except that the data is transmitted over the Internet for storage on a central system for vote tabulation. I wrote earlier that aside from the many objections to Internet voting, there is one simple one. Here it is: just like in Fairfax County, there is only one electronic copy of the vote data, and if that data becomes suspect, then there is no way to recover the original ballots cast by the citizens. But here the problem is magnified. Instead of one device malfunctioning in one precinct, and casting into doubt some hundreds of citizens' votes, an Internet voting scheme could cast into doubt all votes stored on a malfunctioning central system.
In that case "Ouch!" would not even begin to describe the chaos.