Bad Idea, in Voting

I’ve been in my book writing fox hole, so much so that when the storm hit Maryland and D.C. and I did not lose power, I had no idea that nearly half of my state and our neighboring ones had none.  But enough about hiding from the world (and the Internet), there are alarming stories about voting worth sharing now with elections coming up, the only time the public seems to sniffle at the issue.  Internet voting.  One might say, in your dreams, pal, never going to happen.  But in truth it is happening, with calls for more.  Nineteen states offer some form of online voting, mostly for soldiers living overseas.  The Military and Overseas Voter Empowerment Act requires states in most cases to get ballots to military and overseas voters well in advance of regularly scheduled federal elections, which has led states to adopt voting via e-mail and online for soldiers.  (Other states like Maryland allow voters to download ballots online and mail them).  Because these experiments have “worked,” more calls for voting online have been forthcoming on the grounds that people might then actually vote.  It’s my understanding from voting activists that election boards are agitating for online voting, and it is a very bad idea.  To state the utterly obvious, all things online are insecure — the infiltration of Pentagon and countless companies, including financial ones, should instill fear about the sophistication of bad actors looking to steal state secrets, trade secrets, credit card numbers, SSNs, you name it.  And online elections–what a target (think about all of the people who would bother–in a word, lots).  Stuffing ballot boxes in a handful of precincts is quaint as compared to the possibilities of malware, distributed denial of service attacks, and the like in a state and federal election.  It is mind blowing, really.

Scott Wolchok, Eric Wustrow, Dawn Isabel, and J. Alex Halderman of the University of Michigan recently released a study on the ease with which they hacked a pilot project on Internet voting run by Washington D.C.  The authors explain that within 48 hours of the system going live, they gained near-complete control of the election server, successfully changed every vote and revealed almost every secret ballot. Two business days later, election officials detected the intrusion, and probably only because the authors deliberately left a prominent clue.  Some respond to these sorts of concerns with “we bank online and it is safe, so we can vote online, if we just work hard enough at it.”  As the authors explain, banking and voting involve very different activities with very different needs for secrecy as between client/voter and bank/voting precinct.  As the authors explain:

While Internet-based financial applications, such as online banking, share some of the threats faced by Internet voting, there is a fundamental difference in ability to deal with compromises after they have occurred. In the case of online banking, transaction records, statements, and multiple logs allow customers to detect specific fraudulent transactions and in many cases allow the bank to reverse them. Internet voting systems cannot keep such fine-grained transaction logs without violating ballot secrecy for voters. Even with these protections in place, banks suffer a significant amount of online fraud but write it off as part of the cost of doing business; fraudulent election results cannot be so easily excused.

The National Institute of Standards and Technology agrees.  Chief among NIST’s concerns are malware and our lack of an infrastructure for secure electronic voter authentication.  Amazingly, countries like Estonia and Switzerland have adopted Internet voting for national elections.

You may also like...

6 Responses

  1. A.J. Sutter says:

    You’re right, of course. Nonetheless, the analogy between banking and voting sadly rings true in other ways, these days.

  2. Paul Horwitz says:

    I appreciate the concerns and have no particular stake in this debate. But I wonder two things: 1) Isn’t this, like any other choice of policy instrument, simply subject to cost-benefit analysis, in this case balancing the risk of catastrophic harm against the smaller but potentially more certain risks of the current system? 2) What is the actual experience in, eg, Estonia and Switzerland? Have there been costs and harms? Have there been benefits?

  3. Danielle Citron says:

    Great to hear from you both, and A.J., look forward to hearing about your book project. For Paul’s question, the question of cost-benefits is worse than what I portrayed. The federal government post HAVA spent millions upon millions for the states to buy DREs, which have their serious flaws but certainly pose less security risks than Internet voting. So the notion of throwing those out (and states like NY just sunk their fiscal teeth into them as Maryland did just a few years ago buying all new ones) for a less secure option shows that the cost of switching is higher for little benefits (people may vote more but then again their intended votes will be more likely to be switched, flipped, or not counted due to malware, DDOS attacks, etc.) So seriously why would we think about it? Because election officials love shiny new ideas and are not technologists. There are far more odious reasons too.

    All from me, back to the book (and also an essay with David Gray)!

  4. Ken Rhodes says:

    In re Paul’s question/suggestion, we would have to establish the “cost” of a stolen election, the “cost” of compromising the secrecy of the ballots, and the “cost” of a denial-of-service attack that caused many people who expected to be able to use the system, and thus did not make alternative arrangements, to lose their opportunity to vote.

    We don’t generally do cost/benefit assessments on the rights guaranteed us in the Constitution. The “right” to elect our government would seem to me to be subject to a similar consideration.

  5. Paul Horwitz says:

    With all due respect to Ken, I think cost/benefit assessments are a regular, if constrained, part of constitutional rights adjudication–both in constitutional systems that expressly contemplate proportionality analysis and in those, like ours, in which it is not explicitly set out in the constitutional text but happens just the same. And just as we should consider the costs Ken sets out–which are among the very costs I had in mind–we should also consider any potential benefits, including increased voting rates (if, that is, one thinks higher voting rates are a good thing), reductions in lower-level fraud and error, and so on. Again, I’m not for or against it, and I’m fine with Burkean conservatism on this point, and on any other issues on which Danielle wants to be a Burkean conservative! Just asking the questions. I’m still especially interested in the experience of those nations that have actually gone this way, but that’s by way of curiosity, not advocacy of a particular result.

  6. The question isn’t as simple as cost-benefit, because we can’t say “if you spend $X more you reap Y benefit”. (Defining Y, though not easy, is probably feasible, if you try to use something like the percentage of votes that are cast accurately, plus allowances for who will cast a vote.) The problem is that we don’t — and, I think, can’t — have a good handle on the trade-off for elections. Note that the tradeoff exists even without computer issues; better-trained poll workers could make a big difference, as can things like avoiding butterfly ballots. But elections are already costly, and counties are strapped for money. Beyond that, we don’t have a good handle on the extrema; the worst case scenario with computerized voting systems is far worse than with manual systems.

    Computerization makes it worse for two reasons. First, there’s the security issue: can someone hack the voting system? There have been plenty of lab studies, but to my knowledge at most one fraud case tied to DRE (Direct Recording Electronic) voting machines. However, if you care about the subject see the many reports that were part of California’s “Top to Bottom” review; http://www.sos.ca.gov/voting-systems/oversight/ttbr/red-overview.pdf is a good starting point.

    Personally, I worry more about bugs. Code is hard to get right; voting machines are very hard, because of the voter privacy requirement: it means that one can’t keep adequate log files. There have been many reports of buggy code; see, for example, Ed Felten’s hard evidence of results that just can’t happen. These are far from the only ones; one notable one occurred in North Carolina because election officials ran far more votes through a machine than it was designed for — somehow (and the programmer’s logic escapes me; it was much harder to do it this way), not enough digits (actually, binary bits) was allocated to hold the total, so many votes were lost. (My own comments are in my blog; see especially the link to the article on Bernalillo County, New Mexico, in 2000 — a far more interesting, though less reported, story than Florida.)

    I think the best way to understand the problem is to realize that virtually no computer scientists think it’s a good ida. Normally, people push their own field’s products; here, you see the opposite.

    Could a really reliable electronic voting machine be developed? I’m extremely skeptical. I can say that doing so would be extremely expensive; we know that from the cost of developing other ultra-high-reliability systems (aircraft flight control computers, phone switches, etc.). We also know (see Avi Rubin’s paper on Diebold software) that production systems are not developed with that level of care. You get at most what you paid for, and I’m not sure one can pay enough here to get a system where the worst-case result is acceptable.