Welcome back to securing digital democracy. In today's lecture, we are going to focus again on some of the human elements of, of secure and trustworthy voting. Because voters are people, voting systems have to be designed so that they're, they're easy for people to use. I mean, this sounds obvious, but if voting systems aren't easy to use, lots of things can go wrong and some of those things have really important security implications. Before we begin, I'd like to just remind you to please try to keep up with the reading in Broken Ballots, and, we have the chapters to read for this week posted online. The first element of human factors that I want to talk about today, is usability. Usability, in computing systems has been studied very extensively and, and in voting systems too. Usability consists of several different things. It consists of how, how quickly can, people, use a particular system. How satisfied are they with their interactions with that system. But also, and the, the focus of our discussion today will be this last element. Usability includes how correctly and accurately, can people interact, with a particular system. The correctness and accuracy and usability of voting systems, ends up having a big impact on how accurately votes can be counted. So, to see why. Let's think again about our definition of the integrity security property for voting. We defined integrity with two parts. First, votes have to be cast as intended and second, they have to be counted as they are cast. This cast as intended issue turns out to be largely a usability problem. Does the system provide an interface for voters to use that allows the voter to tell the system correctly who it was they wanted to vote for. Now, we've seen several examples already where systems just couldn't do that. One example was the punch card voting system used in Florida during the 2000 presidential election. The, famous butterfly ballot. Because of the design of the ballot, thousands of voters apparently voted for a candidate they didn't intend to. And we can see a little bit of why this design is bad right here. But for a more accurate view, let's think about how the system would actually look to a voter. It's not just going to be, presented flat before their face. But it's, sitting on a table angled just slightly towards the voter. So, if you're looking at the system at an angle, the usability problems are, are even easier to spot. Look at how poorly aligned the punchcard locations are with the candidates there. So, this kind of design where the ballot just creates obstacles to being used correctly can lead to major problems as it did in Florida. Some other problems in this design. You can see that the choices alternate between the left and right faces of the page. This is another thing that likely added to and compounded the voter confusion. We've seen examples of this kind of usability problem with DRE electronic voting machines as well. I'm going to show you this example. This comes from Sarasota, Florida in the 2006 presidential election. The Sarasota ballots had a multiple page design as in the Diebold DREs we saw in previous lectures. And voters would vote by touching square-shaped targets on the screen for each page. So, I'm going to flip through a couple of pages of this ballot now. And I want you to pay close attention, and then I'm gonna ask you a question about them. Okay, let's see if you were paying attention. Did you notice the race for the House of Representatives? It's right there at the top of the second page. It turned out that a lot of voters didn't notice that race. And, the reason we think that, is there was a one percent rate of under votes or people who didn't mark, the, the permitted number of candidates or just left it blank for the governor race that's located on the bottom half of this page. But for the race for Congress for US Representative, there was an under vote rate fourteen%. Fourteen percent of voters left that very important race blank. Now, why was that? Well, well one possibility was that the machines were dishonest. But, based on testing, that, people in the usability field have done, subsequent to this race, that kind of fraud doesn't seem like the most likely explanation. Instead, the reason that we think, fourteen percent of voters appeared to miss this race, has to do with the design of the ballot. So, if you look at the way these two pages of the ballot were laid out. There, we have the first page that had only one race. And then the second page that had two races on it. That's the first usability problem here. That the voter after the first page was only expecting there to be one race. But there's an even more, complicated and interesting problem going on with this design and this in an effect called banner blindness. So, look how, prominent this heading is in the middle of the ballot. The heading that says state. This is the first thing your eye is drawn to. Well, there's an effect that's been very widely documented in usability studies, called banner blindness. This says that people tend to subconsciously ignore text, and especially rectangular shapes that are above the main heading on a page. If you're on a website and it has banner shaped advertisements above the text, you usually ignore those. And this is just reinforcing and training you to do this more. Because of the, the banner blindness effect is so strong. We've been able to reproduce in studies after this election with actual users, the, the same effect in, in, volunteer voters. They'll just not see something up there. So, let us now in the forms if you, also missed that race when you looked quickly at the pages. Another problem, with these machines that just compounded this sort of thing was very slow response. The machines, displayed a lot of lag moving from page to page. Furthermore, there's often a problem with DREs that the displays aren't completely accurately calibrated. This manifests itself in other cases and has in other races has been documenting as causing, what people think is, is a vote swapping effect. You touch one place on the screen, and, the, machine registers, and shows a check in another box. That's usually caused by a miscalibrated display. Unlike say your, your IPad, or your, your, your smartphone, the displays on voting machines tend to use a much older technology that requires, periodic manual calibration. And often, this step is just skipped in the, the, the election procedures, or it's not done correctly and that can lead to, to this miscalibrated display. But that just slows the voter down further, adds aggravation, and adds confusion to the process creating further usability problems. Finally notice that this ballot is 21 pages long. The user would have to go through this again and again and again. The likelihood of mistakes is just going to be compounded as the user is being fatigued by all of this. So, all these different factors combine to make, make for significant usability problems in this style of DRE. Let's look at another kind of usability problem with DRE's though. Remember, when we talked about the voting process for DRE's that the last screen voters are presented with is typically a review screen that looks like this that highlights all of the choices on the race. So, based on some of the things you've learned about usability and some of what you know about, about human nature. What might limit the usefulness and effectiveness of this kind of review screen in actual voting. Let me give you a chance to think about that and then, we'll come back and talk about it. So, while you were gone we took the liberty of setting up this, DRE voting machine that I'm going to use to demonstrate, some of the problems we'll talk about. But before we get to that, I want to look again at the, review screen you saw in the slides. So, this kind of review screen was introduced to help voters, catch any mistakes or errors they made on the ballot. But how well does that actually work in practice? Some researchers decided to, to test this empirically by, creating a mock voting system and getting some volunteers to come in and try it out. But the researchers did, did one thing with the system which is they, they rigged it so that the review screen would, display either additional candidates, or have some, choices missing, or just display the wrong selection. Something different from what the voter really picked. So, when they took the actual voters and, and asked them. 95 percent of those voters said they checked the review screen. But when they actually saw whether they caught errors that were introduced by the researchers, 63 percent of the people being studied failed to detect the changes. That means own, only about a third of people caught the errors the researchers introduced. Now, this is, this is quite a, a shocking result of, of, to me, anyway. We would have thought that the, the review screen would provide a, a much better way to, to catch errors. But it turns out that very few voters, actually, take the time, and have the, have the, concentration to, to catch errors, on such a screen. Now, one question this raises is how many voters would check the tape on, would check a print out on a VVPAT device. So, we've taken the, the, Diebold AccuVote TSX voting machine that, I showed you all the other week, and, one thing that we've done is added this device. So, this wasn't here before. This is a VVPAT add-on that Diebold made to allow jurisdictions to have a printed paper trail to go along with the election. It just attaches to the side of the machine. The way this works is it prints out while the voter is at the review screen. It prints out, on a cash register style tape, a record of each of the voter's selections. Now, the design of this VVPAT unit adds several other usability questions. It creates, it actually creates an impediment to voters checking the tape. Because in order to even see it, you have to open this door on the VVPAT unit. And there's no button or switch or anything. There's nothing to, to force you to do that before you cast your ballot. This is entirely voluntary for the voters. So, a large fraction of voters aren't even going to bother to open the door. Even if you do actually seeing the choices printed on the tape might be pretty hard, because they're printed in very small and crude writing on this cash register style, piece of paper. Diebold has provided a little magnifying glass. In fact, the, the writing is so small you need this to, to make it out. Unless you have excellent vision. But even then the design of this magnifier is such that unless you're looking at it head on, you're not going to be able to see the printing. So, there are a lot of physical obstacles to usability here. But there's another kind of problem with this, this VVPAT, design, and it has to do with the review screen issues. And this actually a pretty subtle but pretty scary attack. So, the VVPAT idea is that if we have a paper trail, and a, a, a physical piece of paper that the computer and the voting machine can't go back and change. Then that's going to provide a, a pretty strong defense against software cheating because, people can go back and audit the results on the tape and make sure they match the count on the computer. But what if you wanted to commit fraud on a device that had a VVPAT? How, how could you go about doing that? Well, one thing you might do, is you tried to have it print out some choices that weren't what the voters selected. And hope that the voter wouldn't catch it. If very few voters actually look at the tape, that, that's going to give you the chance to, to commit, some level of fraud without getting caught. But what if some voters do look at it? I mean, that still provides a defense. Right? It's like if, if those voters are if, if one percent of voters actually look at it and that's a random selection of voters, then it's like you're getting a one percent audit on the correspondence between what's on the screen and what's on the tape. Well, the problem is that the voting machine, still has an option to cheat. So, what it can do is it can make sure that what it's showing on the review screen matches what's on the tape. That way if there is, it, it has to make sure that those match. Otherwise, if someone, if someone notices a discrepancy. If a voter checks and notices, they can just get a poll worker to come over and see that they're not the same. And, that's going to be an obvious way to catch, that the software is cheating. So, that the, if the machine wants to cheat, it has to show the same fraudulent choice, something the voter didn't intend to vote for, on the review screen, as it prints on the tape. Now, it might just decide to do this for, for a certain fraction of voters. But that would. If the poll workers have good procedures in place, that would be something they could chat, they could catch, because they've noticed, a, a very high rate of, of, of voters having to go back and change choices. So, what the machine might do if the designers of the vote stealing software were clever is, is take into account the usability features, and usability failings of the voting population. It could try, for instance, to cheat only in cases where it thinks the voter's not going to check the review screen. And we see that voters, we, we might see from, from research that voters who are really in a hurry, for instance, fail to check the screens. So, if you fly through everything, that might increase the chances that the, that the machine wants to try this kind of attack and change something on the screen. So the, the, the theory here is that it's going to decide whether or not to cheat based on whether it thinks you're gonna look. Another thing it could try to do is, only cheat, if it thinks no one's going to believe the voter. So, if you were having a lot of trouble. If you had to press several times to get a button to go on. That might be a sign that you're, a voter who, is not familiar with the touch screen technology or maybe you're, you're, an older voter who's having trouble. Something like that, then even if you do, notice that the machine is showing something that wasn't your choice on the review screen, a poll worker is going to be likely to assume that it's, it's just a human error. That the user made a mistake. So, all these sorts of problems, add up to, give us a, a bit less, reason for confidence in the VVPAT mechanisms, and their ability to, to reliably stop clever software on the voting machine, from trying to commit fraud.