The Future of Engineering Election Systems
It's been a year of elections around the world. How do engineering and design choices affect the candidates running and results generated in our elections? And why do we have far more channels to rate our Uber drivers than we do our politicians?
Host Guru Madhavan chairs a political summit with:
Charles E. Phelps, provost emeritus of the University of Rochester and expert in health economics.
Eswaran Subrahmanian, research professor at Carnegie Mellon University and expert in decision support systems.
Episode Transcript
GURU MADHAVAN
Union Station, Washington DC. I hopped on a rideshare for a meeting near the White House. Travel time, eight minutes. After the ride my phone lit up again. The app was eager for my feedback on everything from the driver's conversation and coolness factor, their playlist, to parking ability, and cleanliness and safety. All this for an eight minute jaunt. Now compare that to voting for political leaders. One quick ride, endless feedback, years of impact, just a single checkbox. But what if we could rate politicians like we do rideshare drivers? Could this help boost turnout and lead to better policy outcomes? Even public participation? What if we made voting itself more engaging, expressive, and enjoyable? Make it feel less like a chore and more like having a real say. I am Guru Madhavan and in this episode of Create the Future, we will visit the what's and what if’s of voting. What can engineering do to support, even enhance our decisions? In conversation with me are two scholars of decisions making … Charles Phelps from the University of Rochester, New York …
CHARLES PHELPS
The person with the most first place votes is often a very polarising, charismatic figure. And the person with the fewest first place votes is often a person who is second best choice for everybody, a compromise candidate. And there's a very simple example of how the voting rule can completely change not only who wins, but who's likely to be running in the race.
GURU MADHAVAN
And Eswaran Subrahmanian of Carnegie Mellon University in Pennsylvania …
ESWARAN SUBRAHMANIAN
In the case of elections, if you have a system which doesn't bring the best candidate out, but brings a mediocre candidate, then you will be in trouble. It is not only the information, how you consume the information in what order, and what kind of feedback do you have in order to correct the problems you have.
GURU MADHAVAN
With them, we'll launch into the worlds of psychology, politics, and philosophy. We'll look at how we decide on choices, how we should decide as groups on matters of public policy, and how we could decide if we borrowed ideas and insights from engineering systems design. Chuck, please take us to 18th century Paris, and tell us what is social choice.
CHARLES PHELPS
This begins in some sense when Louis XIV officially formed the French Academy of Sciences, 20 January 1699. And as they evolved through the years, they grew and their members died and they had votes to replace them in very clumsy voting methods. So along the way, a couple of members of the Academy tried to figure out ways to efficiently choose new members. First of them was Marquis de Condorcet, who was a mathematician and an economist. He proposed a rule which is astonishingly interesting and acceptable to people. And that is if you let people vote and somebody's vote beats everybody else in the election, they win. And that's called Condorcet voting. It starts out with a rank order list and then you just simply use it that particular way to decide who wins. So anybody that beats everyone else wins the election. Unfortunately, that doesn't always produce a winner. And that's one of the many problems that Condorcet voting has. So a long later came Jean-Charles de Borda and he designed a voting system that also used a rank ordering, but did it in a very different way. So let's say there's five people on the ballot. And if you're ranked first, you get five points. If you're ranked second, you get four, and so on down the line. And then you simply add up the points for everybody. And the one with the highest point total wins. Now that had actually been invented by an early European monk to choose new priests in the monastery. The Borda count is actually fairly widely used in various places. And it was adopted by the French Academy of Sciences to choose new members until it was tossed out in 1801 by Napoleon III, who had his own method of voting. So that began this long pathway towards trying to find a voting method. And that evolved into a series of things. The one we're most familiar with is majority rule. There are literally dozens of ways that have been devised to use rank order. But these all come down to the question of what we really mean by social choice. And that is how do you combine the opinions of a bunch of people into a single choice? It's really quite clear now as you study these voting rules that choosing a voting rule sometimes chooses the winner.
GURU MADHAVAN
Sub, you are a design engineer. You study how people can make better decisions. So tell us more about how design, which is the liberal art of bringing together different voices that Chuck was alluding to, help us here.
ESWARAN SUBRAHMANIAN
In the engineering design literature, there has been a lot of debate over the application of social choice theories, including Condorcet and Borda's count and other ways of identifying what is the choice you want to make among alternatives. But engineering is a little bit different in the sense that you're often identifying a product and a collection of products and their attributes. So to find out which product is better, you take a datum. You take a particular choice and then say how the other choices vary with respect to each one of those attributes. And sometimes it's plus/minus notation that is used. And then you count the number of pluses and minuses in each one of them and then see which one of the alternatives can be ranked with respect to the datum. But in this case, if you change the datum, then the choices also change with respect to that because of the attributes you use. In some ways in engineering, there has been debate saying that we should be using either mainly Borda’s count as a way to deal with choices. But unfortunately, in social choice theory, you don't start with a datum. In engineering, datum is very important because you are trying to choose an alternative which is better than the existing set of alternatives in order to identify the attributes. It's in that sense that engineering differs from the traditional social choice theory. Now sometimes what you may want to do is to change the datum and still see which are the alternatives that come about in terms of what you want to choose. Because the ranking of the alternatives with respect to the datum changes depending upon the datum you choose. So that is the part where actually you're determining the requirements of the product. That's where you are identifying the variation from the datum so that you can position your product with respect to competitors.
GURU MADHAVAN
You have written that the term ‘users’ creates a kind of distancing, which means it creates a space between the technology and the people who actually engage with it. Is it fair to assume that when you're in front of the ballot box, we are a bit more than a user of a technology, but actually designers of a decision?
ESWARAN SUBRAHMANIAN
It's very tricky here because you are a designer of a decision, but at the same time you don't have choices in creating alternatives. And in the case of engineering, I can create a new alternative with new requirements based on existing alternatives and a datum point. And it's there that there is a difference. You can think of the user as a person who's using the voting system to choose a particular requirement. But the voting system has been designed already. Having the right voting system is useful, but not sufficient by itself. So the user is not designing the voting system. The user is participating in the voting system. So it is in that sense that I separate the user from the decision maker. On the other hand, you also see that the individual is choosing an alternative from the given set of alternatives. But when you are designing, that is not the case. So you can see in different circumstances, the user is behaving in different ways. Now, if the user is included in the requirements generation, then it's a different story. They are participating in the requirements themselves, thereby participating in deciding the bounds of the design. So it is in this sense that engineering uses a lot more knowledge embedded in the individual in order to arrive at the requirements.
GURU MADHAVAN
This has been called the year of elections. And you talked about the majority rule earlier. And by and large, every one of those elections in global politics is about who gets the majority and wins. So take us through the panoply of voting methods that are at our disposal.
CHARLES PHELPS
Well, let me start with majority rules. It sounds extremely simple. The person with the majority of the votes wins, but sometimes when you have more than two people on the ballot or more than two choices, sometimes you don't have a winner of the majority rule. And the US voting system, for example, resolves that problem by saying it's not really majority rule that we're running in our presidential elections, but it's a plurality rule. That is to say, the person with the greatest number of votes wins. It's sometimes called first past the post using a horse racing analogy. And I'll come back to a problem about that that's really very visible in US elections in a moment. But the other way to do it instead of having a plurality is to say, OK, we're going to have a run-off. And if we have five people in the election, we're going to toss out the three with the fewest number of votes and then have a second binary choice between the first two, and then we'll have a winner. Now, that's a run-off election. Both of these alternatives, if the majority rule fails, they have the problem that the winner can depend on who's in the race. And let me give you an example from the US presidential elections in 2000. The two major candidates were Republican George Bush, the son of the earlier president, and Al Gore, who later won a Nobel Peace Prize. And a third fringe candidate, Ralph Nader, the political activist who wrote the book Unsafe at Any Speed about the automotive industry. And you come down to, for example, the Florida vote. In the straightforward election, Bush very, very slightly edged out Gore in the plurality count. But about 4% of the votes have been given to Ralph Nader, the alternative candidate, and exit polls showed almost invariably had Nader not been in the race, those votes would have either stayed home or they would have gone to Al Gore. And I think quite readily, almost everybody understands that if Nader had not been in the election, Gore would have won Florida and hence would have won and become the US president. There the choice of the voting system determined the US president. Florida said it was a plurality vote and Bush became the winner. Had they had a run-off, then Nader would have been tossed out. Almost with 100% certainty, Gore would have won a run-off election against Bush. All these elections have problems like this. This one’s formerly called the independence of irrelevant alternatives. An irrelevant alternative is one that doesn't matter when you come to the final choice, in the simplest sense. Whoever's in the race can change the outcome. That's an endemic problem in voting rules. There are some ways to get around that. Let me just move on from majority rule to the ones using rank order lists like Condorcet and Borda use. There are literally, it turns out, several dozen ways of using a rank order list to choose a winner. Let me just give you a simple example. There's two basic processes in these, one of them is you start out with a rank order list and then you toss out candidates. One of the criteria says, let's toss out the candidate with the fewest first place votes. And another way to use that is to say, let's toss out the candidate with the most last place votes. Then you re-vote, you re-vote until you're down to two people. Then you have a winner. Now let's just come back to that first choice. Are you getting rid of the candidate with the fewest first place votes or the most last place votes? Those are very different people. The person with the most first place votes is often a very polarising, charismatic figure. And the person with the fewest first place votes is often the person who is second best choice for everybody, a compromise candidate. And so one of these rules rewards being polarising and the other one rewards being a compromiser or a negotiator. And there's a very simple example of how the voting rule can completely change not only who wins, but who's likely to be running in the race.
GURU MADHAVAN
Now let's get to the psychological aspects of it. We need to make voting more enjoyable, expressive and make it enthusiastic for future use. But ultimately, it is about extracting richer, more nuanced data from the electorate. Correct?
CHARLES PHELPS
Let me go back to majority voting. That's the voting system that has the smallest possible amount of information conveyed back to the people organising the vote. It's a binary piece of information. You can't get any smaller in information theory. The rank order system obviously provides more information about that. But just coming into the question of use, rank ordering actually gets really hard to do if you have more than about five or six candidates. If you're trying to rank order 20 candidates, what you tell people is, give me one at the top and one at the bottom, and then work your way into the centre. And then the people in the centre are just all bunched together, is the way that actually works out. Rank order system is very hard to use in that circumstance. If you impose that on people's voting system, it's going to drive people away from participating. So one of the things you'd like for a voting system to do is be easy to use so that people come out and do it. The second thing is that you have to make sure that people can understand how the voting rule works. There are two things you can do with a vote. One is you can choose a winner or winners, and the other is you can gather information. Those two sometimes are in conflict, but often they're self-reinforcing. And there are ways that people are quite familiar with that can do both. Let me give you an example of that, almost ubiquitous now on Amazon, when you go through the airport, every time you check out of a grocery store, they ask you to give you one or two, three, four or five stars on your experience. That's called star voting. Kind of a simplified form of another voting rule that basically has everybody grade every choice. Now everybody is familiar with grades because they've all been graded in school. So you can take an electorate and ask them to, instead of ranking all of the choices, you could ask them to put a grade on them. That turns out to be really easy to do, first of all. Now then the question is how do you use the grades or how do you use the stars? The stars are kind of undefined grades and what this average star voting does is they take an average. Look at the Amazon.com score. It takes the average number of stars that people have assigned. When you get star voting, there's an incentive for people to vote strategically. You don't have sincere voting. Now a way to avoid that is to not use the average, but use the median vote. And this is a voting system called majority judgement. And in majority judgement, everybody grades the candidate and then you take the middle most grade. So majority voting and grading combined give you this majority judgement, which is actually a wonderful system of voting in many ways. It's easy to use. People understand what grading means, and people understand at least what averages and maybe the middle most means. Now let me give you an example of how this might have worked with the famous Brexit vote. The vote given to them was to leave or stay. It's a binary vote. Again, it's the minimal possible information you convey. Now, if the Parliament and the leadership of the United Kingdom really wanted to understand what voters wanted they would have given them a bunch of choices. Let me give you an example. One of the choices might be stay, but the other might be leave. But in between, you could have things like stay unless we can't control our own border, or stay unless we can't control our own currency, or stay unless we can't control both of those. And you could put a bunch of things on the ballot, and you could, in fact, have preliminary tests to see which were the most important to people. And then you could have had the electorate grade those A, B, C, D, E, F, just the way every United Kingdom citizen would have understood. And it would have vastly better informed the parliament and its negotiators with the European Union about how to proceed. It would have been a completely different world that we live in now if they just changed the voting system.
ESWARAN SUBRAHMANIAN
In the case of elections, if you have a system which doesn't bring the best candidate out, but brings a mediocre candidate, then you will be in trouble. It is not only the information, how you consume the information, in what order, and what kind of feedback do you have in order to correct the problems you have.
GURU MADHAVAN
What are you optimistic about the possibilities at the junction of voting, collective choice, engineering design and public policy?
CHARLES PHELPS
I guess I'm both optimistic and pessimistic in that in the United States, we're beginning to see some experimentation in alternative voting rules. In many ways, the rapid proliferation of star voting, which I think is a kind of a simplified majority judgement that I really like, is becoming so commonplace in people's mindsets that I think it may be possible to have these kinds of choice mechanisms, voting methods moved into the political domain, probably not in my lifetime, but I hope in yours, Guru. If we can achieve that, I think it will be important. It will improve public policy because you'll have much better information fed back to the body politic. I'm a little concerned about the proliferation of information that comes with modern technology as well, with the flood of information on the internet it's almost impossible for the average user to actually understand whether it's valid or not. And so they rapidly begin to associate providers of information with accuracy. What’s been developed, I think, in the sort of polarisation of the American electorate is that people turn over and over again to what are being called echo chambers for their information sources, the same news broadcast, the same bloggers, the same influencers and so on. And I have to say that makes me very fairly pessimistic. I don't know how to change voting and information conveyance so that we learn how to use compromise for mutual benefit, rather than a winner-take-all kinds of strategies that seem to have emerged out of the system we have.
ESWARAN SUBRAHMANIAN
I'm also pessimistic and optimistic. I'm optimistic in the sense that more and more designers of technology are starting to include the target customers as part of their design process. I'm also pessimistic because if you take digital technology, there is this notion that the user doesn't know what he or she wants and we are going to provide them, and we end up with the kind of attention economy we are facing today in the digital world. The digital world is not subject to the same kind of liabilities that the physical artefacts used to have, and this is a major public policy problem we have to face. Recently I just saw a paper which was very interesting. It says, ‘from attention economy to cognitive lock-in’. Especially with the new large language models and so on, what is happening is everybody is using the same large language model. Now slowly everybody's language in the way they write is going to be identical and slowly what is going to happen is what they are consuming, like Chuck pointed out, different groups will consume different pieces and they'll get locked into that particular cognitive perspective. In fact, dividing the society further rather than integrating the society and not enhancing community participation, both locally and globally. So you've got this strange problem where digital technology allows you to scale, while destroying the local community which will bring things together. And in that sense, I think engineering design, public policy, and the kind of voting Chuck is talking about to bring the right kind of issues into the public space becomes extremely important.
GURU MADHAVAN
Consider the math. A traditional election with six candidates offers the voter only six possible expressions. But grading those candidates, A through F, you get 46,000 ways to voice your opinion. Add plus or minus modifiers, and you're looking at 34 million potential expressions. Could more expressive, nuanced voting enable better debates and discourse? Even better campaigns and how we fund them? If democracy and engineering are instruments for choice, shouldn't we have more say in choosing our leaders than about our drivers' vibe? Thanks for listening to Create the Future, a podcast from the Queen Elizabeth Prize for Engineering and Peanut & Crumb. Look out for new episodes every two weeks and follow QEPrize on social channels, Instagram, LinkedIn, Facebook, and X. The show was produced by Tess Davidson and Thomas Scott. And until next time, this is your host, Guru Madhavan.