Regardless of your feelings about the outcome of the election, it’s pretty clear the present voting system is so antiquated it struggles to provide a meaningful mechanism for translating the will of the people into action. For example, I was surprised that in this age of cloud storage and wireless connectivity, I couldn’t vote at the polling place of my choice and still be presented with the relevant ballot issues from my home district. US elections are controlled at the local level, which is one reason why the widespread voter fraud sometimes alleged to exist is, in practice, almost impossible to achieve. But this same local control makes it difficult to implement the kinds of systems that would help voters better understand issues. Even ballot design is typically controlled locally, and is limited to whatever local officials draw up and their own printers can print.
Clearly there are ways modern technology artificial intelligence could improve these processes, and one group is hoping smarter AI could help voters make candidate decisions that better reflect their own goals and priorities. Researchers at Harvard and Carnegie Mellon University have been hard at work devising better methods for collective decision-making, using cutting-edge developments in artificial intelligence and machine learning. The researchers behind the effort, led by computer scientist Prof. Ariel Procaccia, stumbled upon the idea while working on decision making for software agents. They devised RoboVote.org, a free public tool that helps people optimize their group decision making process.The “aha” moment came when he realized the same toolset that could be used to help AI make better decisions could be leveraged to help groups of people make better decisions as well.
Their system comes in two flavors. One helps people assess objectively determinable questions, such as which company to invest in based upon projected revenue. The other is intended for questions with subjective valuations, such as which toppings a group should order on the pizza they are sharing.
While the first type of question is fairly straightforward since it does not involve people’s feelings or opinions, the other is trickier. To answer questions involving subjective valuation, the researchers behind RoboVote turned to utility theory. Utility theory presents the concept of the utility maximizing agent, for whom each decision comes with a concomitant cost and benefit. By deriving a group of individual’s unique utility functions for a given choice, you can pool the results and arrive at an equation that maximizes the collective utility of the group. The idea here is to create the greatest good for the greatest number of people, no matter how insane or untrustworthy their opinion.
To see how this might have played out in the presidential elections, imagine if instead of choosing between two candidates, each voter was asked to rank all the initial 20 or more presidential hopefuls in order of preference from top to bottom. The computer would then create a utility function based upon each person’s ranking, and compare this to the same set of functions generated for everyone else in the electorate. The resulting compound differential equation could be solved to reveal a candidate who would maximize the utility for the entire electorate.
This is conceptually similar to the practice of ranked voting, or preferential voting, which allows a voter to rank candidates rather than simply choosing a single candidate. In a hypothetical three-way election between Jack Johnson, John Jackson, and Richard Nixon, voters choose their first, second, and third choice. In the event that no candidate receives at least 50% of the popular vote, the first-place votes for the least-popular candidate are redistributed to the other two candidates based on the second-place selection of those voters.
This AI-based process is a good deal less intuitive than our current voting system. It would need to be an option that people could consult throughout an election campaign rather than simply at the polling station, and there are questions about how and if users would accept its recommendations as valid — especially if the AI determined that the best candidate for a user to vote for based on their stated goals and priorities was different than the candidate they themselves were inclined to support. But as it’s refined and developed further, it could prove a valuable tool for helping voters evaluate candidates — potentially more valuable than the systems we currently have in place.