AI for the Good of Democracy: How AI Can Redeem Itself and Safeguard Elections


Click here to learn more about Golnar Pooya.

With the stakes (and tension) impossibly high,
November’s fateful election will now be held amidst a global pandemic of
unprecedented scope. As things stand, American voters will still be called on
to fulfill their democratic duty on schedule this year, though under the threat
of an airborne pathogen that causes serious illness and death. The results of
the election will now dictate how America navigates a completely altered global

At the same time, the coronavirus is casting a shadow over how this year’s all-important presidential election can and should be carried out. First, new voter numbers are plummeting as registration efforts simply cannot take place in person. Second, vote-by-mail alternatives — now of greater import with the looming threat of lockdowns — are being criticized for opening the door to possible stolen votes, double votes, and state-by-state disparities.

On top of all this, increased voter
polarization, fueled by fake news, and
attempted foreign political interference during the pandemic are likely to
result in decisions at the ballot box
that do not best reflect the actual political views of individuals.

In a world where the most seemingly simple and innocent things have suddenly become frightening, yesterday’s supposed existential threats may, ironically, come to the rescue. AI — hitherto maligned frequently as a threat to democracy — provides a remarkable tool for tackling each of these issues. The technology promises to correct voting records and boost registrations, empower voters to better follow their personal policy preferences, and bring confidence to mail ballot results. Now more than ever, it’s critical to consider the ways in which AI can, when deployed transparently, secure and bolster democracy by safeguarding the election during COVID-19.

On-The-Ground to Online

Predictions that this year’s election could see the highest voter turnout in the past century have been upended by the emergence of COVID-19. Today’s “new normal” prevents door-to-door campaigns and voter registration drives, leading to concerns that many first-time voters will be left on the sidelines. While virtual events are attempting to fill the void, recent data for March and early April suggests new registrations are plummeting in many states compared with the same period in 2016.

Moreover, the lack of grassroots registration efforts fails to update voter information. One in every eight voting records is inaccurate, with more than 12 million records out-of-date at any one time, including millions of deceased voters. Such administrative problems persist as voter officials still lack sufficient tools to remove the deceased from current lists or to do so in such a way that doesn’t mistakenly remove living voters.

This is where machine learning software’s ability to quickly spot information discrepancies is working to iron out the systemic kinks. ERIC is a sophisticated and secure data-matching tool that identifies out-of-date records by comparing voter registration data between states against motor vehicle licensing agency data and against the Social Security Administration master death index list. ERIC — Electronic Registration Information Center — is already achieving major success, having identified 26 million people who were eligible but unregistered to vote as well as 10 million registered voters with incorrect information between 2012 and 2018.

There are simply too many voters and too many
variables for analog voting registration to work efficiently. With the right
safeguards, then, ERIC promises an overdue overhaul. Importantly, member states
need to agree to stipulations that ensure non-biased use of the tool, like the
requirement to mail out notices to unregistered
voters upon identification. The benefits are wide-ranging, with
member states increasing the accuracy of their voter rolls, improving their
election performance across several measures, and reducing costs. Bringing this
system to even more states, or federally, could help to increase voter
registration in the absence of on-the-ground efforts in the run-up to November.

Polarized Base

Trying to register voters in the age of coronavirus is one thing, but encouraging and ensuring an informed choice during times of major uncertainty is another. American voters have been bombarded with fake news and attempted foreign political interference in recent years, and this has only intensified during COVID-19. As is, trends seem set to continue along a line that predicts a democracy growing less and less healthy.

For the voters who do show up on election day, their decisions are often volatile. Studies show that voters can shift and shape their decision at the ballot box based on the attractiveness of the candidate, recent news coverage, and social expectations. Ultimately, voter apathy and voter polarization lead to the same outcome: votes which do not accurately reflect the political views of individuals. This is because voters who rely on emotion or predisposition do not take into account the entire political landscape of their given vote.

Again, smart solutions are working to help voters better understand candidates and follow their personal policy preferences. Last year, the AI tool helped hundreds of thousands of EU citizens to match voters with their most suitable candidate in the European Parliament elections. To do this, the tool asked users to vote on issues as if they were politicians themselves, with the algorithm then matching them to like-minded candidates.

Machines already make all kinds of difficult
decisions for us — with recommendation engines helping to find the cheapest
flights and the best car insurance — but bringing this into the political realm
undoubtedly raises eyebrows due to the danger of bias. However, if done
correctly, AI promises to improve the caliber of individual decisions by
cutting through the noise of negative campaigning, biased reporting, and
slanted arguments. In the same way that technology improves other aspects of our lives,
voters should not be afraid to turn to AI to aid them in their political
decision-making process.

The Power
of the Pen

COVID-19 changes this election on a
fundamental level. Regular elections entail large crowds, shared surfaces, and
the potential for virus propagation. On the other hand, mail-in ballots are
being seen as a voting alternative that respects social distance. There are,
however, a number of perceived problems associated with mail-in ballots that I
mentioned earlier — the threat of stolen votes,
double votes, and state-by-state disparities. If mail is to
replace live voting, there must be smart security measures in place.

Risk-limiting audits is one potential solution. This statistical method double-checks a sample of paper ballots against
digital tallies to determine whether results have been tabulated correctly.
This security measure detects irregularities that could influence the outcome
of elections, such as cyberattacks and machine and human errors — and this is a
solution that already has government buy-in.

Last year, The Department of Homeland Security’s Cybersecurity and Infrastructure Security Agency announced it was partnering with VotingWorks to further develop and pilot the organization’s AI-powered risk-limiting audit tool, Arlo. Platforms like this will be vital in November if mail-in ballots become the democratic default. States, especially those which experience tightly contested elections, should, therefore, explore risk-limiting audits to guarantee the integrity of their outcome.

Push-Pull of AI in Elections

Without the broader voter and voting benefits that technology offers, results from this election season could be altered and/or jeopardized at a time when the U.S. is in great need of stability. At the same time, however, some critics are not so sure that AI is the electoral answer. This is the same AI that has been used the world over to pinpoint voter views and unduly influence social media feeds, and the same AI that has previously exhibited racial and gender bias.

These are valid concerns that AI solutions in
elections must confront head-on — and exactly why the previously discussed
implementations come loaded with safeguards to ensure transparency. ERIC, for
example, automatically encrypts the non-public data it collects to ensure
privacy. Risk-limiting audits, meanwhile, leave a paper trail while working to
effectively combat the outside manipulation that many are looking to prevent.

The ongoing pandemic requires extraordinary
solutions to secure democratic institutions. AI, when leveraged responsibly,
offers an unrivaled tool for the defense of our democracies. And, with the
whole world looking on, the chance to use AI to bring greater certainty to a moment
of deeply felt doubt represents a historic opportunity.

Credit: Source link