Precise individual microtargeting threatens to remake the political landscape as thoroughly as it has remade marketing. This paper explores the observed uses to date of political microtargeting as well as the many difficulties, some inherent to politics, of measuring its effects. Considering the philosophical difficulties of predictively removing human choice, it then assesses the observed risks of and some potential remedies to the current trajectory and finds that free electoral choice is not doomed to be written out of the system.
More than two centuries ago, the French philosopher Pierre-Simon de Laplace described a hypothetical “intellect which at a certain moment would know all forces that set nature in motion, and all positions of all items of which nature is composed.” This intellect, Laplace argued, could “embrace in a single formula the movements of the greatest bodies of the universe and those of the tiniest atom; for such an intellect nothing would be uncertain and the future just like the past would be present before its eyes” (Laplace 1951, 4). Assuming the predictable and deterministic nature of the universe, he concluded that free will could not exist.
That conclusion echoes loudly in the modern era, where political society is grappling with the possibility that “liberal habits such as democratic elections will become obsolete, because Google will be able to represent even my own political opinions better than I can” (Harari 2016, 343). Will widespread political use of predictive algorithms inhibit free will as would Laplace’s demon? This paper will trace the development of microtargeting, explore the basic purpose for which these algorithms are employed, and demonstrate that while more of them are around us than we may realize, the future need not look so bleak, as proper policy can preserve much of what we value in our democracy.
Microtargeting is a series of relatively simple and intuitive practices that may essentially be summed up as “knowing your audience.” Since the 1920s, the concept has been of central importance in the field of marketing: “Micro-Targeting is a marketing strategy that uses consumer data and demographics to create audience subsets/segments. It’s possible to predict the buying behavior of these like-minded individuals, and to influence that behavior through hyper-targeted advertising” (MNI, n.d.; Sheingate 2018, 8-11). When substituting for a purchase the choice to vote a particular way, political applications of the practice become readily apparent. The first systematic use of political microtargeting was Richard Nixon’s “Southern Strategy” in the 1968 Presidential campaign, where instead of creating one message for a unified nation, he imagined the nation as segmented into North and South and tailored the civil rights language of his advertisements in a way he thought would be appealing to the particular audience (Bunting 2015). Since then, targeting has become precise enough that by the 2000 Presidential election, both major political parties could maintain databases of voter history going down to the individual level.
The strategic benefit of accurate microtargeting is hard to overstate. Campaigns are tightly budgeted operations, and any marginal improvement in turnout can spell the difference between victory and defeat. Crucial to a successful campaign is distinguishing between the “base” of hardcore supporters that will cast their vote come rain or shine and the “persuadables” who, though much fewer in number, can swing the outcome of the election and thus require special attention (Burton et. al. 2015). However, reported voter history from primary elections generally does not provide enough detail to be useful. Additional data is required if the model of voter preference is to make predictions with minimal error. Relevant data can be of nearly any sort, including all manner of physical, psychological, and preferential variables. A campaign with access to seemingly random and innocuous data will not only know who to target, but how to target them with particular arguments.
Surprisingly, frequent implementation of these techniques has taken a long time, only truly coming into widespread practice during the 2012 Presidential election (Issenberg 2012). While the Obama campaign focused heavily on microtargeting, no presidential campaign has used it to greater effect than Donald Trump’s in 2016. Brad Parscale, the Trump campaign chair, recognized early on the advantage of digital ad campaigns, and invested some $90 million of the campaign’s $250 million war chest into online publicity, mostly through Facebook (Martinez 2018). The Clinton campaign, meanwhile, had a much larger budget but chose to spend it on traditional TV and radio ads, casting a wider net at the expense of precise microtargeting (Beckel 2016). The margin of Trump’s victory was slight enough that Hillary Clinton could well be President had her campaign better appreciated the power of the Internet.
This discussion is naturally incomplete without explaining how websites like Facebook collect user data. While using the site, and the many other sites that include Facebook-linked security or advertising mechanisms, exhaustive amounts of data on the content of the user’s browsing will be stored on Facebook servers (Martinez 2018). Google retains an approximate minimum of 5 gigabytes of publicly-accessible data on each user (Curran 2018). When customers pay one of these companies to send a message to a niche market, they describe the traits of the group they wish to appeal to, and the companies’ predictive algorithms locate others in the database with those traits.
However, most political advertisement requests do not involve one-to-one matching of traits, meaning these predictive algorithms face the tremendous task of drawing relevant predictions from the many potentially irrelevant data categories that the parent company tracks. Some algorithms follow a process known as unsupervised learning (MathWorks n.d.). In this situation, the algorithm analyzes a vast set of data and attempts to determine relationships among the data. A common method of doing so is clustering, the grouping together of apparently highly correlated variables. Much like the robots in Amazon’s warehouses whose decisions on where to store items have proved indecipherable to humans, these relationships and the grounds for their efficacy may not be understood by the campaign operatives who analyze them, potentially leading to disastrous electoral strategies (Knight 2015).
There also exists supervised learning, which takes relationships known to exist and creates a model that may predict their future behavior given new data. This method holds particular promise as it could theoretically distinguish between, say, equally politically active Trump supporters and Never-Trump Republicans despite their voting for and donating to the same national party. From there, clustering could be used to identify the factors that imply the likelihood of developing one political affiliation as one ages, or as national conditions change. One could thus theoretically be exposed to advertisements tailored to that general cohort from a young age, guaranteeing by exclusion of alternate messaging that person’s fulfillment of the stereotype.
Because of all this potential, the consequences of this technology should be discussed. The first is something seemingly endogenous to algorithms: the recursive spiral they can fall into when interacting with each other too frequently. Wall Street has already observed the possible catastrophes that can result from algorithms interacting to direct each other into territory neither was designed to operate in (Scharre 2018, 199-201, 203-204). Politicos using algorithms to aid their campaigns should be very careful of which data they train their models on, as one algorithm learning from the work of another may produce distorted outcomes.
A prudent campaign manager should also recall that political conditions are ever-changing. Parties may phase in and out of prominence in a matter of a few years as new issues take their place at the front of the national consciousness. Once-valid relationships drawn by an algorithm may therefore become outdated and wildly inaccurate between two consecutive election cycles. The same voter might also privilege different issues or identities within themselves as time goes on. To correct for this factor, a campaign would either have to regather its data each year or accept some amount of error in their strategies. Perversely, the rule-breaking elections like 2008 or 2016 which require cogent strategies are the ones most likely to inflate that error.
There is the additional and perhaps more frequently articulated concern that to turn such a large share of the current democratic process over to artificial intelligence would destroy some vital part of democracy. The questions are philosophical: can political preferences or choices be meaningful if only one narrative is available to each person? Does narrow exposure to carefully tailored messages shrink citizens’ capacity to exercise free will? Certainly, on its face, a world with ubiquitous microtargeted advertisements providing one solid, personalized drumbeat of one single narrative smacks of dystopia. Less dramatically, current usage of microtargeting has been cited as one reason for the hyper-partisan state of political discourse, and one likely to worsen over time (Illing 2017). Given these concerns, it is worth asking if they are justified.
First we should ask whether campaigns using microtargeting are more successful than otherwise. The answer seems to be yes, but it is far less clear than one might expect (Kreiss 2017). The Ted Cruz primary campaign, for instance, used microtargeting of a more extensive scope than even the Trump campaign, but still could not clinch the Republican nomination (Conick 2016). One should note, though, that Cruz still received the second-greatest number of delegate votes after Trump. The most interesting finding, though, is that microtargeting seems to do an awful job at convincing “persuadables,” but an excellent job at ensuring base turnout (Kreiss 2017). So far, it has remained easier to design advertisements to reinforce partisan loyalties than to change them. With that, it seems that polarization, if not outright identitarianism, is the nation’s political future (Sheingate 2018).
However, comprehensive studies of microtargeting’s effects are barely extant at the national level in America, let alone the state level. This likely stems from the prohibitive cost; estimates for the cost of access to a marginally useful voter file begin at $140,000 (McDonald et. al. 2015). So long as congressional campaigns struggle to end up in the black, the number that can afford microtargeting will remain small and insufficient to constitute a valid sample. Rigorous studies may thus have to wait for this cost to decrease.
Given this fact, policymakers may think it wise to implement protections against microtargeting now, before its most advanced forms become routine. The natural suggestion to increase privacy protections may not actually be enough; studies in Germany have demonstrated that privacy laws much stronger than those in the United States are ineffective because enough information is willingly shared on Facebook and other social networks to construct perfectly valid predictive models (Papakyriakopoulos et. al. 2018). The United States could have an opportunity to tighten its own standards, though, since the case Rosenbach v. Six Flags Entertainment Corp.,dealing directly with the question of biometric data gathering, is likely to be cited as persuasive authority or even be considered by the Supreme Court. Six Flags alleges that the collection and storage of biometric data from customers entering its parks does not constitute “harm” under a strict state privacy law. Setting a firm standard on this malleable word could heavily curtail biometric data gathering for politics nationwide. Furthermore, the United States’ two-party system may provide surprising protection from a 1984-esque terror in which all citizens’ information is rigidly controlled, as neither party would want to let the other get away with maintaining an advantage in microtargeting technology for long. Thus, at the very least, there will always remain a choice between two very alluring and psychologically tailored options.
This paper has considered the realities, possibilities, risks, and possible remedies to a political climate steeped in microtargeting. Tailored and counterfactual political advertisement is no rarity in the present. In spite of this, free will remains in the modern system, and it will take more than microtargeting to do away with it. Perhaps this question remains open, but it may be interesting to note that Laplace’s demon was actually disproven as a constraint to free choice (Wolpert 2008). With this in mind, it is prudent to view the growth of microtargeting as a matter of logistics: difficult, debatable, and important, yes, but not necessarily an apocalyptic sea change.
Beckel, Michael. 2016. “Team Clinton Sponsored 75 Percent of TV Ads in 2016 Presidential Race.” The Center for Public Integrity. November 8, 2016. https://publicintegrity.org/federal-politics/team-clinton-sponsored-75-percent-of-tv-ads-in-2016-presidential-race/
Bunting, Luke. 2015. “The Evolution of American Microtargeting: An Examination of Modern
Political Messaging.” Butler Journal of Undergraduate Research. 1, no.2. https://digitalcommons.butler.edu/cgi/viewcontent.cgi?article=1007&context=bjur
Burton, Michael John, William J. Miller, and Daniel M. Shea. 2015. Campaign Craft: the Strategies, Tactics, and Art of Political Campaign Management. Santa Barbara, CA: Praeger.
Conick, Hal. 2016. “How Social Media, Microtargeting and Big Data Revolutionized Political Marketing.” American Marketing Association. October 1, 2016. https://www.ama.org/publications/MarketingNews/Pages/social-media-big-data-microtargeting-revolutionized-political-marketing.aspx
Curran, Dylan. 2018. “Are You Ready? Here is All the Data Facebook and Google Have on You.” The Guardian. March 30, 2018. https://www.theguardian.com/commentisfree/2018/mar/28/all-the-data-facebook-google-has-on-you-privacy
Harari, Yuval N. 2016. Homo Deus: A History of Tomorrow. Signal. Hersh, Eitan D. 2015. Hacking the Electorate: How Campaigns Perceive Voters. Cambridge: Cambridge University Press.
Illing, Sean. 2017. “A Political Scientist Explains How Big Data is Transforming Politics.” Vox. March 16, 2017. https://www.vox.com/conversations/2017/3/16/14935336/big-data-politics-donald-trump-2016-elections-polarization
Issenberg, Sasha. 2012. “How Obama’s Team Used Big Data to Rally Voters.” MIT Technology Review. December 19, 2012. https://www.technologyreview.com/s/509026/how-obamas-team-used-big-data-to-rally-voters/
Knight, Will. 2015. “Inside Amazon’s Warehouse, Human-Robot Symbiosis.” MIT Technology Review. July 7, 2015. https://www.technologyreview.com/s/538601/inside-amazons-warehouse-human-robot-symbiosis/
Kreiss, Daniel. 2017. “Micro-Targeting, the Quantified Persuasion.” Internet Policy Review 6(4). https://policyreview.info/articles/analysis/micro-targeting-quantified-persuasion
Laplace, Pierre-Simon. 1951. A Philosophical Essay on Probabilities. Translated by F.W. Truscott and F.L. Emory. New York: Dover.
Martinez, Antonio Garcia. 2018. “How Trump Conquered Facebook—Without Russian Ads.” Wired. February 23, 2018. https://www.wired.com/story/how-trump-conquered-facebookwithout-russian-ads/
MathWorks. n.d. “What is Machine Learning? 3 Things You Need to Know.”
McDonald, Michael, Peter Licari, and Lia Merivaki. 2015. “The Big Cost of Using Big Data in Elections.” The Washington Post. October 18, 2015. https://www.washingtonpost.com/opinions/the-big-cost-of-using-big-data-in-elections/2015/10/18/cb7bdf6c-7443-11e5-8248-98e0f5a2e830_story.html?utm_term=.3163b55d5f98
MNI Targeted Media. n.d. “What if Micro-Targeting?” https://www.mni.com/blog/advertmarket/what-is-micro-targeting/article_a12cbac0-137c-11e9-84d4-73b6685fa60c.html
Papakyriakopoulos, Orestis, Simon Hegelich, Morteza Shahrezaye, and Juan Carlos Medina
Serrano. 2018. “Social Media and Microtargeting: Political Data Processing and the Consequences for Germany.” Big Data and Society 5(2): 1-15. https://journals.sagepub.com/doi/pdf/10.1177/2053951718811844
Johnson, Steven. 2008. The Ghost Map: A Street, an Epidemic and Two Men Who Battled to Save Victorian London. New York: Penguin.
Scharre, Paul. 2018. Army of None: Autonomous Weapons and the Future of War. New York: W.W. Norton.
Sheingate, Adam. 2018. Building a Business of Politics: The Rise of Political Consulting and the Transformation of… American Democracy. Oxford: Oxford University Press.
Wolpert, David H. 2008. “Physical Limits of Inference.” Physica D: Nonlinear Phenomena 237(9):1257–1281. https://ti.arc.nasa.gov/m/pub-archive/1476h/1476%20(Wolpert).pdf