USNWR: How Google Could End Democracy
How Google Could End Democracy
Research suggests that Google has the power to manipulate elections around the world.
Let’s get something out of the way. I am not accusing Google of manipulating elections. No whistle blowers are sending me clandestine messages on my disposable cell phone. No statistical analysis that I am aware of has shown that Google is deliberately altering search rankings to favor particular candidates. No one has proven, to my knowledge, that Google is actually evil.
But research I have been conducting with Ronald Robertson has shown that with its virtual monopoly on search, Google has the power to flip the outcomes of close elections easily – and without anyone knowing. Over time, they could change the face of parliaments and congresses worldwide to suit their business needs – keeping regulators at bay, getting favorable tax deals and so on. And because their business is unregulated in most countries at this point, flipping elections in this way would be legal.
So I am not accusing anyone of anything – definitely not, not at all, not even slightly. But if I were Larry Page, Google’s CEO, I would have a crack team of my Mountain View prodigies studying and manipulating elections worldwide 24/7. Not doing so would be counter to the quest for profit.
When you think about some of the other outrageous activities the company has been fined for in recent years – hacking Apple software to track millions of Safari users, vacuuming up unprotected Wi-Fi data in 30 countries with its Street View vehicles, illegally marketing Canadian drugs to U.S. citizens through its AdWords service – it’s hard to imagine that Google executives haven’t taken at least a smidgeon of interest in elections. Dave Eggers’ new Google-bashing novel, The Circle, is about just that possibility – that Google will soon play a dominant role in government. If outsiders can imagine such things, Google employees can too.
The formula is simple: Find a tight race, identify the candidate who best suits your needs, and contribute to his or her campaign. Next, identify undecided voters and repeatedly send them customized search rankings that make the favored candidate look better than the other ones – easy as pie with Google’s massive database of email content and search histories. Over time, that will tip the voting preferences of many people – more than you think (see below).
Your candidate won’t know about your fiddling, and he or she will be beholden because of the contribution. Sweeter still, regulators will have no way of detecting your manipulation because it was directed at a small number of people who received customized rankings, as many of us do these days. Think how targeted the campaign spending was in the final days of the Obama-Romney race; the fate of the election was in the hands of a few thousand people in easily identified counties.
In close elections, it helps to know the right people, and no one is in a better position to identify them – and, it turns out, to manipulate them – than Google.
Search Engine Manipulation Effect
Our first experiments, which we described last year at the annual meeting of the Association for Psychological Science in Washington, D.C., were laboratory-based. Through newspaper ads, we recruited a diverse sample of eligible voters from the San Diego area who in most respects resembled the American voter: in age, income, educational background, and so on.
People were randomly assigned to groups in which search rankings favored either Julia Gillard or Tony Abbott, the candidates in the 2010 prime minister race in Australia; this guaranteed that our participants were “undecided” because they were unfamiliar with the candidates. Participants read brief descriptions about each one and then rated them in various ways and indicated who they would vote for. Then they were given up to 15 minutes to use our customized search engine – Kadoodle – to conduct research on the candidates. They had access to five pages of real search results that linked to real Web pages, and they could navigate through them freely, just as they would on Google.
In the Gillard group, pro-Gillard results had higher ranks; in the Abbott group, pro-Abbott results had higher ranks; and in a control group, the results were mixed. All participants had free access to exactly the same material; only the search rankings differed.
The results, consistent with those of research on consumer behavior, were clear and consistent: 15 percent or more of participants in the bias groups switched their preferences and votes toward the target candidate, and few participants detected that the rankings were biased. There was no shift in the control group. In other experiments, we found that we could mask our manipulation by mixing things up a bit – so well that not a single participant could detect the bias.
Search rankings have this powerful effect on votes for the same reason that they have one on consumer behavior: the higher the ranking, the more people believe and trust the content, mistakenly assuming that some impartial and omniscient genie has carefully evaluated each Web page and put the best ones first. (Not so.)
We recently replicated this “Search Engine Manipulation Effect,” or SEME, as we call it, with an online sample of 2,100 eligible voters throughout the U.S. One new and highly practical finding (Google, take note): Some demographic groups in the U.S. are especially vulnerable to this manipulation, especially – and I’m not making this up – divorcees, Republicans and people from Ohio.
Our previous studies all involved a past election from another country, but we just finished testing SEME in a real election in India. That’s right, we deliberately manipulated the voting preferences of more than 2,000 real voters in the largest democratic election in the history of the world, easily pushing the preferences of undecided voters by more than 12 percent in any direction we chose – double that amount in some demographic groups.
We even have a formula now that indicates, based on projected win margins, exactly which elections can be flipped with virtual certainly by search ranking manipulations. In the India study, the shift we produced in voting preferences was large enough to flip just about any election with a projected win margin under 2.9 percent. Worldwide, upwards of 25 percent of national elections are won by margins under 3 percent.
Even without deliberate manipulation by Google executives, there is another possibility here that is unprecedented. What if, without human intervention, Google’s algorithm boosted a candidate into high search positions? In other words, what if the favoritism happened “organically”? This is exactly what Google claimed was the explanation for Barack Obama’s consistently high search rankings in the months leading up to the 2008 and 2012 presidential elections.
Let’s play this out. An algorithm has boosted a candidate’s search rankings in a race with 100 million voters. Even if only 6 percent of them are undecided voters with Internet access (the actual percentage in most developed countries would be higher), our research suggests that over time biased search rankings will drive nearly a million votes to that candidate. In coming years, as more and more people get their election-related information through the Internet, the impartial algorithm will have even greater impact.
This is a kind of undue influence the founding fathers couldn’t have imagined in their wildest dreams.
If this sounds farfetched to you, think about it this way. In 2006, researchers at the National Bureau of Economic Research found that whenever Fox News entered a cable television market, Republican candidates gained votes. Fox consistently tipped the voting preferences of between 3 and 8 percent of its viewers, the researchers concluded – more than enough to have had a “decisive” impact on the close elections of 2000. Subsequent research found that when, over time, as those markets became more crowded with competing networks, the “Fox News Effect” was mitigated.
So… imagine if Fox News were the only television network in the country. With no competition, it would tip the voting preferences of millions of undecided voters on an ongoing basis.
Whenever a major source of influence is biased, with no possibility that opposing views will be heard, the impact of that influence is overwhelming. This is typically the case in dictatorships, which is why elections in such countries are farces. At the moment, the Google search engine is a source of unopposed influence, which also happens to be highly credible, opaque in its methodology, massive in scope, rapidly increasing in reach and beyond the scrutiny or control of both regulators and candidates. Other sources of influence during elections compete with and largely offset each other, but Google stands alone.
Having now completed five experiments with more than 4,000 participants in two countries, I can tell you with certainty that SEME is no longer a matter of speculation. It’s real, and it’s affecting millions of undecided voters right now. Google’s own data – literally, the “scores” they attach to politicians in the middle of campaigns – show dramatic differences in search activity associated with each candidate, which, according to the company’s own PR materials, helps to boost the rankings of high-scoring candidates. Obama’s Google Scores bested McCain’s 90 out of 92 days just before the 2008 election and 86 out of 92 days just before the 2012 election.
And Narendra Modi, the winner in the recent India election, had Google Scores at least 25 percent higher than those of his opponents for at least 60 days in a row before the polls closed on May 12, 2014 – this in a country in which it is illegal for exit polls to be published until after the polls close. Publishing exit polls is prohibited to prevent a bandwagon effect from tilting an election in a way that cannot be countered by opponent candidates.
But no protections are in place to prevent biased search rankings from having such an impact, which can be large even in developing nations. According to Google’s own estimates, 240 million people in India had Internet access during the recent elections, and that number is expected to grow to more than 600 million by 2019.
High-ranking search results cannot be countered by anyone, and a mathematical model we have developed shows how their impact might even feed back over time to produce landslide support.
Even if Google’s executives are as innocent as lambs, do we really want our leaders to be selected by a computer program? That would raise the absurdity of political theater to a new level, if not a new cosmological dimension.
The point of this research is to save democracy – really. As long as an unregulated company has a monopoly on search, it has the power to make democracy as we know it meaningless.
Recently, though, a colleague threw the proverbial spanner into the works. What, he said, if Google really isn’t evil? What if company executives are entirely unaware of the influence their algorithm is having on voters and never even thought of using search rankings to manipulate elections? What if you’re showing them the way?
I’m rarely speechless, but that got me. He might be right.
All the more reason to get our lawmakers moving swiftly while there still might be some who will listen.
To protect our society from deliberate or, worse yet, accidental election manipulation on a large scale, election-related search needs to be regulated, monitored, and subjected to equal-time rules. The alternative is untenable, if not bizarre.
(Emphasis and minor formatting added)