Authorities in the UK have
finally figured out that fake news stories and Russian-placed ads are
not the real problem. The UK Parliament is
about to
impose stiff
penalties—not on the people who place the ads or write the stories, but
on the Big Tech platforms that determine which ads and stories people
actually see.
Parliament’s plans will almost
surely be energized by the latest leak of damning material from inside
Google’s fortress of secrecy: The Wall Street Journal
recently reported on
emails exchanged among Google employees in January 2017 in which they
strategized about how to alter Google search results and other
“ephemeral experiences” to counter President Donald Trump’s newly
imposed travel ban. The company claims that none of these plans was ever
implemented, but who knows?
While U.S. authorities have
merely held hearings
, EU authorities have taken
dramatic steps in recent years to limit the powers of Big Tech,
most recently with a comprehensive law that protects user privacy—the
General Data Protection
Regulation—and a
whopping
$5.1 billion fine against
Google for monopolistic practices in the mobile device market. Last
year, the European Union also levied a
$2.7 billion fineagainst
Google for filtering and ordering search results in a way that favored
their own products and services. That filtering and ordering, it turns
out, is of crucial importance.
As years of research I’ve been
conducting on
online influence has
shown, content per se is not the real threat these days; what really
matters is (a) which content is selected for users to see, and (b) the
way that content is ordered in search results, search suggestions,
newsfeeds, message feeds, comment lists, and so on. That’s where the
power lies to shift opinions, purchases, and votes, and that power is
held by a disturbingly small group of people.
I say “these days” because the
explosive growth of a handful of massive platforms on the internet—the
largest, by far, being Google and the next largest being Facebook—has
changed everything. Millions of people and organizations are constantly
trying to get their content in front of our eyes, but for more than 2.5
billion people around the world—soon to be more than 4 billion—the
responsibility for
what algorithms do should always lie with the people who wrote the
algorithms and the companies that deployed them.
In randomized, controlled,
peer-reviewed research I’ve conducted with thousands of people, I’ve
shown repeatedly that when people are undecided, I can
shift their opinions on
just about any topic just
by changing how I filter and order the information I show them. I’ve
also shown that when, in
multiple searches,
I show people more and more information that favors one candidate, I can
shift opinions even farther. Even more disturbing, I can do these things
in ways that are
completely invisible to
people and in ways that don’t leave paper trails for authorities to
trace.
Worse still, these new forms of
influence often rely on ephemeral content—information that is generated
on the fly by an algorithm and then disappears forever, which means that
it would be difficult, if not impossible, for authorities to
reconstruct. If, on Election Day this coming November, Mark Zuckerberg
decides to broadcast go-out-and-vote reminders mainly to members of one
political party, how would we be able to detect such a manipulation? If
we can’t detect it, how would we be able to reduce its impact? And how,
days or weeks later, would we be able to turn back the clock to see what
happened?
Of course, companies like
Google and Facebook emphatically reject the idea that their search and
newsfeed algorithms are being tweaked in ways that could meddle in
elections. Doing so would undermine the public’s trust in their
companies, spokespeople have said. They insist that their algorithms are
complicated, constantly changing, and subject to the “organic” activity
of users.
This is, of course, sheer
nonsense. Google can adjust its algorithms to favor any candidate it
chooses no matter what the activity of users might be, just as easily as
I do in my experiments. As legal scholar Frank Pasquale noted in his
recent book “The Black Box Society,” blaming algorithms just doesn’t cut
it; the
responsibility for
what an algorithm does should always lie with the people who wrote the
algorithm and the companies that deployed the algorithm. Alan Murray,
president of Fortune,
recently framed the issue this
way: “Rule one in the Age of AI: Humans remain accountable for
decisions, even when made by machines.”
Given that 95 percent of
donations from
Silicon Valley generally go to Democrats, it’s hard to imagine that the
algorithms of companies like Facebook and Google don’t favor their
favorite candidates. A
newly leaked video of
a 2016 meeting at Google shows without doubt that high-ranking Google
executives share a strong political preference, which could easily be
expressed in algorithms. The favoritism might be deliberately programmed
or occur simply because of unconscious bias. Either way, votes and
opinions shift.
It’s also hard to imagine how,
in any election in the world, with or without intention on the part of
company employees, Google search results would fail to tilt toward one
candidate. Google’s search algorithm certainly has no equal-time rule
built into it; we wouldn’t want it to! We want it to tell us what’s
best, and the algorithm will indeed always favor one dog food over
another, one music service over another, and one political candidate
over another. When the latter happens … votes and opinions shift.
Here are 10 ways—seven of which
I am actively studying and quantifying—that Big Tech companies could use
to shift millions of votes this coming November with no one the wiser.
Let’s hope, of course, that these methods are not being used and will
never be used, but let’s be realistic too; there’s generally no limit to
what people will do when money and power are on the line.
1. Search Engine Manipulation
Effect (SEME)
Ongoing research I began in
January 2013 has shown repeatedly that when one candidate is favored
over another in search results, voting preferences among undecided
voters shift dramatically—by 20 percent or more overall, and by
up to 80 percent in
some demographic groups. This is partly because people place inordinate
trust in algorithmically generated output, thinking, mistakenly, that
algorithms are inherently objective and impartial.
But my research also suggests
that we are
conditioned to
believe in high-ranking search results in much the same way that rats
are conditioned to press levers in Skinner boxes. Because most searches
are for simple facts (“When was Donald Trump born?”), and because
correct answers to simple questions inevitably turn up in the first
position, we are taught, day after day, that the higher a search result
appears in the list, the more true it must be. When we finally search
for information to help us make a tough decision (“Who’s better for the
economy, Trump or Clinton?”), we tend to believe the information on the
web pages to which high-ranking search results link.
As The Washington Post
reported last
year, in 2016, I led a team that developed a system for monitoring the
election-related search results Google, Bing, and Yahoo were showing
users in the months leading up to the presidential election, and I found
pro-Clinton bias in
all 10 search positions on the first page of Google’s search results.
Google responded,
as usual, that it has “never re-ranked search results on any topic
(including elections) to manipulate political sentiment”—but I never
claimed it did. I found what I found, namely that Google’s search
results favored Hillary Clinton; “re-ranking”—an obtuse term Google
seems to have invented to confuse people—is irrelevant.
Because (a) many elections are
very close, (b) 90 percent of online searches in most countries are
conducted on just one search engine (Google), and (c) internet
penetration is high in most countries these days—higher in many
countries than it is in the United States—it is possible that the
outcomes of
upwards of 25 percent of
the world’s national elections are now being determined by Google’s
search algorithm, even without deliberate manipulation on the part of
company employees. Because, as I noted earlier, Google’s search
algorithm is not constrained by equal-time rules, it almost certainly
ends up favoring one candidate over another in most political races, and
that shifts opinions and votes.
2. Search Suggestion Effect
(SSE)
When Google first introduced
autocomplete search suggestions—those short lists you see when you start
to type an item into the Google search bar—it was supposedly meant
to save you some time. Whatever the original rationale, those
suggestions soon turned into a powerful means of manipulation that
Google appears to use aggressively.
My
recent research suggests
that (a) Google starts to manipulate your opinions from the very first
character you type, and (b) by fiddling with the suggestions it shows
you, Google can turn a 50–50 split among undecided voters into a 90–10
split with no one knowing. I call this manipulation the Search
Suggestion Effect (SSE), and it is one of the most powerful behavioral
manipulations I have ever seen in my nearly 40 years as a behavioral
scientist.
How will you know whether
Google is messing with your election-related search suggestions in the
weeks leading up to the election? You won’t.
3. The Targeted Messaging
Effect (TME)
If, on Nov. 8, 2016, Mr.
Zuckerberg had sent go-out-and-vote reminders just to supporters of Mrs.
Clinton, that would likely have given her an additional 450,000 votes.
I’ve extrapolated that number from Facebook’s
own published data.
Because Zuckerberg was
overconfident in 2016, I don’t believe he sent those messages, but he is
surely not overconfident this time around. In fact, it’s possible that,
at this very moment, Facebook and other companies are sending out
targeted register-to-vote reminders, as well as targeted go-out-and-vote
reminders in primary races. Targeted go-out-and-vote reminders might
also favor one party on Election Day in November.
My associates and I are
building systems to monitor such things, but because no systems are
currently in place, there is no sure way to tell whether Twitter,
Google, and Facebook (or Facebook’s influential offshoot, Instagram) are
currently tilting their messaging. No law or regulation specifically
forbids the practice, and it would be an easy and economical way to
serve company needs. Campaign donations cost money, after all, but
tilting your messaging to favor one candidate is free.
4. Opinion Matching Effect
(OME)
In March 2016, and continuing
for more than seven months until Election Day, Tinder’s tens of millions
of users could not only swipe to find sex partners, they could also
swipe to find out whether
they should vote for Trump or Clinton. The website
iSideWith.com—founded
and run by “two friends” with no obvious qualifications—claims to have
helped more than 49 million people match their opinions to the right
candidate. Both
CNN and USA Today have
run similar services, currently inactive.
I am still studying and
quantifying this type of, um, helpful service, but so far it looks like
(a) opinion matching services tend to attract undecided voters—precisely
the kinds of voters who are most vulnerable to manipulation, and (b)
they can easily produce opinion shifts of 30 percent or more without
people’s awareness.
At this writing, iSideWith is
already helping people decide who
they should vote for in the 2018 New York U.S. Senate race, the 2018 New
York gubernatorial race, the 2018 race for New York District 10 of the
U.S. House of Representatives, and, believe it or not, the 2020
presidential race. Keep your eyes open for other matching services as
they turn up, and ask yourself this: Who wrote those algorithms, and how
can we know whether they are biased toward one candidate or party?
5. Answer Bot Effect (ABE)
More and more these days,
people don’t want lists of thousands of search results, they just want
the answer, which is being supplied by personal assistants like Google
Home devices, the Google Assistant on Android devices, Amazon’s Alexa,
Apple’s Siri, and Google’s featured snippets—those
answer boxesat
the top of Google search results. I call the opinion shift produced by
such mechanisms the Answer Bot Effect (ABE).
My research on
Google’s answer boxes shows three things so far: First, they reduce the
time people spend searching for more information. Second, they reduce
the number of times people click on search results. And third, they
appear to shift opinions 10 to 30 percent more than search results alone
do. I don’t yet know exactly how many votes can be shifted by answer
bots, but in a national election in the United States, the number might
be in the low millions.
6. Shadowbanning
Recently, Trump
complained that
Twitter was preventing conservatives from reaching many of their
followers on that platform through
shadowbanning,
the practice of quietly hiding a user’s posts without the user knowing.
The validity of Trump’s specific accusation is arguable, but the fact
remains that any platform on which people have followers or friends can
be rigged in a way to suppress the views and influence of certain
individuals without people knowing the suppression is taking place.
Unfortunately, without aggressive monitoring systems in place, it’s hard
to know for sure when or even whether shadowbanning is occurring.
7. Programmed Virality and
the Digital Bandwagon Effect
Big Tech companies would like
us to believe that virality on platforms like YouTube or Instagram is a
profoundly mysterious phenomenon, even while acknowledging that their
platforms are populated by
tens of millions of fake accounts that might affect virality.
In fact, there is an obvious
situation in which virality is not mysterious at all, and that is when
the tech companies themselves decide to shift high volumes of traffic in
ways that suit their needs. And aren’t they always doing this? Because
Facebook’s algorithms are secret, if an executive decided to bestow
instant Instagram stardom on a pro-Elizabeth Warren college student, we
would have no way of knowing that this was a deliberate act and no way
of countering it.
The same can be said of the
virality of YouTube videos and Twitter campaigns; they are inherently
competitive—except when company employees or executives decide
otherwise. Google has an especially powerful and subtle way of creating
instant virality using a technique I’ve dubbed the
Digital Bandwagon Effect. Because the popularity
of websites drives them higher in search results, and because
high-ranking search results increase the popularity of websites (SEME),
Google has the ability to engineer a sudden explosion of interest in a
candidate or cause with no one—perhaps even people at the companies
themselves—having the slightest idea they’ve done so. In 2015, I
published a
mathematical model showing
how neatly this can work.
8. The Facebook Effect
Because Facebook’s ineptness
and dishonesty have squeezed it into a digital doghouse from which it
might never emerge, it gets its own precinct on my list.
In 2016, I published
an article detailing
five ways that Facebook could shift millions of votes without people
knowing: biasing its trending box, biasing its center newsfeed,
encouraging people to look for election-related material in its search
bar (which it did that year!), sending out targeted register-to-vote
reminders, and sending out targeted go-out-and-vote reminders.
I wrote that article before the
news stories broke about Facebook’s improper sharing of user data with
multiple researchers and companies, not to mention the stories about how
the company permitted fake news stories to proliferate on its platform
during the critical days just before the November election—problems the
company is now
trying hard to
mitigate. With the revelations mounting, on July 26, 2018, Facebook
suffered the
largest one-day drop in
stock value of any company in history, and now it’s facing a
shareholder lawsuit and
multiple
fines and
investigations in
both the United States and the EU.
Facebook desperately needs new
direction, which is why I recently called for
Zuckerberg’s resignation. The company, in my
view, could benefit from the new perspectives that often come with new
leadership.
9. Censorship
I am cheating here by labeling
one category “censorship,” because censorship—the selective and biased
suppression of information—can be perpetrated in so many different ways.
Shadowbanning could be
considered a type of censorship, for example, and in 2016, a
Facebook whistleblower claimed
he had been on a company team that was systematically removing
conservative news stories from Facebook’s newsfeed. Now, because of
Facebook’s carelessness with user data, the company is openly taking
pride in rapidly
shutting down accounts that
appear to be Russia-connected—even though company representatives
sometimes acknowledge that they “
don’t have all the facts.”
Meanwhile, Zuckerberg has
crowed about his magnanimity in
preserving the accounts of
people who deny the Holocaust, never mentioning the fact that
provocative content propels traffic that might make him richer. How
would you know whether Facebook was selectively suppressing material
that favored one candidate or political party? You wouldn’t. (For a
detailed look at nine ways Google censors content, see my essay “
The New Censorship,”
published in 2016.)
10. The Digital Customization
Effect (DCE)
Any marketer can tell you how
important it is to know your customer. Now, think about that simple idea
in a world in which Google has likely collected the equivalent of
millions of Word pages of
information about you. If you randomly display a banner ad on a web
page, out of 10,000 people, only five are likely to click on it; that’s
the CTR—the “
clickthrough rate”
(0.05 percent). But if you target your ad, displaying it only to people
whose interests it matches, you can boost your CTR
a hundredfold.
That’s why Google, Facebook,
and others have become increasingly obsessed with customizing the
information they show you: They want you to be happily and mindlessly
clicking away on the content they show you.
In the research I conduct, my
impact is always larger when I am able to customize information to suit
people’s backgrounds. Because I know very little about the participants
in my experiments, however, I am able to do so in only feeble ways, but
the tech giants know everything about you—even things you don’t know
about yourself. This tells me that the effect sizes I find in my
experiments are probably too low. The impact that companies like Google
are having on our lives is quite possibly much larger than I think it
is. Perhaps that doesn’t scare you, but it sure scares me.
The Same Direction
OK, you say, so much for
Epstein’s list! What about those other shenanigans we’ve heard about:
voter fraud (Trump’s
explanation for why he lost the popular vote),
gerrymandering,
rigged
voting machines,
targeted ads placed by
Cambridge Analytica,
votes cast
over the internet,
or, as I mentioned earlier, those
millions of bots designed
to shift opinions. What about hackers like
Andrés Sepúlveda,
who spent nearly a decade using computer technology to rig elections in
Latin America? What about all the ways new technologies make
dirty tricks easier in
elections? And what about those darn Russians, anyway?
To all that I say: kid stuff.
Dirty tricks have been around since the first election was held
millennia ago. But unlike the new manipulative tools controlled by
Google and Facebook, the old tricks are competitive—it’s your hacker
versus my hacker, your bots versus my bots, your fake news stories
versus my fake news stories—and sometimes illegal, which is why
Sepúlveda’s efforts
failed many
times and why Cambridge Analytica is dust.
“Cyberwar,” a
new book by
political scientist Kathleen
Hall Jamieson, reminds us that targeted ads and fake news stories can
indeed shift votes, but the numbers are necessarily small. It’s hard to overwhelm your
competitor when he or she can play the same games you are playing.
Now, take a look at my numbered
list. The techniques I’ve described can shift millions of votes without
people’s awareness, and because they are controlled by the platforms
themselves, they are entirely noncompetitive. If Google or Facebook or
Twitter wants to shift votes, there is no way to counteract their
manipulations. In fact, at this writing, there is not even a credible
way of detecting those manipulations.
And what if the tech giants are
all leaning in the same political direction? What if the combined weight
of their subtle and untraceable manipulative power
favors one political party? If 150 million people
vote this November in the United States, with 20 percent still undecided
at this writing (that’s 30 million people), I estimate that the combined
weight of Big Tech manipulations could easily shift upwards of 12
million votes without anyone knowing. That’s enough votes to determine
the outcomes of hundreds of close local, state, and congressional races
throughout the country, which makes the free-and-fair election little
more than an illusion.
Full disclosure: I happen to
think that the political party currently in favor in Silicon Valley is,
by a hair (so to speak), the superior party at the moment. But I also
love America and democracy, and I believe that the free-and-fair
election is the bedrock of our political system. I don’t care how
“right” these companies might be; lofty ends do not justify shady means,
especially when those means are difficult to see and not well understood
by either authorities or the public.
Can new regulations or laws
save us from the extraordinary powers of manipulation the Big Tech
companies now possess? Maybe, but our leaders seem to be especially
regulation-shy these days, and I doubt, in any case, whether laws and
regulations will ever be able to keep up with the new kinds of threats
that new technologies will almost certainly pose in coming years.
I don’t believe we are
completely helpless, however. I think that one way to turn Facebook,
Google, and the innovative technology companies that will succeed them,
into responsible citizens is to set up
sophisticated monitoring
systems that
detect, analyze, and archive what they’re showing people—in effect, to
fight technology with technology.
As I mentioned earlier, in
2016, I led a team that
monitored search results on
multiple search engines. That was a start, but we can do much better.
These days, I’m working with business associates and academic colleagues
on three continents to scale up systems to monitor a wide range of
information the Big Tech companies are sharing with their users—even the
spoken answers provided by personal assistants. Ultimately, a worldwide
ecology of passive monitoring systems will make these companies
accountable to the public, with information bias and online manipulation
detectable in real time.
With November drawing near,
there is obviously some urgency here. At this writing, it’s not clear
whether we will be fully operational in time to monitor the midterm
elections, but we’re determined to be ready for 2020.
Robert Epstein is a senior
research psychologist at the American Institute for Behavioral
Research and Technology in California. Epstein, who holds
a doctorate from Harvard University, is the former
editor-in-chief of Psychology Today and has published 15 books and
more than 300 articles on internet influence and other topics. He is
currently working on a book called “Technoslavery: Invisible Influence
in the Internet Age and Beyond.” His research is featured in the new
documentary “The Creepy Line.”
You can find him on Twitter @DrREpstein.
Views expressed in this
article are the opinions of the author and do not necessarily reflect
the views of The Epoch Times.