An increasingly popular tactic challenges conventional wisdom on the spread of
electoral disinformation: the creation of partisan outletsmasquerading as local news organizations. An
investigation by the Tow Center for Digital Journalism at Columbia
Journalism School has discovered at least 450 websites in a network of
local and business news organizations, each distributing thousands of
algorithmically generated articles and a smaller number of reported
stories. Of the 450 sites we discovered, at least 189 were set up as
local news networks across ten states within the last twelve months by
an organization called Metric Media.
Titles like the East
Michigan News, Hickory Sun, and
Grand Canyon Times have appeared on the web ahead of the 2020
election. These networks of sites can be used in a variety of ways: as
‘stage setting’ for events, focusing attention on issues such as voter
fraud and energy pricing, providing the appearance of neutrality for
partisan issues, or to gather data from users that can then be used
for political targeting.
On October 20, the Lansing
State Journal first
broke the story of the network’s existence. About three dozen local
news sites, owned by Metric Media, had appeared in Michigan. Further
reporting by the Michigan
Daily, the Guardian and the New
York Times
identified yet more sites. Ultimately, previous reporting has
identified around 200 of these sites. Our analysis suggests that there
are at least twice that number of publications across a number of
related networks, of which Metric Media is just one component.
Over a two-week period starting
November 26, we tapped into the RSS feeds of these 189 Metric Media
sites, all of which were created this year, and found over fifteen
thousand unique stories had been published (over fifty thousand when
aggregated across the sites), but only about a hundred titles had the
bylines of human reporters. The rest cited automated services or press
releases.
Many of the stories attributed to
automated services (Metric Media News Service, Local Labs News
Service, etc.) relied on data releases from federal programs like the
Department of Education and Department of Health and Human Services,
Congressional Research Service, the Federal Election Commission, and
the Census Bureau. Another frequently used data sources in Metric
Media’s automated stories is GasBuddy, a Boston-based startup that
monitors gas station prices in North America and Australia. Some of
the pieces attributed to human writers also rely on public data, from
a Tax Foundation study that established the cents per gallon each
state collected in wine taxes, in one instance, or a builders
association study that rated structural deficiencies in bridges in
another.
Interspersed between stories about real
estate prices and the best place to purchase premium gas based on zip
code, newsier pieces appear, sometimes to quote a state senator about
how the federal
government should not play a role in education.
Others quoted various
exclusively Republican officials like US senators, chairpeople, or
communications staff of the state GOP, about the impeachment process
being a
witch hunt or abuse
of power.
The networks can be traced back to
conservative businessman Brian Timpone. In 2012, Timpone’s company
Journatic, an outlet known for its low-cost automated story generation
(which became known as ‘pink
slime journalism’), attracted
national attention and outrage for faking bylines and quotes, and for
plagiarism. Journatic rebranded as Locality Labs in 2013; Locality Labs is
behind many of the publications we discovered that mimic the appearance
and output of traditional news organizations. These sites do not bear much
information about their political use or funding, but some of them have
been funded by political candidates and lobbying campaigns. Metric
Media, Locality
Labs (or LocalLabs), Franklin Archer,
the Record Inc.,
and Local Government Information Services (LGIS) are the main organizations involved in
operating these networks of publications, and Timpone is associated in one
way or another with each of them. Michigan
Daily has detailed the
convoluted relationship between these organizations.
Some of these mysterious, partisan local
news sites publish physical newspapers and many have minimal social
media presence. At first, they do not appear to be owned by the
same network or organization, but a number of clues suggest that they
are intimately linked. Our analysis demonstrates the links between the
networks by identifying shared markers, such as unique analytics tokens,
server IP addresses, and even shared design templates and bylines on
articles. Further, the Privacy
Policy and Terms
of Service for many of
these websites—but not all—suggest they are part of Locality Labs,
LLC.
Websites and networks can aid campaigns
to manipulate public opinion by exploiting faith in local media. The
demise of local journalism in many areas creates an information vacuum,
and raises the chance of success for these influence campaigns. The
strategy is further made possible by the low cost of automating news
stories, repurposing press releases (including obituaries from funeral
homes), and replicating design templates, as well as the relative ease
with which political or single-issue campaigns can obscure their funding
and provenance.
One
in five newspapers has closed in the past decade,
and many other outlets have significantly
cut staff and reporting.
This bleak landscape for local news creates space for this cottage
industry of outlets that claim to produce journalism—typically during
election cycles—and create and amplify material to support political
outcomes, sometimes even physically mailing advertising materials that
masquerade as newspapers but are not labeled political advertising.
These publications cover certain candidates and topics, including
limited government, tort reform, and labour unions, with a conservative
bias.
It is not clear how effective the sites
are, but their architecture and strategy is useful to understand the way
they co-opt the language, design and structure of news organizations.
Automation has been touted as a way to create stories where there are
few reporting resources, and it can be used to build credibility. It can
also make a news organization look far more prolific than it is.
Potentially adding to the credibility of these sites is their Google
search ranking: in the case of some of the websites set up in 2015-2016,
we observed that once sites had gained ample authority, they appeared on
the first page of Google Search results just below the official
government and social media pages.
The Lansing
State Journal’s story highlights Metric Media’s intent
to launch hundreds of such
sites to “fill the void in local communities.” And, while Metric Media
hasn’t been embroiled in controversies like its sibling organizations,
the Journal’s story quotes Matt Grossmann, director of
Michigan State University’s Institute for Public Policy and Social
Research, who said, “many [articles] on [Metric Media site The] Lansing Sun
appeared to be right-leaning.”
Using multiple network analysis tools, we
found at least 450 websites including the ones identified by previous
reporting. There are dedicated networks with anywhere between five and
forty sites per state in eleven states. Additionally, every state in the
US (and Washington DC) has a dedicated “business daily” (e.g. Michigan Business Daily,
Utah Business Daily), and there are several international
“business daily” websites (e.g. Toronto
Business Daily, Mexico Business Daily, Manila Business Daily) as well.
These are augmented by
several single-subject sites (e.g. American
Security News, Tobacco Newswire,
Farm Insurance News), and community sites that aggregate events,
news, and business services information (e.g. Metro
East Today, Wilton Review, Palmerton Guide).
450 sites, 12 state networks, 5 corporate entities
The 450 websites span twenty-one
different networks operated by five distinct corporate entities. Below,
we provide further detail about each of these organizations and how
they’re linked.
Figure 1: Breakdown of sites across networks,
including the sites for multiple corporate entities or parent
organizations (“Organizations”) such as LGIS, Franklin Archer, Metro
Business Network, or Dan Proft’s PAC Liberty Principles.
The overall distribution of sites across
the networks is shown in Figure 1, with Franklin Archer’s Local News
Network the single largest network with almost 130 sites. In the case of
Franklin Archer, we conflated nine standalone sites that bear all the
heuristics of their local news sites, even though these sites were not
explicitly listed on the organization’s website. These include Surprise Journal,
El Paso Review,
and Lansing Reporter.
With the Local News Network sites added
to the 51 sites in the Metro Businessnetwork,
the overall number under the Franklin Archer banner stands at 179, a
close second to Metric Media, which has 189 publications across ten
states. Metric Media’s North Carolina operation alone has 46 sites,
making it the third biggest individual network in our dataset.
For clarity, we have also conflated 25
standalone sites—sixteen topical sites, and nine international
sites—under “LocalLabs: International” and “LocalLabs: Topical” in Figure 1.
These sites have a Locality Labs,
LLC privacy policy or shared
analytics identifier, but no other associations. In some cases, the
privacy policy is not even linked on the homepage. The Mexico
Business Daily, for
example, doesn’t mention any affiliation in its “About Us” or “Contact”
pages, and there is no privacy policy on its homepage. This
link suggests it is a
Locality Labs property. CISTRAN
Finance follows the
exact same pattern. In other cases, like the Toronto
Business Daily or the
Balkan News Wire, the
privacy policy is linked to from the homepage, but no other affiliations
are mentioned.
In our analysis so far, websites on the
Metric Media state-centric networks only link to other sites from the
same state, including stories from Franklin Archer’s Metro Business
network. There are, however, few indicators that the sites are operated
by the same organization. On sites in both networks, the privacy
policies and terms of service suggest they are operated by Locality
Labs, LLC. Stories covering vastly different topics across the network
share bylines, as well as similar—often identical—templates.
These anecdotal pieces of evidence can
only be gleaned after identifying the scale of the operation. To measure
the scale of the operation, we relied on website metadata and network
forensics: the sites shared IP addresses and various analytics
identifiers. Within the same network—and even the same organization—that
is expected behavior. It is highly unusual for sites to share IP
addresses and analytics identifiers if they are not connected, but
that’s exactly what we found in these networks, as can be seen
below.
Shared IP addresses
Each domain maps to an IP address, which
identifies the server where the site is hosted. Across this greater
network of networks, including the domains for the organizations, we
found about twenty-five unique IP addresses. Some of these networks,
including the biggest—Franklin Archer: Local News Network—don’t share an
IP address or server with any other network as can be seen in Figure 2.
Some only share IP addresses with other sites that belong to the same
parent organization. For example, some state-level Metric Media
properties—the publications in New Mexico, Arizona and Montana—share a
server.
But, Metric Media’s Maryland properties
(which includes North Baltimore
Journal, Harford
News, Montgomery
News) also sit on the
same server as the LGIS sites, which is more unusual. Similarly, it is
improbable for sites within Franklin Archer’s Business Network to share
an IP address with LocalLabs’ international or topical sites, unless the
two organizations share resources.
Some of the parent organization’s
websites live on the same server as some of the LocalLabs’ topical
properties, too. To wit: The Record (therecordinc.com), Franklin Archer
(franklinarcher.com) and Dan Proft’s super PAC Liberty Principles
(libertyprinciples.com) are on the same server as Tobacco
Newswire, FDA Reporter,
and FDA Health News.
Figure 2: Network graph of shared IP addresses;
each different edge colour represents a different IP address
Shared Google Analytics IDs
Even though there were over twenty Google
Analytics IDs, only five were used on more than one network. Of these,
three identifiers were shared within Metric Media sites only, while
another three were shared among Franklin Archer, LGIS, and the LocalLabs
sites (Figure 3). It is highly unlikely for independent,
supposedly unrelated organizations like LGIS and Franklin Archer to
share analytics identifiers.
Figure 3: Network graph of shared Google
Analytics IDs; each different edge colour represents a different Google
Analytics ID
At the time of writing, the Franklin
Archer organizational website (franklinarcher.com), publications
belonging to LGIS, and topical websites like American
Security News, Power Newswire,
and Higher Education Tribune shared the same Google Analytics ID.
International sites like Gulf
News Journal and a handful
of business dailies also had that same identifier.
A Single Shared NewRelic ID and Quantcast ID
Perhaps no other identifier illustrates
the convoluted nature of these networks as well as the NewRelic ID (Figure 4) and the Quantcast ID (Figure
5).
Most of Metric Media’s properties didn’t
have a corresponding NewRelic ID, nor did the Florida network’s. ther
networks, though, shared the same identifier, including Metric Media’s
Maryland sites. Also, for the first time, we see the publications within
the Record network, the first site of which was established in 2004,
sharing an identifier with the greater network.
Figure 4: Network graph of the single shared
NewRelic ID
Similarly, a single Quantcast identifier
was used across LGIS, Record, and Metric Media sites. (Figure 5).
Much like the NewRelic ID, this too shows the connected nature of the
networks.
Figure 5: Networks connected by a single unique
Quantcast ID. The Franklin Archer publications don’t seem to be relying
on Quantcast identifiers at all, but Metric Media does, as do the Record
network and LGIS.
When we discovered the high volume of
sites intricately linked across a convoluted network, we set about
trying to identify the people behind these organizations, and their
backgrounds. In reporting from the Lansing
State Journal, Michigan Daily,
and the Guardian, as well as our own analysis, one name
appeared time and again: Brian Timpone.
The sites in our dataset go back to 2004
with the inception of the first publication—the Madison-St.
Clair Record—which is now
part of the Record network. In the next section, we’ll trace
through the emergence of various publications, networks, and their
impact.
The Record Network
The Record Network is a system of eleven
publications, the first of which—the Madison-St.
Clair Record—went live in
2004. The network’s publications cover the legal system in eight states;
one of the outlets, Legal Newsline, is not bound to a single state.
Even before Journatic was set up in 2006,
Timpone was the founding publisher of sites in the Record network, which
have come
under scrutiny for being run
by advocacy groups
disguised as media organizations. Timpone is currently listed as
founding publisher of the Madison-St.
Clair Record and the Louisiana Record
identifies him as the publisher. According to the “About Us” page on a
subset of Record network sites (including Legal
Newsline, Madison-St. Clair Record
and Florida Record), they are owned by the US
Chamber Institute for Legal Reform,
an affiliate of the US Chamber of Commerce—an organization, according to
OpenSecrets, that since 2008 has been one of the biggest
“dark money” spenders in US elections. Fifteen years later, these sites
are still very much active.
For the last twenty months or so, the
website for LocalLabs has simply said “We’re improving our site. Stay
tuned for a new and improved LocalLabs.com.” Previously, the website detailed the services the organization provided,
declaring that “LocalLabs offers community news production, advertising
and advertorial content, print special sections and web and mobile
application management.”
We found over forty websites with the
disclaimer “a product of LocalLabs” or a privacy policy/terms of service
page suggesting the sites are operated by Locality Labs, LLC.
Fifteen of these sites, all registered
in 2018, belong to a single
network in a single state: Florida. At the time of writing, we had not
found a Metric Media operation in Florida.
The remaining publications are
single-topic sites or international business sites. Examples of these
sites include Manila Business
Daily, Toronto
Business Daily, and Mexico Business Daily.
The single-topic sites are more US-centric, such as American
Security News, American
Pharmacy News, and Higher
Education Tribune. While
these don’t say they are a product of LocalLabs, their privacy policies
are attributed to Locality Labs.
Bylines on these sites are shared across
other sites in other networks as well.
Newsinator and LGIS
In 2015, Timpone incorporated another organization: Newsinator, a firm that, among other things,
had a “history of doing paid political work and offers marketing
services to companies under the name Interactive Content Services,”
according to the Chicago
Tribune.
In 2015-16, according to the Illinois
State Board of Election Expenditure data,
Newsinator was paid over $300,000 for “Advertising – newspaper” by Liberty Principles,
a conservative super PAC, run by Dan Proft, described as “one of the state’s most visible and
controversial political figures.” According to the data, this $300,000
was split between twenty different candidates’ campaigns, most of whom
were running for state office in Illinois in 2016.
The line between Locality Labs and
Newsinator is blurry: Proft, in
his own words, works with
reporters at Locality Labs and “suggests stories and discusses what
reporters should cover.” Newsinator, on the other hand, was responsible
for mailing publications
created by Proft’s organization to
voters, in the form of weekly eight-page print editions. Additionally, a since-dismissed
Federal Election Commission complaint says
Proft created Local Government
Information Services (LGIS)
in 2016, which “took over production and distribution of newspapers.” State records confirm that Proft was the President of LGIS,
but the complaint also says that as of January 2019, the corporation’s
status was “dissolved” involuntarily. At the time of writing, LGIS has
33 Illinois-based sites, eleven of which have corresponding print
publications.
The extent to which Locality Labs was
involved in the business of LGIS and Newsinator is highlighted in an FEC lawsuit, filed for violating the Federal Election
Campaign Act of 1971 by “re-publishing campaign materials in a format
designed to look like local community papers.” The respondents included
the obvious suspects: the campaign (Khouri for Congress), the PAC
(Liberty Principles), and LGIS (the company publishing the material). In
addition, it also included Locality Labs. These entities were charged
with violating the Federal Election Campaign Act of 1971 by
“re-publishing campaign materials in a format designed to look like
local community papers.” The details of the lawsuit further stipulate
that LGIS “contracts with various entities, including Locality Labs […],
to prepare content for publications” including providing “local and
state news in certain geographical areas in Illinois.” LGIS, however,
has full editorial control. The lawsuit was dismissed in July
2019.
Two networks—DirecTech(Local
News Network) and Metro Business Daily—and almost 180 sites sail under the banner of
Franklin Archer,
a company established in 2018. Brian
Timpone’s brother, Michael
Timpone, is CEO. On the Franklin Archer website, the company boasts that
it is “the largest producer of local news in the United States” with
over 200 active websites, over three million monthly page views, and 600
stories produced daily.
The 128 Local News Network sites we found
promise communities a way to “browse the latest news, investigate
upcoming events, stay in the know” by creating and maintaining “the most
comprehensive and easily searchable database of all local businesses by
placing an emphasis on consumer needs.” They are all “a product of
LocalLabs.” (note: some of the sites mentioned under “Our
Publications” lead to dead
links; we also found nine websites that follow the exact same pattern,
but are not explicitly mentioned on the Franklin Archer website.)
The Metro
Business Network website
also doesn’t explicitly state its connection to either Franklin Archer
or LocalLabs, but its sites are the same as those in Franklin Archer’s
“Metro Business Daily” network: one “business daily” per state and one
for Washington DC. And, like the Metric Media websites, their privacy policy
and terms of service indicate that they are Locality Labs
properties.
Today, the Franklin Archer organizational
website lives on the same server as Dan Proft’s super PAC: Liberty
Principles. A Franklin Archer worker, who asked not to be named, said
the company’s writers are paid by Newsinator. According to the Iowa corporate registry, Newsinator and Franklin Archer are both
alternative business entity names for DirecTech, LLC.
Metric Media
The Metric Media Foundation was created
in 2019 by Bradley Cameron, whose link to Timpone (if any) remains
unclear. However, his sites too sport Locality Labs’ privacy policy and
terms of service, and share servers and analytics identifiers with some
of the other organizations.
Our analysis found Metric Media has 189
sites across ten states. According to their “About
Us” page, the Metric Media
Foundation “funds more than 900 news sites that generate over 3 million
monthly page views” and its operation “also produces over 600 stories
daily, making it the largest producer of local news content in the
nation.” The names of these news sites are not listed on their
website.
The numbers boasted by Metric Media are
identical to the numbers boasted by Franklin Archer: the 3 million
monthly page views and 600 stories produced daily. But the relationship
between Franklin Archer and Metric Media is not articulated on either
organization’s website.
When we asked about the emergence,
structure and scale of these networks as well as their business model,
Timpone said, “Locality Labs has used its proprietary technology to
transform public records into news for major newspaper/media companies
around the US, as well as digital start-ups like LGIS.” Previously,
Brian Timpone has said, “I’m a biased guy. I’m a Republican.”
Despite the different organization and
network names, it is evident these sites are connected. Other than
simply sharing network metadata as described above, they also share
bylines (including “Metric Media News Service” and “Local Labs News
Service” for templated stories), servers, layouts, and templates.
Typically, creating entities that focus
on communities, local news, and single issues important to the general
public would be a worthwhile endeavour. But, the partisan material
present on the more established networks along with the ideological
leanings of some of the key personnel give us pause. The approach
adopted by Dan Proft’s LGIS network during the 2016 and 2018 election
cycles provide a model for these relatively newer networks. The Lansing State Journal
has highlighted stories about voter fraud using data from the Heritage
Foundation on the Metric Media Michigan sites. Similar stories have also
been published on the Arizona, Kentucky, North
Carolina, and Iowa sites.
Over the course of the 2020 campaign the
Tow Center will continue to monitor these networks and others like it to
better understand how these types of activity affect the course of
campaigns, how they work, what their effects are, and, crucially, what
the overall impact might mean for both news consumers and audiences. We
intend to augment our initial analysis with further qualitative research
across all the networks we find as the campaign progresses. In due
course, we intend to release our datasets for other journalists and
researchers to examine.
If you are a local reporter and have
examples of similar networks operating in your area, please get in touch
with us here. If you have written a piece about the
phenomenon in your local community, then please send us links. The
phenomenon is often associated with right-leaning causes and
personalities, but we believe the strategies could be adopted across the
political spectrum.
How we established the scope of the network
The majority of this research was
conducted using a combination of tools and products available online.
Below we provide a short introduction to these tools and outline our
process for students/journalists wishing to conduct similar analyses. If
you’d like further help or guidance for similar investigations, please
get in touch.
To create a website, two basic steps need
to be followed: the domain or web address needs to be registered, and a
web host set up to house the actual content of the website. When the
first step is taken, a WHOIS
record is created, which
may include the details of the registrar (private WHOIS entries, where
the details of the registrant are hidden or redacted, are becoming
increasingly common, especially due to recent
regulation in Europe).
And, when a website goes live, the
content is pulled from a web host or server, which has a corresponding
IP address (analogous to a phone number for a person). The IP address
not only tells us where the content for the website is hosted, but it
can also reveal what other websites reside on the same IP address.
The code that drives the website can
contain metadata, which reveals other identifying traits, including
analytics identifiers. Tracking identifiers allow websites to get
aggregate user information and analytics, including demographic details,
the amount of time spent on a specific page on the site, and how users’
navigated the site. It is highly unlikely for unrelated websites to
share tracking identifiers, because these analytics services provide
business intelligence and insights that can result in competitive
advantage.
Therefore, by finding domains registered
by the same people, websites residing on the same server (i.e. websites
pointing to the same IP address), and websites sharing the same
analytics IDs, it is relatively straightforward to build out the
network. There are various free, freemium, commercial, and open-source
tools available that help extract information at this level.
To ensure thoroughness, manual analysis
is necessary coupled with original reporting from outside sources and
research tools that allow us to explore corporate (including non-profit)
entities, look up campaign finance details, and see historical content
of websites. And, to ensure no false positives (sites incorrectly
identified to be part of the network) slip in, manual confirmation is
necessary.
For our analysis, we used:
RiskIQ to identify websites that shared tracking
identifiers (Google Analytics, Facebook Pixel, Quantcast, or NewRelic)
SecurityTrails and Farsight DNSDB to determine which websites were housed on
the same servers
Lexis Nexis to research the corporate entities and key
people
Starting with one of the domains from the
original Lansing State
Journal story, we found
that many of the stories were sourced from Michigan
Business Daily. A quick
look at the Michigan Business
Daily website immediately
informed us that similar business dailies existed in every state and
Washington D.C.
We looked up one of these business
dailies (Rhode Island Business
Daily) on RiskIQ,
a self-proclaimed “cyberthreat intelligence tool,” to find which other
websites shared the same identifiers (Figure 6). One identifier type
proved to be key: the NewRelic ID. RiskIQ found 138 sites shared the
same NewRelic identifier. The sites included Chicago
City Wire, Cook County Record,
Legal Newsline,
Florida Record, and Blockshopper.
Figure 6: Search by Tracking ID on RiskIQ.
The link to Blockshopper—and the
subsequent WHOIS lookup—confirmed there was an overlap between sites
registered by Brian Timpone and the sites identified by RiskIQ,
including Chicago City Wire, Kane
County Reporter, and
SE Illinois News. Manually exploring some of these
sites, it was evident that some of the underlying technologies and
templates they were using were identical, and yet part of separate
networks. Anecdotally, one more thing stood out: different stories
across states, topics, and networks had many of the same bylines.
Using these 138 sites as our base set, we
input each of the domains into SecurityTrails and Farsight DNSDB to find
which other sites were on the same IP address. Farsight DNSDB and
Security Trails are both passive DNS tools, which amongst other things
can give users details about website infrastructure, including what is
the IP address of the server the website lives on (Figure 7.1) and
who its mail provider is. Further, given an IP address, passive DNS
tools provide users with all websites that share the same IP address
(Figure 7.2). Websites with the same IP address are not necessarily
related though. Websites hosted by companies like Squarespace and Wix
(and many others) are likely to share IP addresses (lookups on two
websites—one hosted by Squarespace and one by Wix—showed more than five
million domains for each of the two IP addresses). Similarly, “parked”
domains—domains that have been registered but not linked to any
web-hosting—are likely to share IP addresses, as the web registrar will
simply assign an IP address to said domain.
Figure 7.1: Extracting the IP address from a
domain in SecurityTrails Figure
Figure 7.2: IP address lookup on on Farsight
DNSDB using DNSDB Scout
We used both tools, plugging in the IP
addresses of the domains we had identified, to get all other websites
that were hosted on the same server. It might seem redundant to use
multiple services to do exactly the same thing. However, the algorithms
and technologies that back each of these services are different, and the
extent of their archival/historical data vary as well. Consolidating
data across services increases the odds of getting a more complete
dataset. (Note: you can use both tools to get both, the IP address of a
website and the list of websites pointing to the same IP address; there
are also quicker ways of doing this, but this is the easiest)
Once we extracted still more domains from
passive DNS lookups, we cross-checked these domains on RiskIQ, and the
cycle continued. Combining the outputs from these sources, we were able
to map out the network.
We are grateful for the guidance and
advice from Farsight Security and REN-ISAC (The Research and Education
Networking Information Sharing and Analysis Center).
We can put 500,000 bot generated comments on any website on Earth.
We can rig America's Got Talent and American Idol Votes. We can make a
100% fake national election happen. We are "GOOGLE"!
Text-generation software is already good enough to fool most people most
of the time. It’s writing news stories, particularly in sportsandfinance. It’s talking with customers on merchant
websites. It’s writing convincing op-eds on topics in the news
(though there are limitations). And it’s being used to bulk
up “pink-slime journalism”—websites meant to appear like legitimate local
news outlets but that publish propaganda instead.
There’s a record of algorithmic content pretending to be from
individuals, as well. In 2017, the Federal Communications Commission had
an online public-commenting period for its plans to repeal net neutrality.
A staggering 22 million comments were received. Many of them—maybe half—werefake, using stolen identities. These comments were also
crude; 1.3 million were generated from the same template, with some words
altered to make them appear unique. They didn’t stand up to even cursory
scrutiny.
These efforts will only get more sophisticated. In a recent experiment,
the Harvard senior Max Weiss used a text-generation program to create 1,000 comments in response to a
government call on a Medicaid issue. These comments were all unique, and
sounded like real people advocating for a specific policy position. They
fooled the Medicaid.gov administrators, who accepted them as genuine
concerns from actual human beings. This being research, Weiss subsequently
identified the comments and asked for them to be removed, so that no
actual policy debate would be unfairly biased. The next group to try this
won’t be so honorable.
Chatbots have been skewing social-media discussions for years. About
a fifth
of all tweets about the 2016 presidential election were
published by bots, according to one estimate, as were about a third of all tweets about that year’s Brexit
vote. An Oxford Internet Institute report from last year found evidence of bots
being used to spread propaganda in 50 countries. These tended to be simple
programs mindlessly repeating slogans: a quarter million pro-Saudi “We all
have trust in Mohammed bin Salman” tweets following the 2018 murder of Jamal
Khashoggi, for example. Detecting many bots with a few followers each is
harder than detecting a few bots with lots of followers. And measuring the
effectiveness of these bots is difficult. The best analyses indicate that they did not affect
the 2016 U.S. presidential election. More likely, they distort people’s
sense of public sentiment and their faith in reasoned political debate. We
are all in the middle of a novel social experiment.
Over the years, algorithmic bots have evolved to have personas. They have fake names, fake bios, and
fake photos—sometimes generated by AI. Instead of endlessly spewing
propaganda, they post only occasionally. Researchers can detect that these
are bots and not people based on their patterns of posting, but the bot
technology is getting better all the time, outpacing tracking attempts.
Future groups won’t be so easily identified. They’ll embed themselves in
human social groups better. Their propaganda will be subtle, and
interwoven in tweets about topics relevant to those social groups.