SaTC PI Meeting ’19 Kate Starbird

SaTC PI Meeting ’19 Kate Starbird


>>Great so, we’re going to move onto our
first keynote talk, Kate, maybe you want to come up. We are please to do have Kate Starbird from
the University of Washington who will talk about her work on understanding online disinformation
and its spread. Kate is at the University of Washington. Her research situated between information
technologies are used during crisis events. Her work is I think an excellent example for
highlighting the expanding reach and influence of cyber space and how establishing a defending
trustworthiness in today’s online information ecosystems as well as future ecosystems requires
new approaches and perspectives beyond maybe what we traditionally thought of as cyber
security. So I am delighted to have Kate with us.>>Thank you for that introduction and I have
a couple more affiliations to mention. One is that I’m founding PI of the center
for an informed public at the University of Washington going up and I’m currently a visiting
professor at Stanford at the cyber policy center while on sabbatical. I’m going to jump in here to this talk so
fake news. Political propaganda, active measures, information
war, in the last few years we’ve been introduced to a lot of new terms or old terms reintroduced. These terms that are meant to describe some
of the toxicities we’ve been receiving in our information spaces and I’m going to focus
on a subsection of this problem and on one term in particular. Disinformation. And I’m going to talk to you about that in
the context that we’ve been studying it. Online during crisis events. We’ve actually been looking at disinformation
several years now but we didn’t start out aiming on this. We just sort of just stumbled down the rabbit
hole. My first line of research was examining digital
volunteers during crisis events. How volunteers organize themselves in online
tools during political protests. In 2013 my colleagues and I set out to examine
these during crisis events because we saw this as an issue becoming more salient and
in early 2016 we began to realize that we weren’t seeing accidental rumors and misinformation
we were seeing intentional and pervasive disinformation. And soon that became the focus of our research
so a few different goals for this talk. My first is to understand what disinformation
is. I’m going to define the term and give it a
bit of a historical context and I will provide case studies of online disinformation and
connect those to misperceptions that we have. Try to correct those and add a little nuance
to our understanding and finally I will talk about the implications for the design of technology,
policy and some other things as well and perhaps have some questions to talk about more broadly
of what the implications of this in this way are for some of your work. Throughout this talk I hope to communicate
why disinformation should be a core concern of those who care about secure and trustworthy
systems. The first thing I want to do is distinguish
between misinformation and disinformation. The differences between the two are especially
important when we start to think of solutions. And one high level way to characterize the
distinction is simply as a matter of intent. Misinformation is information that’s fault. But not necessarily deliberately so. While disinformation is intentionally false
information. From a high level, I think this definition
works. This is the definition I give my undergraduate
students when they want me to break this down a little more simply. But in another view, disinformation isn’t
simply about intent There’s other specific means. It’s been used to describe political propaganda
and connected to this term. Originating in Soviet intelligence. Our understanding is that it is formed by
the work of a practitioner of Soviet disinformation. A member of Czech intelligence and defended
the United States in 1968 and later became a professor of disinformation studies. He’s one of the few academics we can cite
in the literature and orient our study around that. Much of what we know we learned from him. Borrowing from Bittman we define disinformation
as a suite of tactics to achieve strategic political objectives. There’s other kinds of disinformation out
in the space but we try to focus on this. It’s important to note that not all this information
is connected to Russia. It’s extremely important I think the new emerging
threats are more of domestic disinformation between domestic and foreign actors. My talk will focus on case studies in direct
connection with Russia. Because we want to appreciated them for how
they defined this term and used it so I’m going to present three recent case studies. Two with a strong connection to Russian intelligence
and the other is a bit more tenuously connected with some historical connections. These are long-term studies. They took place over the course of multiple
years in our lab and I’m going to take a — the perspective I’m going to take is from my field. We look and we’ve listen look at disinformation
as a kind of collaborative work between agents and crowds. And I’m going to hopefully show you a bit
of that and why that has implications for our work. We apply this sociotechnical lens that they
were talk about here. We apply the sociotechnical lens exploring
the relationship between social structures and technical structures and human action. So in other words, we want to look at how
technology, social norms and social behaviors are mutually shaping. We theorize that disinformation is another
shaping force. And through this lens we focus on collaborative
work that produces and distributes disinformation so we study how diverse and distributed groups
of people connect and work together towards various outcomes, intentional and otherwise. And how that (media is offline). How all these are mutually shaped (media is
offline) the other platforms our data often start with twitter data collections, and we
use those to map out surrounding information spaces. This is a limitation. We’re hoping to start to work around it, talking
to techs. But I know the academic community has been
a beneficiary in some ways of twitter’s open data policy but that also has some limitations
we can talk about at the end. So I’m going to jump into the three case studies
and they might be a little intense as we move across them but this is the meat of the talk. All right. So our first case study is on Russian interference
in the 2016 election. And this case as I mentioned before, we actually
didn’t go in with the intention of studying mis or disinformation. In this case I had an undergraduate student
who really wanted to study how the Black Lives Matter community was organizing and he was
really interested in that and so we began a study — and we began to look at framing
contests within that discourse. So we were looking at primarily Black Lives
Matter activists who were advocating systemic injustice and political violence against African
Americans and those who challenge the message. This conversation was highly polarized. From data collected related to shooting events
in 2016 we generated a dataset that had tweets that had either Black Lives Matter or Blue
Lives Matter in the text or the content of the tweet. And we created a structural graph. In this graph each node or circle is an account
and the accounts are — the edges are invisible but the accounts are clustered closer together
if they retweeted each other and they’re farther apart if they didn’t and the graph reveals
two separate communities or echo chambers on either side of this conversation. There’s pro-Black Lives Matter on the left. So these accounts were political left leaning
and anti-Black Lives Matter on the right were they were right leaning or conservative. When people talk about online polarization
this is what it looks like structurally. The two sides, if we start to look at the
content the two sides spread frames about police shootings of African American citizens. At times the discourse was emotional, divisive
and uncomfortable for our research team. We managed to complete the study in October
2017. We published a paper at CSCW, our core conference
and a few weeks later in November 2017, twitter released its first list, they were working
with — they were actually pressured by the Senate intel committee or the House intel
committee actually to release a list of accounts they had identified as part of Russia’s internet
operations. The disinformation operations that they were
running in 2016 and so twitter had identified a bunch of accounts that they found were working
with the Internet Research Agency in St. Petersburg if you don’t know much about this I will let
you look it up. So these accounts were associated with the
IRA. So they put out the list and I was on twitter
one night, late, and I saw my community was saying, hey, there’s this list out there and
so I went there and I looked at the list and I started looking through the accounts. And I recognized some of them. And I recognized them from our qualitative
research that we’d been doing on this graph and I said, wow, these kind of look familiar
so I sent the list off to my students who luckily weren’t sitting on twitter that night
but doing other better things and they decided to cross-reference the list of the IRA Russian
trolls or agents or the accounts that they were running with our graph. And this might not surprise you now it certainly
surprised us then. This is that graph. And those trolls were active on both sides
of the conversation. So the troll accounts here are in orange and
we highlighted the retweets with edges in this case. So you can see who was retweeting whom and
so this graph reveals that Russian agents were active on both sides of the Black Lives
Matter conversation. A few among the most influential. An orange account on the left was retweeted
by Jack at twitter and me. Several orange accounts on the right had integrated
into other grassroots efforts on the right side. They were connected to other organizing efforts
happening there my Ph.D. student found these accounts were across platforms that played
on stereotypes of African Americans on one side and white U.S. conservatives on the other. They were impersonate ing activists but moderating
others. Reflecting norms but also to some extent possibly
likely shaping them. Often, their content wasn’t superficially
problematic or any different from what others in the space were doing. They were tweeting about supporting strong
black voices on the left or U.S. veterans on the right. They were becoming part of these communities. In other places they were so sowing and amplifying
division. Some of the content was the most vitriol on
the space. Using racial epithets on the right. In a few places we can see them holding arguments
with themselves. Like a puppeteer they had one account arguing
on the right to one on the left. So we now have a much more complete view of
the IRA activity in 2018 through twitter’s project sunlight they released all of the
tweets they associated with this set of IRA based accounts. And they found additional accounts and released
those accounts and tweets as well. In total in fact this total’s probably higher
than this now but in this set here they had about 3,000 different accounts adds — as
part of their operations. This is focused on English here. This is a temporal graph of their activity
with tweets down at the bottom and in the darker color gray are the retweets. What you can see here is for years IRA troll
accounts toiled in relative obscurity. There’s a few here. And what happens is — you know, they’re picking
up few retweets, they’re picking up few followers but that all begin to change in December 2015
and what happened in 2015 December is they began to create these political personas we
saw in our Black Lives Matter data and when they began to create those personas that’s
when things took off. There’s about 300 accounts that are associated
with sort of these online personas of African American Black Lives Matter activists and
white U.S. conservatives which is red in our next graph. When they began to perform these personas,
their accounts started finally to gain traction. And in the end nearly 85% of the retweets
that the IRA accounts received through 2018, were retweets of less than 300 accounts dedicated
to the political personas we shown above in other words the IRA trolls the disinformation
operations were most effective not in spreading fake news or doing some of the other things
they were trying to do but in infiltrating online political communities. Stepping back, we can see this as a multiyear
campaign to infiltrate and shape political discourse in the United States. It was orchestrated by a group with Russian
tie to the government and we can see them burrowing into otherwise organic online communities. We can categorize this activity as orchestrated
which is going to contrast with some of our upcoming examples. It was coordinated by a single organization
in a building in St. Petersburg. We don’t know how it will work. We have people who sent information out but
don’t have a complete picture. However we can assume it’s a part — in part
organized. And another way to look at this activity is
sort of a new campaign. As opposed to what we’re going to see later
which are campaigns that have taken root and are a little bit older. All right. So that’s the first case study and I’m going
to switch very quickly to the next case study. Which is in fact probably the hardest one
for me to give. It’s a — it was — it has been very intense
to study this content. And it is only getting worse as the days go
on. But our second case study looks at disinformation
during armed conflict since 2017 we’ve been conducting research on online discourse related
to the Civil War in Syria. In particular we focus on conversations around
a Syrian civil defense. They did, my verb tenses are changing right
now as I’m not sure they’re able to be active anymore or they’re kind of probably at the
end but I think they are still trying to do search and rescue and providing medical aid
to people impacted by the Civil War there but again, they’re located in rebel areas. They were not — they are targets of the government. And not only do they work and try to respond
to these disasters they also documented the impacts and that’s where I think they ended
up becoming a target of a disinformation campaign. They started calling out the impact of air
strikes by the Syrian government and the Russian allies. And also documenting evidence of chemical
weapons attacks starting I think in 2014 and moving into 2015-16 the one above calls out
the Assad regimes Russian allies and claiming they specifically target medical aid workers. So this content she were sharing especially
in about 2016 they were starting to get their message out into the western world with their
cause to wake up the world that there were these atrocities going on there that were
effects citizens and in 2016 they were the focus of an Oscar winning documentary which
likely catalyzed this campaign that I’m going to show you against them. So in 2017-18 when we started studying this
conversation if you searched twitter or Google for white helmets you may have found tweets
supporting them but you most likely would have found content like this challenging the
white helmets, equating them with terrorists, calling them crisis actors or accusing them
of any number of atrocities. We were working with folks from the Harvard
humanitarian coalition. And we decided that, you know, we didn’t have
a lot of access to data so we started with twitter and we’re actually going to be looking
at the English speaking conversations so the disinformation campaign that’s targeted towards
western audiences which would be different than how it might look in other languages. This is one representation of that set. That dataset. It has about two million tweets over the course
of a year. In the center of this graph is a retweet graph
network. And in this graph the accounts are sized by
the number of retweets they get so the larger accounts are are influential in this space. And the edges here are visible. Around the outside is a different kind of
note I’m going to talk a little bit about that later. So using a community detection algorithm we
identified two major clusters it’s not hard to see them on the graph. These two major clusters on the account so
we got pro-white helmets in blue and anti-white helmets in red. Their activity is persistent over time. Many of the same accounts repeatedly tweeting
and retweeting comments. They’re literally drowning out of the voices
on the other side. The center account is the official account
of the white helmets they do get some retweets but their content over time is drowned out
by what is going on on the other side. Who are the accounts on the right? So the top influencers include self-described
journalists. They consistently produce content aligning
with Syrian and government interests. Some of them visited as guests of Assad in
Syria. There are also officials of the Syrian government. And there are few of what seem to be inauthentic
and likely sort of paid or organizationally controlled agents on that side. We’re not sure who to attribute them to but
we have a hunch that some of those accounts aren’t real. But they only make up a tiny percentage of
the conversation. The retweeters causing some of these other
accounts to be big are not trolls or bots. They’re actually sincere activists. We went looking for bots in this conversation
and except in a few rare cases we really don’t find them what. We see is actually a community of online activists. So most (video buffering). That renders them particularly vulnerable
to additional disinformation that aligns with the metanarratives of disinformation campaigns. Perhaps these are views of the long — of
long-term effects of pervasive disinformation which is something that does keep me up at
night because it’s frightening. All right, so considering these two together,
unit of analysis and that disinformation isn’t always false — or isn’t simply false. As we turn to questions of detection which
is important for research and especially online environments this means that identifying disinformation
isn’t about determining this about the post. But how that post and account fit into an
underlying campaign where the intent is to lead in a strategic purpose. This disinformation remains disinformation
regardless of who the intermediary is. A person does not need to be aware of their
role to be a participant in a campaign we want to address the spread of disinformation
because you can’t treat people and sincere activists who are taking up these messages
in the same way as a paid agent. How do we start to navigate these bases? In ways that are very complex and I do not
recommend taking action against activists. This gets to the point I highlighted in the
talk. A lot of this focus is on authentic activity. But disinformation is not just about bots
and trolls. As we see in our study, it targets, infiltrates,
cultivates, shapes and ultimately leverages online crowds and communities. Bittman talked about the role that unwitting
agents played in historical disinformation campaigns and here we can see agents working
with, collaborating with unwitting crowds to achieve their goals. I encourage researchers and others not just
to focus on the orange parts of the graph, of the IRA graph but to think about the blue
and the whole part of the other graph as well especially on the red to look at — to study
how online communities are targeted and affected, shaped by this information. How some members of these unwitting crowds
take and it make it their own. We’re seeing this as well in sort of alt-right
communities as well and not featuring them here but this is something that is happening
across our information spaces. Once an effort takes root inside an online
community the problem becomes much more difficult to address as freedom of speech concerns by
the platform replace the sort of bot and troll detection work so how do we start talking
about the difficulty of those conversations? Another point I think is really important
and in fact I think for researchers and those that talk about this, this is one of the things
we have to keep in mind is that disinformation doesn’t just affect some other person, somebody
who’s got mental illness or someone who’s not smart. These are not stereotypes that play out. Disinformation can target all of us and we’re
all in some ways vulnerable. For our researchers, we’ve experienced this,
we’ve been affected by disinformation. I was affected by the disinformation in the
content I studied around the conspiracy theories of crisis events. I was disoriented for months. It had the lead researcher for our team talked
about how hard it was to distinguish the content. He had a hard time seeing the content that
was pro-Black Lives Matter as propaganda. He says this looks like everyday messaging. In the white helmets conversation it took
our researchers weeks, in some case months, in one case I’m not sure this has been achieved
to recognize that what we were seeing was disinformation. We had a team of researchers reading this
content. A lot of it was people like us. It was disorienting and compelling. Things got easier when we began to see the
patterns of information sharing and the connection two Russia’s information apparatus. But we were still left with a lingering doubt
about the White Helmets and their mission which was the whole point. And though few of the folks here are Sandy
Hook deniers like some of the people I studied we’re constantly look for patterns and we
— this isn’t about intelligence or mental health or something else. If we’re all vulnerable to seeing the world
in these ways and all vulnerable to disinformation and we’re all being targeted especially in
our political identities when we’re in online spaces and other information spaces. So a few more things as I round this out and
move onto questions. So beyond a single campaign. Disinformation has consequences for societies
experiencing it. One is distrust in information. Disinformation reduces our ability to know
whom and what — whom or what to trust with the idea that when people don’t know where
to go for information that they can trust, they lose their agency. Their ability to take decisions based on knowledge
in the world. You can see this as both cause and effect
within the pervasive spread of conspiracy theories. This kind of lost of trust in the information
people see and they — and their explicitly reflecting and attacking information providers. The IRA trolls play to this distrust. Directly attacking mainstream media. This is a tweet from one of the trolls sent
in 2017 encouraging its audience to doubt media narratives and we see this CNN as ISIS
here. We see that same sort of hashtag spreading
in the anti-white helmeting conversation in our other case study. These same sort of memes are moving into these
different kinds of conversations which does show a connection at the very least between
the kinds of people that are participating in these information spaces. Interestingly this one came from a far right
account and in the anti-White Helmets we see it on the left, but the meaning is the same. And this has second order effects on trust
as well. It’s not just the fake news and main stream
media that we can’t trust as we’re losing trust. It’s about anything we encounter online, so
we go online and we’re mad at somebody and we’re like that must be a troll. I don’t believe them. I heard about these Russian trolls. I don’t have to consider what they have to
say because they’re lying to me or something goes on in the world. It’s not real. It’s the media lying to me again. When I saw the videos I think of the Notre
Dame cathedral on fire I thought it was a fake because we lose trust in what we see
and how we see it. In their insightful and frightening essay
on the menace of unreality it was described how the purpose of disinformation isn’t to
convince you of something it’s to confuse you. To create this muddled thinking across society
with this idea with a society that doesn’t know where to go for trusted information is
a society that’s easily manipulated and one that may turn to more authoritarian power
structures. This final slide that disinformation is actually
a threat to democratic societies directly. Because this lost of trust feeds into this
— into sort of a more pervasive issue. Disinformation destabilize s the common ground
that we need to stand on. It undermines our shared reality. We lose trust in information, in the confidence
that we know what we know. That things are as they seem. Notice the positioning of the orange clusters
here. On the outside of these communities. Helping to pull these communities further
apart. It’s empirical but also metaphorical. Disinformation tears at the social fabric,
amplifying existing divisions, and if the division becomes too extreme we can’t come
together. We have no common ground from which to govern
ourselves and, again, we’re seeing people start to turn towards more autocratic or authoritarian
leadership ideas in the wake of this so it’s a danger to democracy. All right, so I’d like to shift to talk about
solutions. I will introduce a couple slides, talk a little
bit and open up for questions because I think your questions will guide this a little better. I will, yeah, I’ll get to this. Okay. So this is a hard problem. And I told you all I’ve — I’m an optimistic
person and I’ve not been too optimistic for the last few years but I’m try ing to end
this talk on a happier note. I’m starting to feel a little more optimistic
and I have a story around this. A year ago I was here in D.C. at a meeting,
the Swedish embassy and I was on a panel and we were talking about all these problems and
how horrible it was and people kept asking maybe do this and I was like you can’t do
this. Oh, this will work and no, that’s too simple,
and it won’t work because of this and we shot down every idea that anyone in the crowd had
and we basically were like we’re doomed. And there was this ambassador in the back
of the room, it turns out, I just figured out who it was and I told him the story and
he was like that was me. Daniel Freed. And he came to the front of the room and he
scolded us and said how dare you get up here and do this to us. We’ve had hard problems before. We had the Cold War and we were at the — nuclear
war was like something that we were in fear of everyday. And we had hard problems and no one said,
oh, we can’t solve it. We chipped away at the problems. We worked at it from all these different sides
and eventually those problems were gone and now we have a whole new set of problems but
you can’t just give up. You have to, you know, you have to chip away
at these problems and I agree. We’re not going to solve it with one particular
thing. There’s not one digital media campaign that’s
going to fix it or one change in technology, Facebook would only do this. We’re going to solve the problem. Absolutely not. We need to chip away at this from all sides
that’s why it’s important to have a diverse group of researchers to look at how we build
technology that’s more resistant. How do we train and educate people to become
more resilient and sort of interact in these places with more — in more productive ways. How do we build trust in these information
systems? How do we build trust in journalism? How do we develop policies? Policies are coming. Again, I was back in D.C. earlier this week
and policy is coming for the technology companies whether they like it or not and they’re at
the table and there are policymakers. I think we need more people here to think
about what kinds of policies, technology companies should be held to as we’re trying to work
on these spaces. I think that’s one of the things I would really
suggest is for you to start engaging with policy if you’re not already. And to start thinking about in your papers
the implication for policy. I think it’s really important that we think
about how do we, you know, address some of these problems without giving up the things
we really care about in some of these platforms. How do we address disinformation without squashing
online activism and how do we support freedom of speech without supporting the weaponization
of speech I think that will be important there. Journalism I won’t feature here but I think
is a huge part of the solution as well. There’s good critique of media right now as
they try to adapt to new conditions and that have been disrupted because of technology
and I think we need to support journalists in different kinds of ways but they also have
some work to do to kind of adjust their practices in this new world to kind of start building
trust in their own products as well so I’m not going to totally let them off the hook
and the education side, we absolutely need new media literacy things. I don’t think media literacy alone is going
to work. But –. Work. It’s teaching us how to understand how we’re
being emotionally targeted in our political identities and kind of tune in to how our
own outrage at the other is being used against us as a society. And think about how we can be healthier and
more productive, and one thing I want to say is this is a K-99 problem and I think it’s
really important that we’re not just focusing on educational interventions at the younger
ages. The kids are going to be okay I think. They’re pretty savvy. They understand things. Definitely, you know, I’m not advocating education
— I am not saying we should educate them but I think we have to keep the world around
long enough for them to take over and I think people my age and over are very susceptible
to disinformation and we need to take a look at our own practices so I want to encourage
that as well. And I think my last slide is I want to leave you
with this as Ambassador Freed was encouraging me last year to chip away at
the problem. We’re not going to solve it with one little
thing but I think each of us has something to contribute from each perspective. The technology itself making it more secure
and encryption and all these things versus the psychology — it’s a psychology problem,
a sociology, engineering problem. It’s across the board. Whatever skills you have if you have the interest
to come into this space I think it’s important for us to be working together
and just making small, small, small gains are
going to be really important and that’s it. I’m going to say thank you. I want to say thank you to all my students and collaborators. I
had great students helping me with this and then, yeah. Open up for questions. (Applause).

Author: Kennedi Daugherty

Leave a Reply

Your email address will not be published. Required fields are marked *