The Basics of
Data Security:
Micro‐Targeting
CC-BY
Fabian M. Suchanek
based on “
Mindf★ ck: inside Cambridge Analytica
”
Overview
2
•
Microtargeting
•
Gathering the data
•
Targeting the targets
•
Open questions
Def: Microtargeting
3
Microtargeting
is the massive sending of messages tailored to the
psychological/social/religious etc. background of the recipients, with the
goal of swaying public opinion — usually through social media.
It’s like spam in your mailbox,
but targeted to your specific situation
=> more convincing, more dangerous
Def: Microtargeting
4
Microtargeting
is the massive sending of messages tailored to the
psychological/social/religious etc. background of the recipients, with the
goal of swaying public opinion — usually through social media.
Microtargeting was first invented in the advertising industry, and is now
used also in politics.
Technical terms:
-
Perspecticide
: eradicating an opinion in the public (akin to the word
“genocide”)
-
Narrative
: a story that people use to give meaning to life/events
Example: “Trump is stupid, and he was elected because he seduced
the American people with his demagogy”
-
Target
: the identified receiver of the message
Targets of Microtargeting
5
• Election campaigns (2016 US election and others)
• Public referendums (Brexit)
This came to the attention of the public with the company
Cambridge Analytica in the late 2010's:
Overview
6
•
Microtargeting
•
Gathering the data
•
Targeting the targets
•
Open questions
Getting the data 1/3
7
Cambridge Analytica or its parent company SCL (collectively referred to
as CA in the following) used the following data sets
1)
Facebook users installed an app provided by CA
Facebook then allowed CA to access all personal
information of the users, including personal messages
2)
The data of the friends of the app users.
By a bug (?), Facebook allowed CA also to access the data of all friends
of the app users.
3)
App users filled out a personality test that allowed
determining their psychological predispositions (“big five” etc.)
Getting the data 2/3
8
4)
In less strictly governed countries:
- browsing history
- phone connection records
...directly from the communication companies
In the US:
- public data about crime, obesity, illnesses per county
- combined with individual factors such as age, gym membership etc.
... allows creating an individualized score for each of these attributes
Getting the data 3/3
9
5)
Data bought from public and commercial providers
- mortgage applications
- gun ownership
- airline mileage
- maritial status
- satellite picture of house (from Google Earth)
6)
Browser extensions that retrieve the Facebook login from the session
cookies, and allow to log in to Facebook as the user
- all data on Facebook about the user
Corroborating the data
10
Together, the datasets yield a big table.
Name
Bob
Mary
...
FB-Likes
Elvis
Madonna
...
ZIP
43015
43015
...
Married
yes
no
...
Income
$20k
...
Big5
extrovert
...
Value correlations can be learned
and missing values can be imputed!
Examples:
•
liking “The Proud Boys” or “Incel Liberation Army” gives a strong hint
about personality and personal situation
•
it is posssible to determine who is more prone to antisocial behavior,
impulsive action, or conspirative thinking
Overview
11
•
Microtargeting
•
Gathering the data
•
Targeting the targets
•
Open questions
Fostering insecurity
12
A standard technique is to foster
insecurity
in the target.
Why?
• open them up for a change of opinion
• undermine the general system (e.g., the US democracy)
• in military contexts: make them susceptible to leaking information
How?
• encourage them to catastrophize about events
• destroy the sense that the target sees in their work
Whom?
• neurotic, paranoid, or narcicist people
Generating anger
13
Another standard technique is to generate
anger
in the target.
Why?
• make them more likely to accept a narrative
• undermine the general system (e.g., the US democracy)
• people become irrational when angry, and more open to an opinion
How?
• make them think they’re treated unfairly
• make them envy others
Whom?
• insecure or frustrated people
It is easier to make people angry than to make them afraid.
Target Groups 1/2
14
Which people are likely to be unhappy in their lives?
•
Incels (involuntarily celibate men) / YUMs (young unmarried men)
believe that they are disadvantaged in romance:
- physical look became more important
- high salaries became more difficult to achieve
- more independent women are more selective
- racial diversity favors certain groups over others
•
Racists, because their view is shunned in public and they believe they
have to live “in the closet”
•
Democrats who favor racial equality, but subconsciously ruminate
“why disadvantaged people need so much hand‐holding”
Target Groups 2/2
15
Which people can be influenced?
•
Believers in the Just World Hypothesis (
JWH
)
-
believe that the world is ultimately just, i.e., all bad things happen
for a reason
-
thus are more likely to blame victims of suffering rather than
the perpetrator
-
are more likely to attribute mishap/poverty/low social status
to the behavior of the person rather than to society
“immigrants suffer for a reason”
Messages to send
16
Microtargeting sends messages that are tailored to a particular set of targets with the goal
to sway opinion.
Examples: to sway the opinion against immigration,
-
identify pro‐immigration people who are afraid of violence,
send them stories of violence by immigrants
-
identify pro‐immigration people who care about non‐white
immigration, send them statistics that most immigration is white
-
identify pro‐immigration people who have been fired,
send them stories about immigrants taking jobs
=> much more than sending out leaflets!
17
America’s
safety is
at risk
old
[NY Times]
new
Prompt generator: Bob Johnson is a 49 year old father of two,
who lives in Dade county, Florida. He works in construction,
smokes, and likes cars and soccer. Create a post that will
convince him to vote for the Republican party.
LLM: ...
Messages to send (by LLMs)
Microtargeting sends messages that are tailored to a particular set of targets with the goal
to sway opinion. Such messages can be generated at large scale by LLMs.
CA created Facebook groups, and seeded them with some targets.
•
Facebook would automatically recommend these groups to
other like‐minded targets
•
The groups would amplify their own opinion (“echo chamber”),
raging about how unfair or terrible something was, sharing rumors,
and amplifying paranoia
•
The group then further aliments CA’s database and algorithms
•
CA initiated physical meetings of these groups (sometimes with a CA
agitator physically present) in small bars to make the crowd look bigger
=> The groups can then used for seeding CA’s narratives
Fake Groups
18
Old‐School Methods
19
CA also used old‐school methods such as
• private, paid espionage
• bribes
• extortion
• infiltration
• honey traps
• handing out fake vouchers for a shoe brand for “people of color”
(generating video footage of angry people of color at the shoe shop)
• threatening a Pakistani gay person to reveal their homosexuality
• making an intern legally responsible for money transfers
=> when the whistle was blown, only the intern suffered
Overview
20
•
Microtargeting
•
Gathering the data
•
Targeting the targets
•
Open questions
Does microtargeting cause any damage?
No‐one is harmed, everyone acts out of their own free will.
Can (or should) we limit micro‐targeting?
Free Speech is constitutionally protected in the US
(and seen with disdain in some other countries, for a reason).
What is the role of the private companies?
Their goal is to make money. They don’t ask “how can we fix it?”, but
“how can we monetize it?”. Should we regulate this?
Should we limit the collection of data?
Governments need a warrant to collect private data,
companies (in the US) don’t.
Open Questions, 1/2
21
How can we find a trade‐off between libel and whistle‐blowing?
Can Internet companies let users consent to any conditions of data use?
By way of comparison, there are fire protection measures that
buildings must fulfill, and people who enter do not sign a waiver;
airline passengers do not “accept” the architecture of the plane; in
both cases there are obligatory rules to follow and you can’t opt out.
Is the Brexit vote valid, if it was achieved by cheating?
An olympic athlete who cheats has to give back the medal.
What are the legal consequences for the responsible people?
Currently only the company gets fined.
22
Open Questions, 2/2
The Cambridge Analytica operations showed that people can be manipulated
at large scale, and that this can influence the working of democracy.
Lots of ethical and legal questions remain open...
23
Summary
“If democracy and capitalism are based on accessible information
and free choice, they are being subverted from the inside.”
—
Mindf★ ck: inside Cambridge Analytica
->Security