confessions (and lessons) of a "recovering" data broker
DESCRIPTION
Jim AdlerTRANSCRIPT
Confessions (and Lessons) of a "Recovering" Data Broker: Responsible Innovation in the Age of Big Data, Big Brother, and the Coming Skynet Terminators
Jim AdlerVice President, ProductsMetanautix
[email protected]@jim_adlerhttp://jimadler.me
Usenix Security ConferenceWashington, DC
Aug 15 2013
About Metanautix
• Mission: Next generation big data management & analysis– Transparency into the big data supply chain within and across
organizations
• Engineering led, product company– Team of veterans (10-20+ years experience)– Built massive data analysis systems at Google, FB, MSFT, AMZN, Pixar– Solidly funded by top VCs– Working with lighthouse customers
• Started Nov 2012, slowly emerging from stealth mode
• Stay tuned!
I am not an attorney.Confession
Intelligence
Social IneptitudeObsession Dork
NerdGeek Dweeb
You can often do more from the inside than the outside.
• 20B public records
• 30M visitors per month
• 50M+ reports sold
Confession
Listen to your toughest criticsLesson
Can I have a little narcissism with my voyeurism?
• What does my background check say?
• Privacy controls– Suppress single address or
phone number
• Comment on your own public profile
Lesson
Data from Feb 2013
Data from 40,000 BC thru 2002
20Exabytes
(20M TB)
Lots of data created, transferred, & stored.Confession
The “Public” Data Supply Chain of YouConfession
Search
Blogs
Criminal Records
Risk
Civil Suits Marketing
Addresses
Directory
Phone Numbers
Payments
Background
Resumes
Public Posts Name
s
Data Uses
Government
Commercial
Self-Reported
Big Data
Engines
FIND OWNER OF DOG’S RELATIVE FOR TRANSPLANT
SINGLES CURIOUS ABOUT THE PEOPLE THEY MEET
PARENTS ENSURING WHO THEIR KIDS
SAFETY
GENEALOGISTS CULTIVATING THEIR
FAMILY TREE
BUSINESSES THAT NEED TO UPDATE CONTACT INFORMATION ON
CUSTOMERS
FINDING LONG-LOST FRIENDS, MILITARY BUDDIES, ROOMMATES, OR
CLASSMATES
ANYONE CURIOUS ABOUT WHO'S EMAILING OR CALLING THEM
THOSE IN LEGALLY ENTANGLED LOOKING FOR COURT RECORDS ANYONE WHO NEED ADDRESS
HISTORIES FOR PASSPORTS
SOCIAL NETWORKERS LOOKING TO EXPAND THEIR FRIENDS LIST PROFESSIONALS LEARNING ABOUT
COLLEAGUES AT CONFERENCES
RECONNECTING OUT-OF-TOUCH FAMILY MEMBERS
ONLINE SHOPPERS VERIFYING ONLINE
SELLERS
INVESTIGATIVE JOURNALISTS RUNNING
DOWN LEADSSALES PROFESSIONALS
LOOKING FOR NEW PROSPECTS
NETWORKERS SEEKING BUSINESS OPPORTUNITIES
LAW ENFORCEME
NT
NON-PROFIT ORGANIZATIONS LOOKING
FOR SUPPORTERS
FIANCÉS AND THEIR CURIOUS FAMILY
MEMBERS
SOCIAL WORKERS WHO NEED TO KNOW MORE ABOUT THEIR CLIENTS
CALLER ID OF HARASSING PHONE
CALLS
ALUMNI GROUPS ARRANGING REUNIONS
ADOPTED KIDS SEEKING THEIR BIOLOGICAL
PARENTS
AIRLINES TRYING TO RETURN LOST
LUGGAGE
CHECKING OUT A PROSPECTIVE SOCIAL NETWORK CONNECTION
CHECKING OUT A PROSPECTIVE DATE
CHECKING OUT A PROSPECTIVE
TENANT
FINDING PEOPLE THAT HAVE THE SAME ILLNESS AS
YOU
RESEARCHING A PROSPECTIVE
EMPLOYEE
ANYONE RETRIEVING
COURT RECORDS
REGULATED
Lots of uses for your data … some regulated
LAWYERS NEEDING QUICK ACCESS TO COURT RECORDS
BANKING SERVICES
RESEARCH
SHARING
LEARNING ABOUT A BUSINESS
Confession
Billions of RecordsMillions of People
Jim AdlerHouston, TX
Age 70
Jim AdlerRedmond, WA
Age 50Jim Adler
Denver, COAge 48
Jim AdlerMcKinney, TX
Age 57
Jim AdlerCanaan, NH
Age 59
Jim AdlerHastings, NE
Age 32
213 records linkedto the correct 37 Jim Adlers
Philip Collins
375 PeopleJim Adler
213 Records37 People
Randolph Hutchins5 People
Gwen Fleming2 PeopleCarol Brooks
9800 Records1250 People
Confession We don’t know you all that well
Opt-out doesn’t always mean deletionConfession
Jane Hampton 06/23/1998 123 Main Peoria, IL
Jane Hampton 123 Main Peoria, IL [email protected]
Jane Hampton [email protected]
Jane Hampton 06/23/1998 123 Main Peoria, IL
Jane Hampton 123 Main Peoria, IL [email protected]
Jane Hampton [email protected]
When towns were small, personal anonymity was low …
“The only thing worse than being talked about, is not being talked about.”
− Oscar Wilde
Lesson
“Good Fences Make Good Neighbors”
− Robert Frost
Urban populations grew along with personal anonymity…
Lesson
1850 1870 1890 1910 1930 1950 1970 1990 2010 20300
20
40
60
80
100
120
Pri
vacy E
xp
ecta
tion
s
“Rockwell” Era “Good Fences” Era “PrivacyVertigo”
Era O
nline Density ↓
Personal Anonymity ↓
… we’re suffering from Privacy Vertigo.Confession
Urban Density ↓Personal Anonymity ↓
Urban Densit
y ↑
Personal A
nonymity
↑
Privacy is deeply cultural.Lesson
Discretion
Disclosure
EU “Right of Personality”
• Inviolable right of dignity
• Germany: BGB, 1900, post WWII US: Brandeis & Warren, 1890
• Germany: “Source Right”
• Esra “kiss & tell” plaintiff won, book banned
US “Privacy Torts”
• Statutory harm
• US: Prosser’s Torts, 1960
• US: Sectorial privacy regime
• Bonome “kiss & tell” case dismissed
EU Rights versus US Torts
http://paulschwartz.net/pdf/SchwartzPeifer_Prosser_FINAL.pdf
How to unpack Privacy? Think PPP.
PRIVACY
PERILS
PLAC
ESPLAYERS
Lesson
Sometimes you’re in a public place when you think you’re in a private place.
“Gaydar”
A 2009 MIT study found it was possible to predict men’s sexual orientation by analyzing the gender and sexuality of their social network contacts – even if the rest of the information on their profile was set to private.
Confession
“To Serve Man” is a Cookbook.Confession
“If you’re not paying for the product, you are the product.”
− Claire Wolfe (paraphrased)
Peer to Peer
Corporati
on & Custo
mer/Employe
e
Govern
ment & Citize
n
Your God &
You
Power Disparity
Pri
vacy
Rig
hts
In privacy contexts, Power matters.Lesson
Target knows you’re pregnant and when you’re due. So, what’s so perilous?
Confession
Secrets are power equalizers.Confession
Public PlacesPowerful Players
Private PlacesPowerful Players
Private PlacesWeaker Players
Public PlacesWeaker Players
M O R E P R I V A T E P L A C E S
MO
RE
PL
AY
ER
PO
WE
R G
AP
Mapping Places-Players-Perils Cases
M O R E P R I V A T E P L A C E S
MO
RE
PL
AY
ER
PO
WE
R G
AP
Places-Players-Perils Cases
A head in the clouds < 20 yearsPrediction
2012 2014 2016 2018 2020 2022 2024 2026 2028 2030 2032 2034 2036 2038 2040 2042$1
$10
$100
$1,000
$10,000
$100,000
$27,100
$13,500
$6,800
$3,400
$1,700
$850
$420
$210
$100
$53
$26
$13
$7
$3$2
$1
Year
Cost
per
Mon
th (
000s
) • 20,000 TFlops• 2,500 Terabytes• Less than $700K per year
Chris Westbury, University of Alberta
“Watch your thoughts, they become words.Watch your words, they become actions.Watch your actions, they become habits.Watch your habits, they become your character.Watch your character, it becomes your destiny.”
– Lao Tzu
“… the essential crime that contained all others in itself. Thoughtcrime, they called it.”
– George Orwell
Big data inferences are not thoughtcrimesLesson
Felon Classifier Overview
ModelLearner250 M Defendants Feature
Extraction
15K Labels
15K PredictorsClea
ning
Link
ing
Sam
plin
g
ObjectiveIf someone has minor offenses on their criminal
record, do they also have any felonies?
Today’s Bloomberg piece: http://bloom.bg/1eMtnug
How does the Felon Classifier work?
Gender Eye Color Tattoos Criminal Offenses Score
Over Threshold
of 3.5?Likely Felon?
NO
YES
Hazel (+1.7)
Male Blue 2 + (+1.3) Traffic only
Female (-0.5) Brown < 2 4 or fewermisdemeanors (+1.9)
YES
Green 8 or fewermisdemeanors
NO
4.4
Hazel
Male (+0.1) Blue 2 + Traffic only (-0.5)
Female Brown (+1.2) < 2 (+0.1) 4 or fewermisdemeanors
YES
Green 8 or fewermisdemeanors
NO
0.9
widget: http://jimadler.me
Classifiers depend on policy as much as technology
AN
AR
CH
Y
T Y R A N N Y
0.0% 5.0% 10.0% 15.0% 20.0%0.0%
20.0%
40.0%
60.0%
80.0%
100.0%
False Positive Rate
Fals
e N
egat
ive
Rat
e
Threshold: 0.66FP Rate: 5% FN Rate: 22%
Threshold: 1.1FP Rate: 1% FN Rate: 40%
Threshold: -1.82FP Rate: 19% FN Rate: 0%
Confession
Privacy, Reasonable Suspicion, & Probable Cause
• Courts have upheld profiling– US v Sokolow
• Predictive information never enough1. Reliable 2. Efficient3. Particularized4. Detailed5. Timely6. Corroborated• More• Andrew Guthrie Ferguson, Predictive Policing
http://ssrn.com/abstract_id=2050001 • Bernard Harcourt, Rethinking Racial Profiling
http://www.law.uchicago.edu/files/files/rethinking_racial_profiling.pdf• The “Not Ready For Prime Time” Classifier
http://jimadler.me/post/47374264398/the-not-ready-for-prime-time-classifier
NYC Stop & Frisk Found UnconstitutionalRuling
“The city … believes that blacks and Hispanics should be stopped at the same rate as their proportion of the local criminal suspect population.”
− US District Judge Shira Scheindlin
100% greater chance that a minority stop is justified over a random stop
NYC assumes 87%
Minorities in NYC
If it’s not ok to stop & frisk 95% of the general population for nothing,why is it ok to stop & frisk 90% of minorities for nothing?
PA|M = Chance a minority should be arrestedP~A|M = Chance a minority should not be arrestedPM|A = Chance that someone arrested is a minorityPM = Chance someone is a minorityPA = Chance someone is arrested
Bayes’ Rule
M O R E P R I V A T E P L A C E S
MO
RE
PL
AY
ER
PO
WE
R G
AP
Big brother is watching (duh)
“We’re being asked to trust without being able to verify.”
− Alex Howard
Pres. Obama calls for more transparency in FISA court and surveillance laws
NSA chief announces plan to replace 1,000 sysadmins with machines
Confession
Technology grows exponentially. Wisdom grows linearly.
• Gov’t doesn’t trust people (at least sysadmins) but does trust machines
• Little Transparency
• Wisdom is hard to come by
• Sentient (?) brain in the cloud in < 20 years
Lesson
Wisdom
Knowledge
Information
Data
Hilary Mason’s Maxim
Math + Code = Awesome
QuantsMaking a killing on Wall Street but still can’t impress the chicks
Weakonomics.com
Lesson
Corollary to Mason’s Maxim
Values * (Math + Code) = Awesome
Lesson
Norio OhgaSony President
74 min CD
Jeff JonasBig Data
Privacy by DesignMark Zuckerberg
Real Names
Steve Jobs‘nuff said
Eclectic generalists drive innovation
Richard FeynmanPhysics
Lesson
We’re making this up as we go.
Austin Alleman @allemanau
Innocence Frontier Regulation Innovation
Confession
“Can’t we all just get along?”Plea
Geeks
WonksSuits
SocialEntrepreneur
High-TechMercenary
ResponsibleInnovator
TraditionalCapitalist
− Rodney King
Lesson “No one here gets out alive.”− Jim Morrison
Adapt, Invent,
Test
Question, Scrutinize,
Criticize
Watch, Listen, Learn Geeks
WonksSuits