BIG DATA: BEHAVIOR IN BEHAVIOR OUT Summerschool - Big Data In Clinical Medicine Grolsch Veste, June 30, 2014 Johnny Hartz Søraker Assistant Professor Dept. of Philosophy University of Twente j.h.soraker@utwente.nl
BEHAVIOR IN BEHAVIOR OUT THE REAL ETHICAL PROBLEMS COME BEFORE AND AFTER BIG DATA INDIVIDUAL GROUP INDIVIDUAL Behavior determining data Data determining profile Profile determining behavior Big Data affects us as individuals, both before and after the actual processing of the data. It primarily affects our freedom to decide our own behavior, and our freedom to construct our own identities. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 2
TARGETING PREGNANCY EFFECTS BEFORE AND AFTER INDIVIDUAL GROUP INDIVIDUAL Ethical problem: How to make private companies and governments use big data responsibly? Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 3
TARGETING PREGNANCY EFFECTS BEFORE AND AFTER INDIVIDUAL GROUP INDIVIDUAL Ethical problem: How to make private companies and governments use big data responsibly? Ethical problem: Do decisions based on big data affect us in undesirable ways? Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 4
TARGETING PREGNANCY EFFECTS BEFORE AND AFTER INDIVIDUAL GROUP INDIVIDUAL Ethical problem: Do big data algorithms change our behavior in undesirable ways? Ethical problem: How to make private companies and governments use big data responsibly? Ethical problem: Do decisions based on big data affect us in undesirable ways? Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 5
FROM RUSSIA WITH LOVE - JEREMY AND SAMUEL BENTHAM S INSPECTION HOUSE Samuel Bentham Jeremy Bentham Grigory Potemkin Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 6
THE INVISIBLE GAZE MICHEL FOUCAULT ON THE PANOPTICON The major effect of the Panopticon: to induce in the inmate a state of conscious and permanent visibility that assures the automatic functioning of power [so] that the perfection of power should tend to render its actual exercise unnecessary [...] To achieve this, it is at once too much and too little that the prisoner should be constantly observed by an inspector: too little, for what matters is that he knows himself to be observed; too much, because he has no need in fact of being so. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 7
WHY PRIVACY? NOTHING TO HIDE = NOTHING TO FEAR? The Panopticon effect: The mere belief that you are being watched changes behavior dramatically, towards that which is considered socially acceptable. Big data is increasingly used for consumer recommendations based on behavior of you and your network, which can give rise to Group polarization: When not subjected to opposite views and tastes, your existing views and tastes become more entrenched and less nuanced. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 8
YOU MAY ALSO LIKE BIG DATA AND RECOMMENDATIONS Algorithms determine what you listen to, what you read, what you watch, whom you connect with, what you purchase, what you should vote, Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 9
YOU MAY ALSO LIKE BIG DATA AND RECOMMENDATIONS Politicians determine what you listen to, what you read, what you watch, whom you connect with, what you purchase, what you should vote, Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 10
YOU MAY ALSO LIKE BIG DATA AND RECOMMENDATIONS Algorithms determine what you listen to, what you read, what you watch, whom you connect with, what you purchase, what you should vote, Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 11
YOU MAY ALSO LIKE BIG DATA AND RECOMMENDATIONS Politicians determine what you listen to, what you read, what you watch, whom you connect with, what you purchase, what you should vote, Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 12
WHY PRIVACY? NOTHING TO HIDE = NOTHING TO FEAR? The Panopticon effect: The mere belief that you are being watched changes behavior dramatically, towards that which is considered socially acceptable. Big data is increasingly used for recommendations ( you may also like this ) based on behavior of you and your network, which can give rise to Group polarization: When not subjected to opposite views and tastes, your existing views and tastes become more entrenched and less nuanced. Diversity depends on freedom of choice and behavior. Diversity is necessary for innovation, democracy, and flourishing relationships We need the freaks and geeks! Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 13
Name: Johnny Hartz Søraker Address: Mooienhof 14, 7511EC Occupation: Philosopher Education: MA and PhD, philosophy Born: 7/11/1987 Ethnicity: Norwegian Loan purpose: Company startup, sector 134-9 Prior bankruptcies: [none] Databases used: Bankruptcies in sector 134-9, 2005-2012 Bankruptcies by age, education and gender Income statistics: gender and ethnicity Education and Income, 2001-2006 Recommendation: Score: 0.78 Loan ACCEPTED Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 14
Name: Johnny Hartz Søraker Address: Mooienhof 14, 7511EC Occupation: Philosopher Education: MA and PhD, philosophy Born: 7/11/1987 Ethnicity: Norwegian Loan purpose: Company startup, sector 134-9 Prior bankruptcies: [none] Databases used: Bankruptcies in sector 134-9, 2005-2012 Bankruptcies by age, education and gender Income statistics: gender and ethnicity Education and Income, 2001-2006 Facebook connect (music likes) Recommendation: Score: 0.71 Loan DENIED Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 15
Name: Johnny Hartz Søraker Address: Mooienhof 14, 7511EC Occupation: Philosopher Education: MA and PhD, philosophy Born: 7/11/1987 Ethnicity: Norwegian Loan purpose: Company startup, sector 134-9 Prior bankruptcies: [none] Databases used: Bankruptcies in sector 134-9, 2005-2012 Bankruptcies by age, education and gender Income statistics: gender and ethnicity Education and Income, 2001-2006 Purchase history Recommendation: Score: 0.84 Loan ACCEPTED Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 16
Name: Johnny Hartz Søraker Address: Mooienhof 14, 7511EC Occupation: Philosopher Education: MA and PhD, philosophy Born: 7/11/1987 Ethnicity: Norwegian African-American Loan purpose: Company startup, sector 134-9 Prior bankruptcies: [none] Databases used: Bankruptcies in sector 134-9, 2005-1014 Bankruptcies by age, education and gender Income statistics: gender and ethnicity Education and Income, 2001-2006 Recommendation: Score: 0.48 Loan DENIED Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 17
A FUTURE OF COMPUTER SAYS NO BIAS IN BIAS OUT Decisions can be made on basis of which big data pattern you belong to, created from merging separately (and voluntarily) collected information, often combined with statistical population data and other big data. There may be several biases against "your" group: - Spurious, outdated correlations (e.g. mobile vs. home phone) - Dubious inferences (e.g. several jobs=lack of dedication) - Discrimination fed into data (e.g. workplace discrimination) - Complete unknowns (second hand orange cars less defective) Many of these biases would be discriminatory, even outright racist, if performed by human, making it an advantage for the company that the algorithms are not transparent. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 18
COMPUTER SAYS NO IN CLINICAL MEDICINE BIAS IN BIAS OUT Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 19
CAPITALIZING ON BIG DATA THE MARKETPLACE DICTATES Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 20
COMPUTER SAYS NO IN CLINICAL MEDICINE BIAS IN BIAS OUT Patients may become obsessive in conforming to expected behaviour. This may very well improve their physical health, but what about their mental well-being? Subjective perception of own health determines well-being more than objective state of health. Big data may exacerbate problems related to pre-diagnostics, geneticism ( genes for ), and DIY healthcare In the future, we may not only be scared by our own symptom searches, but also by automated reports, made worse by poor understanding of statistics (5% risk of having cancer?) Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 21
COMPUTER SAYS NO IN CLINICAL MEDICINE BIAS IN BIAS OUT Under certain regimes (e.g. Obamacare) and health insurances, prediction record is valued far beyond treatment record; re-admittance needs to be avoided. Big data may increasingly replace causation with correlation in medical research. As with all complex data/neural networking, the better we get at predicting, the worse we may get at understanding. Existing practices entrenched in big data may uphold discrimination, e.g. black Americans receiving less health care than white Americans on basis of correlation alone, black spending habits become associated with unhealthy lifestyle. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 22
WHAT CAN WE DO? 1. Social Scientists, statisticians etc: Do more research on algorithms that can find and adjust biases in big data. 2. Individuals: We are the 99%, our behavior is the 99% (cf. ING) 3. Big data harvesters (researchers): Be very careful, especially when releasing open data. Safeguard privacy through architecture rather than policies. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 23
Privacy stages WHAT CAN WE DO? ANONYMIZE YOUR RESEARCH DATA identifiability Approach Linkability System Characteristics 0 1 identified privacy by policy (notice and choice) 2 pseudonymous privacy by architecture linked linkable with reasonable & automatable effort not linkable with reasonable effort 3 anonymous unlinkable unique identifiers across databases contact information stored with profile information no unique identifiers across databases common attributes across databases contact information stored separately from profile or transaction information no unique identifiers across databases no common attributes across databases random identifiers contact information stored separately from profile or transaction information collection of long term person characteristics on a low level of granularity technically enforced deletion of profile details at regular intervals no collection of contact information no collection of long term person characteristics Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 24 k-anonymity with large value of k
WHAT CAN WE DO? 1. Social Scientists, statisticians etc: Do more research on algorithms that can find and adjust biases in big data. 2. Individuals: We are the 99%, our behavior is the 99% (cf. ING) 3. Big data harvesters: Be very careful, especially when releasing open data. Safeguard privacy with architecture rather than policies. 4. Researchers: Use big data correlations as starting point for causation studies: 5. Clinical Practitioners: Understand the psychological effects on patients. Summerschool - Big Data In Clinical Medicine - j.h.soraker@utwente.nl June 30, 2014 25
COMMENTS, QUESTIONS? J.H.SORAKER@UTWENTE.NL TWITTER.COM/METUS Johnny Hartz Søraker Assistant Professor Dept. of Philosophy University of Twente j.h.soraker@utwente.nl