algorithmic accountability: computational journalism week 11

15
Fron%ers of Computa%onal Journalism Columbia Journalism School Week 11: Algorithmic Accountability November 21, 2014

Upload: jonathan-stray

Post on 24-Dec-2015

11 views

Category:

Documents


1 download

DESCRIPTION

Columbia University, Fall 2014 Syllabus at http://www.compjournalism.com/?p=113

TRANSCRIPT

Fron%ers  of    Computa%onal  Journalism  

Columbia  Journalism  School    

Week  11:  Algorithmic  Accountability  November  21,  2014  

     

What  types  of  algorithms  can  have  large  consequences  for  someone's  life?  

Algorithms  in  our  lives  

•  Personalized  search  •  Poli%cal  microtarge%ng  •  Credit  score  /  loans  /  insurance  •  Predic%ve  policing  •  Price  discrimina%on  •  Algorithmic  trading  /  markets  •  Terrorist  threat  predic%on  •  Hiring  models  

From  myfico.com  

Predicted  crime  %mes  and  loca%ons  in  the  PredPol  system.    

All  The  Stops,  Thomas  Rhiel,  Bklynr.com,  2012  

Title  VII  of  Civil  Rights  Act,  1964  It  shall  be  an  unlawful  employment  prac%ce  for  an  employer  -­‐    (1)  to  fail  or  refuse  to  hire  or  to  discharge  any  individual,  or  otherwise  to  discriminate  against  any  individual  with  respect  to  his  compensa%on,  terms,  condi%ons,  or  privileges  of  employment,  because  of  such  individual’s  race,  color,  religion,  sex,  or  na%onal  origin;  or    (2)  to  limit,  segregate,  or  classify  his  employees  or  applicants  for  employment  in  any  way  which  would  deprive  or  tend  to  deprive  any  individual  of  employment  opportuni%es  or  otherwise  adversely  affect  his  status  as  an  employee,  because  of  such  individual’s  race,  color,  religion,  sex,  or  na%onal  origin.  

Learning  from  Facebook  likes  

From  Kosinski  et.  al.,  Private  traits  and  a4ributes  are  predictable  from  digital  records  of  human  behavior    

Predic%ng  gender  from  Twi_er  

From  Zamal  et.  al.,  Homophily  and  Latent  A4ribute  Inference:  Inferring  Latent  A4ributes  of  Twi4er  Users  from  Neighbors  

Predic%ng  race  from  Twi_er  

From  Pennacchioa  and  Popescu,  A  Machine  Learning  Approach  to  Twi4er  User  ClassificaFon  

Even  if  two  groups  of  the  popula%on  admit  simple  classifiers,  the  whole  popula%on  may  not(from  How  Big  Data  is  Unfair)  

Websites  Vary  Prices,  Deals  Based  on  Users'  InformaFon  Valen%no-­‐Devries,  Singer-­‐Vine  and  Soltani,  WSJ,  2012  

Message  Machine  Jeff  Larson,  Al  Shaw,  ProPublica,  2012