Page 148 - ITU Journal - ICT Discoveries - Volume 1, No. 2, December 2018 - Second special issue on Data for Good
P. 148

,78 -2851$/  ,&7 'LVFRYHULHV  9RO        'HFHPEHU



          human  development.  Second,  there  is  no  need  to   systems  and  governance  standards  for  building  a
          encroach  on  privacy  for  such  a  system  to  work;   human AI. If successful, it will be expanded to other
          aggregated   anonymized     (strictly   speaking,    countries  and  industries.  OPAL and other cases
          “pseudonymized”) indicators suffice.                 point  to  the  fundamental  discursive  function  of
                                                               data,  and  to  the  importance  of  processes,  for
          The human AI is an aspirational analogy. It is a call   instilling positive social change. Setting up a project
          for  building  human  systems  where  facts  matter;   such as OPAL requires aligning incentives of large
          where the efficiency and relevance of policies and   organizations  around  a  common  objective;  this
          programs,  and  a  multitude  of  sociopolitical     process alone has many virtues.
          processes  and  outcomes,  can  be  assessed,
          discussed, and improved on the basis of data.        Another example of the value of processes that can
                                                               be  facilitated  by  data  and  algorithms  is  that  of  a
          5.   REQUIREMENTS AND PRIORITIES FOR                 controversy around changes to school bus routes in
               DESIGNING A HUMAN AI                            Boston, as recounted by Joi Ito, Director of the MIT
                                                               Media Lab [19]. Protests over changes to school bus
          What is required for a human AI? It will take several   routes in Boston ‘decided’ by an algorithm designed
          key  ingredients.  It  will  take  nurturing  a  strong,   by MIT researchers led to the (human) decision not
          healthy  data  culture,  including  widespread  data   to use it. What appeared in hindsight was that the
          literacy, with more trust and interest in evidence-  protesters  were  predominately  wealthier  families
          informed debates among the public. It will also take   who had ‘lost out’ as a result of the changes. In the
          building better public governance for the systems    words  of  Joi  Ito:”  “While I'm not sure privileged
          that provide the data that can power a human AI,     families would give up their good start times to help
          including private sector data systems, allowing key   poor families voluntarily, I think that if people had
          data to be tapped into safely and ethically.         understood what the algorithm was optimizing for—
                                                               sleep health of high school kids, getting elementary
          These are, among others, the key objectives of the   school kids home before dark, supporting kids with
          Open Algorithms (OPAL) project [18]. OPAL aims to    special needs, lowering costs, and increasing equity
          allow accredited users to query private sector data   overall—they would agree  that the new schedule
          through open algorithms running on the servers of    was, on the whole, better than the previous one.”
          partner private companies, behind their firewalls,
          to  extract  key  aggregated  indicators  of  interest,   What   lacked   there   was   basic   human
          from  cell-phone  activity,  bank  transactions,     communication.  This  looks  like  a  missed
          possibly  hospital  records,  police  data,  and  more.   opportunity  to  leverage  the  power  of  AI  for  the
          With OPAL, no sensitive data ever leaves the servers   common good as a result of human flaws. Discussing
          of  the  data  partner  organizations.  All  queries  are   complex social issues through the lens of data and
          logged, auditable; all algorithms are open, subject to   algorithms,  by  transparently  explaining  and
          scrutiny and redress.                                discussing the objectives, features, potential pitfalls,
                                                               of  various  algorithms,  could  change  people’s
          OPAL  also  aims  to  develop  governance standards   perceptions  and  attitudes  towards  fellow  citizens
          and processes that will allow data subjects to weigh   and social problems.
          in on the kinds of analyses done using data about
          themselves;  including  through  local  oversight    Whether or not empathy is what ultimately needs
          bodies referred to as Councils for the Orientation of   strengthening  as  the  core  human  ingredient  for
          Development  and  Ethics,  or  CODEs.  Sensitive  use   building  better  human  systems  is  a  matter  of
          cases are presented to the local CODEs, which may    debate.  It  would  seem  like  more  empathy  within
          determine  that  a  specific  indicator,  for  example,   and between groups could curb behaviors that hurt
          population  density  estimates,  should  not  be     others, and lead to better overall social outcomes.
          provided  beyond  a  certain  level  of  temporal  and   This  sounds  consistent  with  and  conducive  to  a
          geographic granularity for security reasons.         human  AI  that  would  “reward  and  reinforce
                                                               decisions  and  actions  of  its  members  that
          Currently piloted in Colombia and Senegal with two   contribute  to  the  common  good,  through  and
          leading  telecommunication  operators  and  their    feeding candid fact-based discussions between its
          national  statistical  offices,  OPAL  is  the  first  ever   members.”
          real-world  attempt  at  setting  up  technological




                                             ‹ ,QWHUQDWLRQDO 7HOHFRPPXQLFDWLRQ 8QLRQ
   143   144   145   146   147   148   149   150   151   152   153