Democra(zing  Data  Science  

Sophie  Chou  @mpe(tchou  

William  Li  @williampli  

Ramesh  Sridharan  @tweetsbyramesh  

{soph,wpli,rameshvs}@mit.edu  

Some  Links  •  Paper:  bit.ly/DDSpaper  

•  Cathy  O’Neil’s  Blog  (@mathbabedotorg):  bit.ly/DDSblog  

•  TwiMer:  @mpe3tchou,  @williampli,  @tweetsbyramesh  

2  

[insert technology]  for  Social  Good  

Technology  is  a  force  mul(plier,    for  beMer  or  worse  

3  

What  is  Data  Science?  •  Our  working  defini(on:  transforming  data  into  insights/solu(ons/products  1.  collec(on  &  storage  2.  cleaning  &  structuring  3.  analyzing  &  finding  paMerns  4.  visualizing  &  communica(ng  results  

4  

What  is  “Democra(zing  Data  Science”?  

The  applica(on  of  data  science  is  undemocra(c:    problems  that  promote  the  common  good  

receive  insufficient  aMen(on.  

5  

Ford/MacArthur  Founda(on,  2013    “Technology  talent  is  a  key  need  in  

government  and  civil  society,    but  the  current  state  of  the  pipeline  is  

inadequate  to  meet  that  need.”      Source:  hMp://bit.ly/FordMacArthurReport  

6  

Why?  

misalloca3on  of  data  science  

lack  of  diversity  in  data  science  

incen(ves  for  data  scien(sts  

data  science  educa(on  

7  

Outline  •  Incen(ves  for  data  scien(sts  

•  Democra(zing  data  science  educa(on  

•  Poten(al  solu(ons  

8  

Sources  of  Power  in  Data  Science  

capital   data  

people  9  

Human  exper(se      “...[salaries]  between  $200,000  and  $300,000  a  year…100  recruiter  emails  a  day”    “...working  for  a  consumer  Internet  firm  can  be  surprisingly  rewarding.”  

Source:  hMp://bit.ly/WSJDataScience  10  

Sources  of  Power  in  Data  Science  

capital   data  

people  11  

Outline  •  Structural  inequali(es  in  data  science  

•  Democra(zing  data  science  educa(on  

•  Poten(al  solu(ons  

12  

Nega(ve  feedback  loop  Lack  of  

diversity  in  data  scien(sts  

Lack  of  diversity  in  educa(on  

Lack  of  diversity  in  applica(on  

13  

Why  care?  •  Diversity  is  key  to  innova(on  (Forbes  Insights,  2011)  

•  Lack  of  diversity  perpetuates  misalloca(on  

14  

Racial  inequality  in  science  and  engineering  

•  Misrepresenta(on  isn’t  going  away  

•  Fewer  minori(es  receive  degrees  

15  Source:  NSF  Science  and  Engineering  Indicators  2014  

Women  in  Compu(ng  •  1985:  37%  2000:  29%  2014:  18%  

•  2000  to  2014:  –  5%  decrease  in  math  –  2%  decrease  in  engineering  

16  Source:  NSF  Science  and  Engineering  Indicators  2014  

Graduate  degrees  •  Women  fare  even  worse  

•  2x  as  many  white  males  receiving  degrees  as  all  minori:es  combined  

•  Only  1  in  5  PhDs  female  

Source:  NSF  Science  and  Engineering  Indicators  2014   17  

Unlocking  the  Clubhouse:  A  Case  Study  

•  2014  incoming  class:  40%  women  

Source:  Unlocking  the  Clubhouse,  2002   18  

Triggering  Posi(ve  Change  

“insuring  science  and  technology  are  considered  in  their  social  context  may  be  the  most  important  

change  that  can  be  made  in  science  teaching  for  all  people,  both  male  and  female.”    

Source:  Taggart  &  O’Gara,  2000   19  

Posi(ve  feedback  loop  Relevant  

applica(ons  

Increase  educa(onal  opportuni(es  

AMract  diverse  demographics  

20  

Outline  •  Structural  inequali(es  in  data  science  

•  Democra(zing  data  science  educa(on  

•  Poten(al  solu(ons  

21  

Technologies  •  Recent  technologies  target  broader  audiences  

•  Ooen  require  significant  technical  literacy  

•  Can  we  broaden  access  further  s(ll?  

22  

Crowd-­‐based  efforts  •  Machine  learning  compe((ons  – Can  promote  meaningful  problems  – Low  barrier  to  entry  

•  Ooen  run  by  for-­‐profit  en((es  •  Can  we  encourage  more  ini(a(ves  like  KDD  Cup  2014?  

23  

Educa(on:  MOOCs  •  Tremendous  poten(al  for  reaching  students  

•  Most  ooen  taken  by  professionals  and  people  with  graduate  degrees  

24  

Private  Sector  Opportuni(es  •  Reaching  out  to  underserved  groups  –  Tap  new  markets  

•  Pro  bono  work  could  service  groups  and  bring  in  new  customers  

•  Meaningful  “small  data”  to  serve  the  long  tail  25  

Academic  Research  •  Research  promo(ng  social  good  is  par(cularly  accessible  to  academics  – Social  welfare  problems  ooen  rely  on  public  data  – Academia  is  well-­‐suited  to  interdisciplinary  research  

•  Need  for  focus  on  meaningful  problems  

26  

Your  Solu(on  Here!  •  We  believe  the  community  has  a  responsibility  to  solve  these  problems  

•  Exper(se  in  policy,  business,  sta(s(cs,  healthcare,  computer  science  will  all  be  crucial  

•  Undemocra(c  inequali(es  persist  in  data  science  applica(ons  

•  All  of  us  can  be  part  of  the  solu(on  @mpe(tchou    @williampli    @tweetsbyramesh   27