Machine Learning: An Applied Econometric Approach

https://scholar.harvard.edu/files/sendhil/files/jep.31.2.87.pdf


Machines are increasingly doing “intelligent” things: Facebook recognizes
faces in photos, Siri understands voices, and Google translates websites.
The fundamental insight behind these breakthroughs is as much statis-
tical  as  computational.  Machine  intelligence  became  possible  once  researchers 
stopped  approaching  intelligence  tasks  procedurally  and  began  tackling  them 
empirically. Face recognition algorithms, for example, do not consist of hard-wired
rules  to  scan  for  certain  pixel  combinations,  based  on  human  understanding  of 
what  constitutes  a  face.  Instead,  these  algorithms  use  a  large  dataset  of  photos 
labeled  as  having  a  face  or  not  to  estimate  a  function 
f
(
x
)  that  predicts  the  pres-
ence
y
 of a face from pixels
x
. This similarity to econometrics raises questions: Are
these algorithms merely applying standard techniques to novel and large datasets?
If there are fundamentally new empirical tools, how do they fit with what we know?
As empirical economists, how can we use them?
1