Jens Ludwig (Georgetown University; NBER ) and Sendhil Mullainathan (University of Chicago) have posted “Fragile Algorithms and Fallible Decision-Makers: Lessons from the Justice System” on SSRN. Here is an excerpt:
One reason for their fragility comes from important econometric problems that are often overlooked in building algorithms. Decades of empirical work by economists show that in almost every data application the data is incomplete, not fully representing either the objectives or the information that decision-makers possess. For example, judges rely on much more information
than is available to algorithms, and judges’ goals are often not well-represented by the outcomes provided to algorithms. These problems, familiar to economists, riddle every case where algorithms are being applied. […] Existing regulations provide
weak incentives for those building or buying algorithms, and little ability to police these choices.
For a method of providing stronger incentives for those building and buying algorithms, see Frank Fagan & Saul Levmore, Competing Algorithms for Law, 88 U. Chicago Law Rev.
AI Law Blawg 