Software models death row outcomes: Two reseachers created a software program to examine the demographics of death row inmates. The results were highly predictive…which is bad, if you read the second paragraph below.
The neural network, which learns by constantly scanning the data for patterns, was given 1,000 cases from 1973 to 2000 where the outcome was known. Once trained on that information, it was fed another 300 cases but without the outcome included. That’s when its prediction proved highly accurate.
What some observers find alarming about the outcome is that the 19 points of data supplied on each death-row inmate contained no details of the case. Only facts such as age, race, sex, and marital status were included, along with the date and type of offense.