Science News Archive

Computer predicts who will walk on Death Row

A US researcher has developed a computer programme that can predict with 90% accuracy who will be the dead man walking on death row.

Texas A&M University computer scientist Stamos Karamouzis, writing in the International Journal of Law and Information Technology, built a computer-based neural network to examine the characteristics of the death row inmates who never get executed versus those who do.  He first trained the system by feeding in the profiles of over 1000 inmates from 1973 to 2000, half of whom had been executed half of whom had not.  

Details of 18 different parameters including sex, race, age, educational level and marital status were entered.  The team then fed in the details of 300 further inmates and asked the network to predict the outcome on the basis of what it had just learned.

Astonishingly it correctly predicted the fates of the 300 with 90% accuracy.  To find out how the team then re-entered the data but skipped single details each time to identify which factors were the most significant in controlling the outcome.  Predictably, gender played a very large role, and women are almost never executed in the US, but surprisingly race, which has often been held up as a key determinant in execution cases, was found to play little part.

A major predictor, however, turned out to educational level, probably because it governs how well an inmate is to able to defend himself. According to one commentator, Simon Shepherd, from the capital punishment action group Death Watch International, "this finding confirms that being executed is not about what you've done, but more about your ability to defend yourself."

29th Jun 2008