death penalty news March 16, 2005
USA: Researchers Able to Predict Death Penalty Outcomes Description Researchers at Loyola University New Orleans have developed, trained and tested an artificial neural network that is more than 90 percent accurate at predicting whether a convicted capital offender will be executed or not. Following a Supreme Court decision prohibiting the execution of minors which could have ramifications for the future of the death penalty, researchers at Loyola University New Orleans have found further evidence questioning the fairness of the capital punishment process. The researchers have developed, trained and tested an artificial neural network that is more than 90 percent accurate at predicting whether a convicted capital offender will be executed or not. "Our research can specify the post death-conviction process and can add evidence concerning the fairness or unfairness of the process," says Dr. Dee Wood Harper, professor of sociology and criminology. "Predicting execution outcomes for prisoners under a sentence of death utilizing attributes that have no direct bearing on the judicial process has serious implications concerning the fairness of the death penalty." An artificial neural network (ANN) is a multiprocessor computing system that resembles the way biological nervous systems process information. ANNs are capable of learning on their own or by example through a process that involves adjustments to the connections that exist between the neurons. For this project, researchers Harper and Dr. Stamos Karamouzis, associate professor of computer science and developer of the network, reconstructed the profiles of more than 1,300 death row inmates from a national population by using simple attributes such as the inmates' race, sex, age and highest year of education completed at the time of first imprisonment for capital offense. "We took a thousand of those profiles and used them to 'train' the network," says Harper of the process that works in the same way a human brain would study for a test. "We then tested the ANN using 300 profiles that the network never witnessed before," says Karamouzis. "The network was capable of correctly predicting execution/non-execution at a rate higher than 90 percent." "What that says to me from a policy point of view is even after conviction the process is arbitrary and seems to be biased in some way," says Harper. "If a machine can tell who's dead and who's not, that's pretty serious. Even after an offender is convicted, these variables influence who's executed and who's not. That's an important finding." Details of the network's functionality and implications have been announced in Amsterdam at the 2004 Conference of the European Society of Criminology and in Austria at the 2005 International IASTED Conference on Artificial Intelligence Applications. The work was published February in Artificial Intelligence and Applications. Please let me know if there is anything further that I can provide, or if you would like a copy of the paper, which is titled "Artificial Neural Networks for Predicting Death Penalty Outcomes." We help Loyola University New Orleans with some of its public affairs work. (source: newswise.com)