CLEAN - LSE Seminar Series - Gary Wagner (University of Louisiana at Lafayette)


The next CLEAN seminar, in collaboration with LSE, will take place on Monday, April 8th 2024, from 4-5 p.m. Italian time, in Room 2-B3-SR02 of the Roentgen building.

The presenter will be Gary Wagner (University of Louisiana at Lafayette). The title of the paper is "Understanding the Punitive Cost of a Mug Shot: A Machine Learning Application".



Applications for machine learning and AI techniques have expanded exponentially in recent years, but the ethical understanding around appropriateness of these uses is still a contentious debate. First, this paper delves into a meta-analysis of the (limited) existing research regarding application of these approaches within a criminal justice context. Second, we expand on the literature relating attractiveness and criminal behavior, by using machine learning to analyze mug shot images from adults arrested in Pulaski County, Arkansas in 2014 to explain conviction, sentencing, and recidivism outcomes. This novel dataset of more than 4,000 arrestees is linked to court records and criminal history. The images themselves are analyzed using the 10k US Adult Faces Database, providing rigorous and objective ratings of facial characteristics for offenders in the dataset. Ratings include 40 characteristics such as: attractiveness, trustworthiness, friendliness, and aggressiveness. Specifically, we consider the impact of facial characteristics at two distinct stages of criminal justice processing: conviction and sentencing. We compare traditional economic model predictions with those of machine learning models (ie. XGBoost). This case study provides a tangible example to more thoroughly consider the applicability and limitations of machine learning technologies in informing criminal justice policy. While this study focuses on criminal justice policy, the exercise is easily generalized to other areas of societal importance.


The seminar will take place in Room 2-B3-SR02, 2nd floor, via Roentgen 1.

You may follow the event online via Zoom Meetings at the following link: - Meeting ID: 935 3789 3727 - Passcode: 327396