Golf and GameForge: Innovative Analytics for Recommender Systems; Development and Distribution of Rekognition, an Amazon Facial Recognition Software Tool, in Relation to Virtue Ethics

The technical and STS research project outlined in my thesis explore the research
question of how to mitigate the socio-technical problem of biases and disadvantages that arise
when Artificial Intelligence(AI) based predictive modeling methods are used to create selection
tools. The two projects work to create a deliverable and an understanding to better create
data-driven tools that are equitable and efficient. In this synthesis I will outline the two projects I
worked on to explore the overarching research question present and discuss the insights I gained
while working on them.
The technical project component of my overarching research question, consists of
building a data-driven AI tool using predictive modeling to help streamline the collegiate golf
recruitment process. The tool will pair golf players to teams based on optimizing both parties'
statistics and future growth. The goal of this deliverable is to help alleviate the current
fragmented recruitment process and to create a platform that is accessible to all golf players and
coaches. The new system will build upon the existing Gameforge features using predictive
modeling, artificial intelligence, machine learning, publicly available data, and proprietary
metrics from Gameforge. The enhancement will consist of three components: a two way
recommender system for players and coaches, a player archetype creation to identify strengths
and skill gaps, a new holistic ranking system that goes beyond player scores. In conjunction with
the technical project, I worked on a STS research project that analyzed the inherent flaws and
bias present in a data-driven, artificial intelligence based facial recognition tool (FRT). In order
to carry out this analysis I explored Rekognition, a FRT available in the public market,
developed by Amazon that was created to identify individuals by examining and mapping facial
features. The tool learns to identify individuals by looking for patterns in large data sets that
serve as the baseline training data set for facial matches made. Due to public concern and outrage
over Rekognition being used in law enforcement settings and leading to wrongful arrests, two
MIT researchers decided to explore the accuracy of the facial matches made by the tool. They
found that the error rates for gender classification were higher for females than they were for
males, and higher for darker skin toned individuals than for lighter skin toned individuals.
Analyzing these two complementary projects, pertaining to my overarching research
question, allowed me to gain insight that I would have not gained from them individually. The
technical project work on developing a recruitment selection platform for collegiate athletes,
added to my understanding of utilizing data in order to make informed decisions that have been
verified and analyzed to contribute to overarching goals. The STS research project helped me see
that AI tools designed to streamline tedious processes may overlook qualitative components.
Both projects led me to recognize that AI based predictive modeling tools can create a
socio-technical world where selection processes become closed off to new experiences or metrics
that could be indicators of a successful selection and can exacerbate biases present in training
datasets. These projects have shown me that engineering is not just about creating efficient and
well designed artifacts but rather about building a holistic socio-technical world while
innovating

School of Engineering and Applied Science
Bachelor of Science in Systems Engineering
Technical Advisor: William Scherer
STS Advisor: Benjamin Joseph Laugelli
Technical Team Members: Rachel Krietzer, Sam Roberts, Jerry Lu, Thomas Twomey, Zachary Kay, Steven Wasserman