Machine Learning: Enhancing Data Accuracy; The Arup Deepfake Scam: An Actor-Network Theory Analysis of AI-Enabled Financial Fraud

My technical project and my STS research paper are linked through the exploration of how emerging technologies significantly impact corporate processes and decision-making. Both projects demonstrate how advanced technologies like machine learning in my technical work and deepfake technology in my STS research transform traditional workflows within organizations. However, they approach this transformation from different perspectives. My technical project focuses on leveraging machine learning to improve efficiency and accuracy within an established corporate data system, whereas my STS research investigates how deepfake technology disrupts organizational trust and authority structures. Thus, while my technical project applies new technologies to strengthen corporate systems, my STS research critically analyzes how similar technological advancements can also introduce vulnerabilities within these systems. My technical report describes the implementation of a machine learning-based system at Fannie Mae designed to automate the identification of incorrect job key data entries. The system initially used BERTScore, a sophisticated semantic similarity model, but transitioned to ROUGE due to practical constraints regarding computational resources and company policies. This automated matching system, implemented in Python and deployed using Docker and Flask, significantly enhanced Fannie Mae’s data accuracy and efficiency. Specifically, the project resulted in a 70% reduction in manual data verification time, allowing engineers to focus more strategically on resolving issues rather than manually identifying them. In my STS research paper, I examined the implications of deepfake technology in corporate financial fraud using Actor-Network Theory (ANT). The analysis centered on the Arup deepfake scam, where cybercriminals successfully impersonated a senior executive to fraudulently extract $25 million. My argument is that this fraud succeeded due to a sociotechnical network where deepfake technology, corporate hierarchy, human trust, and insufficient regulatory frameworks were strategically aligned by cybercriminals. ANT enabled a nuanced analysis, framing deepfakes as active participants in networks rather than isolated technological tools, highlighting how deepfake technology integrates into corporate networks and exploits organizational vulnerabilities. Simultaneously working on both projects provided me with significant insight into how technologies reshape corporate environments, revealing both their potentials and risks. My technical work at Fannie Mae showed how carefully implemented machine learning can dramatically enhance corporate efficiency, demonstrating the constructive potential of technology integration. Conversely, the STS research underscored the need for vigilance and enhanced protective measures in organizational settings, recognizing that technological advancement can also create significant new risks. Understanding this dual nature of technology informs my perspective as a computer scientist, emphasizing the ethical implications and responsibilities accompanying technological innovation. Moving forward, I intend to apply the insights from my STS research to future technical endeavors, ensuring that technology implementations account for both their intended benefits and unintended vulnerabilities within organizational systems.