Abstract
This research portfolio focuses on the growing importance of verifying and critically analyzing AI-generated outputs as AI (artificial intelligence) continues to take on increasingly prevalent and influential roles in everyday systems. In this portfolio, the technical capstone component details proposed improvements to the University of Virginia's Software Engineering course (CS 3240) to better prepare students for real-world software development, particularly with AI. It highlights safer and more maintainable ways in which students can be taught to build software systems with generative tools, emphasizing the importance of careful scrutiny of AI solutions rather than blind acceptance, along with human oversight in AI-assisted development to ensure outputs are consistently validated through practices such as testing and code reviews. In parallel, the STS research paper for this portfolio explores the concept of trustworthy AI in the context of governance, specifically examining the implications of allowing an AI system in Albania to participate in making decisions in its historically corruption-prone procurement sector. While the technical capstone stresses the importance of human oversight in software development, the STS research paper similarly underscores the possible risks of delegating critical decisions to AI systems entirely, illustrating the need for structured oversight and critical evaluation of such systems to ensure safe, responsible use.
The University of Virginia’s upper-level Software Engineering course (CS 3240) has adapted to modern development practices by allowing students to build their semester-long projects with the help of generative AI (GAI) tools. Although GAI can improve efficiency and serve as a useful starting point for code generation and development, relying on it too heavily in academic settings can leave students insufficiently prepared to design robust systems, critically assess AI outputs, and/or ensure correctness. This gap can carry into industry, where weak validation practices may result in unreliable, difficult-to-maintain software that hinders innovation and erodes consumer trust. To address these risks, the technical capstone proposes the introduction of structured peer code reviews alongside the implementation of stronger software testing requirements within the course. Code reviews would push students to carefully examine both their own code and AI-generated contributions, encouraging deliberate evaluation rather than passive acceptance while also promoting collaboration and accountability. Expanding the role of testing in the course would further reinforce the need to confirm system behavior and catch errors early in the development process, especially in AI-assisted workflows. Together, these changes aim to better align the university’s course with industry expectations and standards to help students develop the critical thinking and quality assurance skills necessary to work responsibly and effectively with AI.
In contrast, the STS research paper of this portfolio shifts to a different domain where AI is gaining traction: governance. The STS research examines Albania’s recent rollout of the AI system Diella, which has been introduced to support decision-making in the government's procurement sector, an area historically affected by corruption and abuse. The paper analyzes how the Albanian government has promoted and anthropomorphized Diella, framing it as an “AI minister” to build public trust and signal governmental transparency and also explores the motivations behind adopting this technology, including efforts to reduce human bias and increase efficiency with the help of algorithms that remain objective. The paper also raises questions on the broader implications of integrating AI into governance, drawing out both the good and bad that come with its use, evaluating potential benefits—such as improved accountability and consistency—and risks—such as algorithmic bias and loss of accountability.
Together, the topics and ideas introduced in both the technical capstone and the STS research paper highlight the tensions between human and machine judgment, both raising broader questions about trust, human authority, accountability, and the responsible integration of AI in both technical and societal domains, ultimately advocating for a collaborative model where human oversight remains central.