Bridging Ethical Gaps in LLM Research and Design Beyond the Sandbox

Jonathan Beever
Professor and Chair
Department of Engineering and Society

Abstract Artificial intelligence—including from generative to agentic LLMs—is fundamentally altering academic and research workflows. While the productivity gains in scientific and engineering research are widely celebrated, a critical disconnect persists: we routinely treat the ethics of AI research as distinct from its societal and educational impacts and implementations. This division creates a false dichotomy between the obligations of system designers and the responsibilities of end users. In this talk, I unpack the conceptual confusion surrounding contemporary AI and examine how it distorts ethical awareness. Through real-world research and industry cases, I outline core ethical dilemmas facing AI-engaged researchers today—from local-model deployments to agentic integrations. We will explore concrete ethical principles to address pivotal questions: When should researchers exercise the “right to refuse” or refrain from building certain capabilities? How much AI integration is “too much” in scientific workflows? Where do we draw the line on disclosure? By tracking the rapid shifts in both technical design and ethical responsibility, this talk provides actionable frameworks for building the ethics literacy necessary for ethical AI research and integrations.

Bio Jonathan Beever is a philosopher by training whose research engages the intersections among ethics and engineering. He joined UVA this fall as the Chair and Professor of the Department of Engineering and Society and serves nationally as the chair-elect of the Association of Practical and Professional Ethics, Ethics Board member of the American Society for Engineering Education, and co-director of the Online Ethics Center. His current research emphasizes theoretical and practical issues in engineering ethics, including around AI and digital technologies as they related to agency and autonomy, and tensions between built and natural environments. His work is currently funded by the National Science Foundation and by Meta.