Hybrid Horizons: Defining the Boundaries of Machine and Human Intelligence in AI-Assisted Bioethics Analysis

Abstract: The growing capacity of large language models (LLMs) to effectively summarize large corpora of natural language has bioethicists and social scientists wondering whether LLMs can effectively assist in qualitative data analysis. This talk presents findings from a multisite collaboration (Baylor College of Medicine / Harvard Medical School) to explore capacities and limitations of using LLMs for qualitative bioethics analysis of three separate datasets comprised of interviews exploring ethics of 1) AI-assisted deep brain stimulation for neurogenerative conditions; 2) algorithmic risk prediction for addressing heart failure; and 3) integration of AI-based computer perception into clinical care. We identify key strengths of LLMs (deductively highlighting known constructs or topics) and enduring challenges (to inductively identify unanticipated or unknown constructs) and discuss their implications for pursuing distinct paradigms (e.g., thematic; interpretive phenomenological; discourse analysis; etc.) in empirical bioethics research. We highlight concrete prompts and approaches (e.g., one-shot vs. many-shot; chain of thought) particularly suited to pursuing different analytic goals. Importantly, we contextualize our findings within a broader discourse around combined human-AI or “hybrid” intelligence and discuss unique concerns around applying LLMs in bioethics versus other domains of inquiry.

Amanda Merner – Center for Bioethics – Harvard Medical School; Jennifer Blumenthal-Barby – Center for Medical Ethics and Health Policy – Baylor College of Medicine; Gabriel Lazáro-Muñoz – Center for Bioethics – Harvard Medical School