E108 Advancing Dialogic Communication Skills through LLM-Based Simulation (Short paper) Introduction Communication skills are increasingly recognized as essential for science students (Akin et al., 2021). The benefits associated with improved communication skills span multiple levels. First, effective communication benefits society: Clear and engaging communication fosters public engagement with science, supports informed decision-making, and promotes a scientifically literate citizenry (Brownell et al., 2013). It also plays a key role in combating misinformation and building public trust, particularly in times of crisis, such as pandemics (Fischhoff, 2013). Second, stronger communication skills benefit scientists themselves, making researchers better positioned to advocate for science policy and funding (Baron, 2010), enhance the impact of their work, and advance their academic opportunities (Akin et al., 2021). Third, communication benefits science itself. It underpins interdisciplinary collaboration, which advances the collective growth of scientific knowledge. Moreover, engaging with the public and policymakers can help scientists align their research with societal needs, and learn from lay expertise, thereby enhancing the relevance and impact of their work (Besley et al., 2015). Recent years brought an increase in science communication training programs for science professionals (Baram-Tsabari & Lewenstein, 2017; Besley et al., 2015; Coletti et al., 2023). However, while most existing training programs focus on one-way dissemination of knowledge (Reincke et al., 2020), research increasingly points to the value of practicing dialogue: engaging in two-way conversations that acknowledge the perspectives, concerns, and knowledge of diverse audiences (Lewenstein & Baram-Tsabari, 2022; Worthington et al., 2024). Yet meaningful practice in dialogic communication typically requires human actors and expert facilitators to provide feedback, resources that are costly, time-consuming, and often unavailable (Chu & Goodell, 2024; Kerr et al., 2020). Recent advances in generative AI, and especially the advent of large language models (LLMs), offer promising opportunities to overcome these barriers by providing accessible, adaptive, and scalable tools for practicing dialogic communication. Research Goal and Questions Our goal in this research was to explore whether an LLM-based simulator can offer a scalable, accessible alternative for dialogic science communication training. Specifically, our research questions were: RQ1: Do students improve their dialogic communication performance following training with an LLM-based communication simulator? RQ2: How do students experience the tool, and what benefits do they perceive in using it? Theoretical Framework For AI to be able to effectively support dialogic communication training, we need to first clearly define to the machine what productive dialogue entails, what its characteristics and features are. To this end, we developed Prodigy (Productive Dialogue for Generative AI), a theoretical framework that defines what constitutes productive dialogue by drawing on key ideas from science communication, academic productive dialogue, and nonviolent communication. Prodigy comprises 15 features organized into four dimensions: Content (e.g. clarity, credibility, reasoning), Interpersonal Rapport (e.g. empathy, respect, sharing personal details), Perspective Taking & Listening (e.g. inviting the partner to share their ideas, building on the partner ideas), and Integrity & Humility (e.g. expressing intellectual humility, transparency regarding one’s own stance or agenda). It serves both as a guiding structure for LLM-generated feedback and as an analytic rubric for evaluating communicative performance.
RkJQdWJsaXNoZXIy Mjk0MjAwOQ==