AI as We Describe It: How Large Language Models and Their Applications in Health are Represented Across Channels of Public Discourse

TL;DR

This paper analyzes how LLMs and health applications are represented across five public channels (news, research press, YouTube, TikTok, Reddit) over a two-year period, using three analytical dimensions: lexical style, informational content, and symbolic representation. It finds that discourse is broadly positive and episodic, with risk discussions limited to information-quality incidents and rare explanations of generative mechanisms. Cross-channel differences reveal that lay platforms emphasize wellbeing and anthropomorphism, while professional outlets foreground clinical and systemic contexts. By treating public discourse as a diagnostic tool for literacy and governance gaps, the study informs strategies for more informed engagement with LLMs in health and for governance that matches public concerns and information needs.

Abstract

Representation shapes public attitudes and behaviors. With the arrival and rapid adoption of LLMs, the way these systems are introduced will negotiate societal expectations for their role in high-stakes domains like health. Yet it remains unclear whether current narratives present a balanced view. We analyzed five prominent discourse channels (news, research press, YouTube, TikTok, and Reddit) over a two-year period on lexical style, informational content, and symbolic representation. Discussions were generally positive and episodic, with positivity increasing over time. Risk communication was unthorough and often reduced to information quality incidents, while explanations of LLMs' generative nature were rare. Compared with professional outlets, TikTok and Reddit highlighted wellbeing applications and showed greater variations in tone and anthropomorphism but little attention to risks. We discuss implications for public discourse as a diagnostic tool in identifying literacy and governance gaps, and for communication and design strategies to support more informed LLM engagement.

Figures (8)

• Figure 1: Overview of Study. This large-scale quantitative description investigates how public discourse introduces LLMs and their applications in the health domain, by examining five prominent discourse channels (news, research press, YouTube, TikTok, and Reddit) between December 2022 and December 2024. Drawing on agenda-setting theory, it studies three core dimensions of discourse: (1) Lexical Style: Examining the overall presentation style of discourse through emotional tone and writing formality. (2) Informational Content: Analyzing how messages frame LLMs' implications, risks, mechanisms, and potential across health domains. (3) Symbolic Representation: Studying the level of anthropomorphic representation of LLM entities.
• Figure 2: Trends in median emotional tone. TikTok data ended in August 2024 due to an unresolved API internal error (see Sec. \ref{['section:data_collection']}). We applied LOESS (Locally Estimated Scatterplot Smoothing) to capture overall trends across five data sources of varying size and nature.
• Figure 3: Lexical style in public discourse on LLMs for health across. Kruskal–Wallis H-tests were performed to determine whether there was a significant difference across channels (*** $p$<0.001, ** $p$<0.01, * $p$<0.05), where emotional tone $H$ = 1515.19 (***), analytic $H$ = 3250.69 (***), clout $H$ = 2645.57 (***), and authentic $H$ = 304.95 (***).
• Figure 4: Framing type and dimensions in public discourse on LLMs for health. Kruskal–Wallis H-tests were performed to determine whether there was a significant difference across channels (*** $p$<0.001, ** $p$<0.01, * $p$<0.05). For Framing dimensions: Capacity and Resources $H$ = 29.82 (***), Crime and Punishment $H$ = 85.30 (***), Cultural Identity $H$ = 101.83 (***), Economic $H$ = 821.39 (***), External Regulation and Reputation $H$ = 655.86 (***), Fairness and Equality $H$ = 145.74 (***), Health and Safety $H$ = 1105.16 (***), Legality, Constitutionality, Jurisdiction $H$ = 63.97 (***), Morality and Ethics $H$ = 34.47 (***), Policy Prescription and Evaluation $H$ = 1522.48 (***), Political Factors and Implications $H$ = 2701.61 (***), Public Sentiment $H$ = 140.98 (***), Quality of Life $H$ = 109.28 (***), Security and Defense $H$ = 681.08 (***).
• Figure 5: Disclosure of risks and generative nature in public discourse on LLMs for health. Kruskal–Wallis H-tests were performed to determine whether there was a significant difference across channels (*** $p$<0.001, ** $p$<0.01, * $p$<0.05), where risks to individual behaviors $H$ = 1515.19 (***), risks to human-centered care $H$ = 1389.36 (***), risks to information ecosystems $H$ = 4973.30 (***), risks to technology accountability $H$ = 3184.23 (***), and explanations of generative nature $H$ = 4203.25 (***).