Georgia Institute of Technology
Assessing and Communicating Risks of Generative AI in Public Health Information Ecosystems
Abstract
dc:description.abstractTechnology is increasingly influencing the way we interact with information, where both the quality of that information and the traits and affordances of information technologies shape people's attitudes and actions. Generative Artificial Intelligence (AI), such as large language models (LLMs), differs fundamentally from prior information and communication technologies (ICTs) that were primarily task-centric, as it can produce new content in a probabilistic manner and at scale. This generative nature is both its power and its pitfall. It enables accessible and fluent responses to complex informational needs, but it can also produce content that is inaccurate, biased, or oversimplified in ways that existing tools struggle to detect. Yet these systems are being rapidly adopted as standalone applications or embedded into existing platforms, including in public health contexts, often without users' awareness of how they generate content or what their limitations are. This imbalance between adoption speed and public understanding raises pressing questions about how generative AI disrupts the ecosystems through which people seek, evaluate, and act on health information. I posit that generative AI introduces ecological disruptions to public health information ecosystems that require contextually grounded risk assessment and literacy-centered risk communication to understand and address underlying harms. Situated within the context of a variety of public health crises and challenges, this dissertation adopts computational, qualitative, and experimental methods alongside deep-seated domain collaborations with experts in the fields of public health, communication, and natural language processing. Amid the formative years of generative AI, this dissertation examines these disruptions at multiple levels. I first demonstrate that low-quality information types co-occur and interact to amplify harm, and that perceived severity varies by both content composition and reader context, showing the limits of content-level categorization as a basis for harm assessment. Building on this, I show that AI-generated low-quality content differs from human-created equivalents in emotional appeals, fabricated evidence, and authoritative language, and that these differences degrade the performance of both algorithmic detectors and human guidelines. I then present stakeholder perspectives on how LLM adoption introduces ecological risks spanning individuals, human-centered care relationships, information ecosystems, and technology accountability. Examining how these risks are represented in public discourse, I find that narratives are broadly positive, risk communication is shallow and uneven across channels, and explanations of LLMs' generative nature are nearly absent. Finally, I show that while structured risk communication reduces general trust in LLMs, it does not necessarily improve decision confidence, and that its effects are contingent on recipients' existing AI literacy, pointing to baseline AI knowledge as a necessary precondition for risk communication to support informed engagement. Collectively, this dissertation reframes the problem of generative AI in public health information ecosystems in two directions: what AI does to the ecosystem by compressing distributed judgment into synthesized responses and simulating the signals users relied on to assess trustworthiness, and what the ecosystem fails to do in response by communicating risks to users who lack the conceptual foundation to act on them. In doing so, it contributes to human-computer interaction, social computing, and responsible AI research, and advances empirical grounding for more responsible adoption, communication, and governance of generative AI for health, with implications for researchers, practitioners, and policymakers evaluating what meaningful informed engagement with AI requires.
Degree
thesis:*- Name thesis:degree_name
- Human-Centered Computing, PhD
- Department thesis:degree_department
- Interactive Computing
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhou, Jiawei
- Advisor dc:contributor.advisor
-
- De Choudhury, Munmun
- Committee members dc:contributor.committeemember
-
- Parker, Andrea
- Kumar, Srijan
- Liao, Vera
- Diakopoulos, Nicholas
- ORCID person:identifier.orcid
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/82132