{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52951"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52951","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"The Bending Effect : Leveraging Large Language Models to Mitigate the Dilemmas of Deliberative Democracy","abstract":"Thesis (Ph.D.)--Michigan State University. Information and Media - Doctor of Philosophy, 2026","abstract_html":"Thesis (Ph.D.)--Michigan State University. Information and Media - Doctor of Philosophy, 2026","abstract_has_math":false,"creators":["Kang, Taewoo"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Thorson, Kjerstin","Vos, Tim","Hiaeshutter-Rice, Dan","Peng, Wei"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T18:59:34Z","subjects":["Communication","Political science","Artificial intelligence"],"languages":["eng"],"rights":["In Copyright"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52951","local:Kang_grad.msu_0128D_21963"],"render_values":[{"text":"etd:52951","href":null,"code":true},{"text":"local:Kang_grad.msu_0128D_21963","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/qsmw-qk79","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/qsmw-qk79","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_52951","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Thorson, Kjerstin","Vos, Tim","Hiaeshutter-Rice, Dan","Peng, Wei"]},{"key":"dc:creator","label":"Author","values":["Kang, Taewoo"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Communication","Political science","Artificial intelligence"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52951","local:Kang_grad.msu_0128D_21963","https://doi.org/doi:10.25335/qsmw-qk79"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Information and Media - Doctor of Philosophy, 2026","This study investigates the potential of Large Language Models (LLMs) to serve as a gatekeeper of public information capable of enhancing citizens’ democratic political participation. Recent scholarship suggests that LLMs can be leveraged to foster political engagement and democratic understanding by facilitating normative cross-cutting exchange. However, this potential is not immune to persistent dilemmas of deliberative democracy: the inverse relationship between the quality of discourse and the likelihood of participation.Cross-cutting interventions, while often promoting democratic communication in small-group settings, tend to be suppressed at the public level. Ergo, it remains an empirical question whether the democratic potential of LLMs can persist within a contemporary platform ecosystem that often drives political engagement through the reflection of like-minded interests. In this study, the persistence of LLMs’ democratic capacity at the macro-level is conceptualized as the bending effect—the extent to which AI systems can ease the dilemmatic trade-off between deliberative quality and participatory scale via algorithmic mediation. To examine this effect, this study employs an experimental design mirroring an AI-platform equipped with a personalized LLM-powered news service. In December 2025, a representative sample of 711 US adults participated in a virtual beta test of “Create Your Own News” (CYON), a fine-tuned GPT-4o-based web application. Participants were randomly assigned to one of four system-prompt conditions designed to generate news texts regarding the US withdrawal from the Paris Climate Agreement. Drawing on the theory of communicative action and the three-level dialectic intersubjectivity framework, the system prompts were engineered to reflect varying levels of intersubjectivity: (1) a self-confirming perspective (DISP 0), (2) direct perspectives (paralleling agreement/disagreement; DISP 1), (3) meta-perspectives (counterarguments; DISP 2), or (4) contextual meta-perspectives (mutual mis/understanding within a broader social context; DISP 3). Regression analyses revealed that, relative to DISP 0, DISP 3 elicited stronger intentions to share news with dissenters and prompted the use of normative expressions when initiating cross-cutting conversations online. Furthermore, using a competitive communication mediation model, the study explored whether DISP 3 could surmount the cognitive hurdles of like-minded orientation in the broader platform system. It was posited that contextual meta-perspective cues (DISP 3) directly facilitate democratic communication while indirectly discouraging it by lowering the perceived affordance of perspective recognition (PAPR)—users’ perception that the algorithmic system recognizes their specific viewpoint. A multiple nonparametric mediation analysis showed that the total effect of DISP 3 on the deliberative quality of cross-cutting messages remained significant, indicating that DISP 3 preserves the bending effect, thereby fostering democratic discourse within the digital communication ecology. However, the hypothesized backfire effects for DISP 1 and 2 groups were not supported. These findings suggest that LLMs can be leveraged to scale their capacity as a democratic mediator across the mass-oriented digital communication ecology, addressing the long-standing feasibility of deliberative democracy at the macro-level. At the same time, this theoretical silver lining highlights a critical risk: platform operators with exclusive access to system prompts may substantially shape the public sphere. This threat underscores the need for regulatory frameworks that enable civil society to intervene in AI system design in strict accordance with normative standards of democratic reciprocity and journalistic professionalism.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (252 p.) : ill."]},{"key":"dc:title","label":"Title","values":["The Bending Effect : Leveraging Large Language Models to Mitigate the Dilemmas of Deliberative Democracy"]}]}],"canonical_facts":{"dc:contributor":["Thorson, Kjerstin","Vos, Tim","Hiaeshutter-Rice, Dan","Peng, Wei"],"dc:creator":["Kang, Taewoo"],"dc:date":["2026"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Information and Media - Doctor of Philosophy, 2026","This study investigates the potential of Large Language Models (LLMs) to serve as a gatekeeper of public information capable of enhancing citizens’ democratic political participation. Recent scholarship suggests that LLMs can be leveraged to foster political engagement and democratic understanding by facilitating normative cross-cutting exchange. However, this potential is not immune to persistent dilemmas of deliberative democracy: the inverse relationship between the quality of discourse and the likelihood of participation.Cross-cutting interventions, while often promoting democratic communication in small-group settings, tend to be suppressed at the public level. Ergo, it remains an empirical question whether the democratic potential of LLMs can persist within a contemporary platform ecosystem that often drives political engagement through the reflection of like-minded interests. In this study, the persistence of LLMs’ democratic capacity at the macro-level is conceptualized as the bending effect—the extent to which AI systems can ease the dilemmatic trade-off between deliberative quality and participatory scale via algorithmic mediation. To examine this effect, this study employs an experimental design mirroring an AI-platform equipped with a personalized LLM-powered news service. In December 2025, a representative sample of 711 US adults participated in a virtual beta test of “Create Your Own News” (CYON), a fine-tuned GPT-4o-based web application. Participants were randomly assigned to one of four system-prompt conditions designed to generate news texts regarding the US withdrawal from the Paris Climate Agreement. Drawing on the theory of communicative action and the three-level dialectic intersubjectivity framework, the system prompts were engineered to reflect varying levels of intersubjectivity: (1) a self-confirming perspective (DISP 0), (2) direct perspectives (paralleling agreement/disagreement; DISP 1), (3) meta-perspectives (counterarguments; DISP 2), or (4) contextual meta-perspectives (mutual mis/understanding within a broader social context; DISP 3). Regression analyses revealed that, relative to DISP 0, DISP 3 elicited stronger intentions to share news with dissenters and prompted the use of normative expressions when initiating cross-cutting conversations online. Furthermore, using a competitive communication mediation model, the study explored whether DISP 3 could surmount the cognitive hurdles of like-minded orientation in the broader platform system. It was posited that contextual meta-perspective cues (DISP 3) directly facilitate democratic communication while indirectly discouraging it by lowering the perceived affordance of perspective recognition (PAPR)—users’ perception that the algorithmic system recognizes their specific viewpoint. A multiple nonparametric mediation analysis showed that the total effect of DISP 3 on the deliberative quality of cross-cutting messages remained significant, indicating that DISP 3 preserves the bending effect, thereby fostering democratic discourse within the digital communication ecology. However, the hypothesized backfire effects for DISP 1 and 2 groups were not supported. These findings suggest that LLMs can be leveraged to scale their capacity as a democratic mediator across the mass-oriented digital communication ecology, addressing the long-standing feasibility of deliberative democracy at the macro-level. At the same time, this theoretical silver lining highlights a critical risk: platform operators with exclusive access to system prompts may substantially shape the public sphere. This threat underscores the need for regulatory frameworks that enable civil society to intervene in AI system design in strict accordance with normative standards of democratic reciprocity and journalistic professionalism.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (252 p.) : ill."],"dc:identifier":["etd:52951","local:Kang_grad.msu_0128D_21963","https://doi.org/doi:10.25335/qsmw-qk79"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["In Copyright"],"dc:subject":["Communication","Political science","Artificial intelligence"],"dc:title":["The Bending Effect : Leveraging Large Language Models to Mitigate the Dilemmas of Deliberative Democracy"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:34Z"}