AI is increasingly embedded in civic technologies to moderate, structure and summarize large-scale public discussions. While these capabilities can scale participation, they also introduce an interpretability gap: the transformations performed by AI mediation pipelines (e.g., clustering, summarization, ranking) often remain hard to understand, verify and contest for heterogeneous publics, with potential consequences for democratic legitimacy. This thesis reframes explainability in civic deliberation as a communication design problem. It combines a systematic literature review across XAI, human-centered XAI, information visualization, data storytelling and civic tech, with a Research Through Design process conducted within the EU ORBIS project. Insights from the review are translated into a consolidated set of requirements and design principles, aligned with ORBIS constraints, and operationalized in Il Corollario—an interactive knowledge-graph interface that represents positions, arguments, AI-generated clusters and keywords, and supports progressive disclosure, traceability to source contributions and uncertainty framing. The prototype is integrated into the ORBIS reporting ecosystem and evaluated through user feedback and a final expert validation against the requirements. Results indicate that combining network visualization with narrative and interaction scaffolds can reduce cognitive load, improve orientation and support critical engagement with AI outputs, while highlighting remaining risks of misinterpretation and the need for clearer cues and verification pathways. The work contributes (1) a requirements-driven framework for communication design in AI-supported deliberation, and (2) a validated design artifact showing how interface-level mediation can support intelligibility and contestability of AI-assisted collective representations.
L’AI viene sempre più integrata nelle tecnologie applicate al contesto civico per moderare, strutturare e sintetizzare discussioni pubbliche su larga scala. Sebbene queste capacità possano aumentare la partecipazione, introducono anche un divario di interpretabilità: le trasformazioni eseguite dai processi di mediazione dell’intelligenza artificiale (ad esempio clustering, sintesi e classificazione) rimangono spesso difficili da comprendere, verificare e contestare per un pubblico eterogeneo, con potenziali implicazioni per la legittimità democratica. Questa tesi riformula la spiegabilità nella deliberazione civica come problema di communication design. Il lavoro combina una systematic literature review su XAI, Human-Centered XAI, information visualization, data storytelling e civic tech con un percorso di Research Through Design svolto nel contesto del progetto europeo ORBIS. Le evidenze della review vengono tradotte in un set consolidato di requisiti e principi di progetto, coerenti con vincoli e obiettivi di ORBIS, e implementate in Il Corollario: un’interfaccia basata su knowledge graph che rappresenta posizioni, argomenti, cluster e keyword generate dall’AI e abilita la rivelazione progressiva dei contenuti, tracciabilità alle fonti ed inquadramento dell’incertezza. Il prototipo è integrato nell’ecosistema di reportistica ORBIS e valutato tramite feedback degli utenti e una validazione finale con esperti, allineata ai requisiti. I risultati suggeriscono che l’integrazione di network visualization con strutture narrative e di interazione può ridurre il carico cognitivo, migliorare l’orientamento e sostenere un ingaggio critico verso gli output dell’AI, evidenziando al contempo rischi residui di errata interpretazione e la necessità di indicazioni e percorsi di verifica più chiari ed espliciti. Il lavoro contribuisce con (1) un framework di requisiti per il design della comunicazione applicato alla deliberazione AI-enhanced e (2) un artefatto progettuale validato che mostra come la mediazione a livello di interfaccia possa supportare intelligibilità e contestabilità delle rappresentazioni collettive assistite dall’AI.
From knowledge graphs to civic sensemaking: communication design for human-centered explainable AI in ai-enhanced deliberation
Garetto, Giacomo
2024/2025
Abstract
AI is increasingly embedded in civic technologies to moderate, structure and summarize large-scale public discussions. While these capabilities can scale participation, they also introduce an interpretability gap: the transformations performed by AI mediation pipelines (e.g., clustering, summarization, ranking) often remain hard to understand, verify and contest for heterogeneous publics, with potential consequences for democratic legitimacy. This thesis reframes explainability in civic deliberation as a communication design problem. It combines a systematic literature review across XAI, human-centered XAI, information visualization, data storytelling and civic tech, with a Research Through Design process conducted within the EU ORBIS project. Insights from the review are translated into a consolidated set of requirements and design principles, aligned with ORBIS constraints, and operationalized in Il Corollario—an interactive knowledge-graph interface that represents positions, arguments, AI-generated clusters and keywords, and supports progressive disclosure, traceability to source contributions and uncertainty framing. The prototype is integrated into the ORBIS reporting ecosystem and evaluated through user feedback and a final expert validation against the requirements. Results indicate that combining network visualization with narrative and interaction scaffolds can reduce cognitive load, improve orientation and support critical engagement with AI outputs, while highlighting remaining risks of misinterpretation and the need for clearer cues and verification pathways. The work contributes (1) a requirements-driven framework for communication design in AI-supported deliberation, and (2) a validated design artifact showing how interface-level mediation can support intelligibility and contestability of AI-assisted collective representations.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Garetto.pdf
accessibile in internet per tutti
Descrizione: Tesi
Dimensione
5.3 MB
Formato
Adobe PDF
|
5.3 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/252101