Affective Dimensions of Trust in Conversational AI: A Qualitative Study of International Students Navigating Public Services in Finland
Intiser, Md Aseer (2026)
2026
Master's Programme in Computing Sciences and Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2026-06-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202606087129
https://urn.fi/URN:NBN:fi:tuni-202606087129
Tiivistelmä
The public-service landscape in Finland is institutionally complex, digitally mediated and difficult to read from outside for international students arriving in Finland. Increasingly, conversational AI is suggested as a means to take up some of this cognitive and emotional load, but the current trust frameworks in human-computer interaction (HCI) view trust as a cognitive assessment of accuracy. This thesis investigates the formation, maintenance and breakdown of trust as affect in the interactions between international students and a purpose-built conversational AI assistant for navigating public services in Finland.
The study is qualitative and interpretive. It involved fifteen international students at Tampere University who participated in single-session studies of approximately thirty-five minutes, which included a brief pre-interaction interview, four scenario-based interactions with the Lumi prototype (covering student housing, banking, healthcare, and university support), and a post-interaction interview consisting of seven questions. A recall framing asked participants to consider the scenarios from the point of view of their earlier first-month selves. The data were analyzed using Braun and Clarke's reflexive thematic analysis, while the Trust-as-Affect framework was used as a sensitizing concept, not as a closed model.
Seven themes were developed across three analytical dimensions: the foundations of affective trust (authority-anchored reassurance, calm through structure), negotiation of affective trust (recognition as reliability-for-one's-situation, the verification disposition), and dynamics and limits of affective trust (asymmetric build-and-crash, vulnerability and identity as amplifiers, and the orientation-action gap). The findings reposition trust in this domain as the coexistence of knowledge-based verification and emotionally felt experience (epistemic and affective registers) operating simultaneously, rather than as a purely cognitive judgment of informational accuracy. The thesis elaborates the Trust-as-Affect framework on three points: the co-constitutive relationship between epistemic and affective registers, the asymmetric temporal dynamics of trust (building incrementally and collapsing instantly) and recognition as a latent condition without which even accurate information produces residual uncertainty. The thesis further reframes conversational AI in public-service settings as institutional voice, instead of consumer product. From these findings the thesis derives a coherent design posture for conversational AI in newcomer high-load public-service contexts. Effective systems in this domain should be source-anchored, structurally minimal, scope-stated, verification-supporting, hallucination-conservative, vulnerability-aware, and action-closing, each principle corresponding directly to one of the seven analytical themes.
The study is qualitative and interpretive. It involved fifteen international students at Tampere University who participated in single-session studies of approximately thirty-five minutes, which included a brief pre-interaction interview, four scenario-based interactions with the Lumi prototype (covering student housing, banking, healthcare, and university support), and a post-interaction interview consisting of seven questions. A recall framing asked participants to consider the scenarios from the point of view of their earlier first-month selves. The data were analyzed using Braun and Clarke's reflexive thematic analysis, while the Trust-as-Affect framework was used as a sensitizing concept, not as a closed model.
Seven themes were developed across three analytical dimensions: the foundations of affective trust (authority-anchored reassurance, calm through structure), negotiation of affective trust (recognition as reliability-for-one's-situation, the verification disposition), and dynamics and limits of affective trust (asymmetric build-and-crash, vulnerability and identity as amplifiers, and the orientation-action gap). The findings reposition trust in this domain as the coexistence of knowledge-based verification and emotionally felt experience (epistemic and affective registers) operating simultaneously, rather than as a purely cognitive judgment of informational accuracy. The thesis elaborates the Trust-as-Affect framework on three points: the co-constitutive relationship between epistemic and affective registers, the asymmetric temporal dynamics of trust (building incrementally and collapsing instantly) and recognition as a latent condition without which even accurate information produces residual uncertainty. The thesis further reframes conversational AI in public-service settings as institutional voice, instead of consumer product. From these findings the thesis derives a coherent design posture for conversational AI in newcomer high-load public-service contexts. Effective systems in this domain should be source-anchored, structurally minimal, scope-stated, verification-supporting, hallucination-conservative, vulnerability-aware, and action-closing, each principle corresponding directly to one of the seven analytical themes.