Abstract

As AI companions evolve from mere tools into relational partners, millions of users are forming deep parasocial attachments. However, the industry currently lacks ethical protocols for termination, often treating the end of these interactions as purely technical events. This oversight can lead to “digital grief” when companions vanish without warning. We conducted a between-subjects experiment(𝑁 = 26) to test whether forewarning mitigates user distress. Participants interacted with an AI companion before experiencing either sudden termination or advance notice. Results indicated that fore-warning significantly reduced the sense of loss (ANCOVA: 𝑝 = .011, 𝜂2𝑝 = .252), with the most pronounced effects observed in levels of sadness. Crucially, this intervention had no negative commercial impact, validating it as a low-cost ethical solution with no apparent short-term drawbacks. Users with high levels of parasocial interaction were particularly vulnerable to sudden termination but benefited most from forewarning (𝑑 = −1.55 vs. −0.81). We propose an “Ethical Offboarding Framework” based on these findings. Our study demonstrates that AI ethics must encompass relational endings; “compassionate AI” requires designing not just how agents arrive, but how they leave.

인용 정보

BibTeX
@inproceedings{10.1145/3772363.3798687,
author = {Kim, Gahui and Choi, Yebom and Kim, Yoojeong and Lee, Changjun},
title = {No Time to Say Goodbye: Emotional Loss Responses to Sudden Termination in Immersive AI Interactions},
year = {2026},
isbn = {9798400722813},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3772363.3798687},
doi = {10.1145/3772363.3798687},
abstract = {As AI companions evolve from mere tools into relational partners, millions of users are forming deep parasocial attachments. However, the industry currently lacks ethical protocols for termination, often treating the end of these interactions as purely technical events. This oversight can lead to “digital grief” when companions vanish without warning. We conducted a between-subjects experiment (N = 26) to test whether forewarning mitigates user distress. Participants interacted with an AI companion before experiencing either sudden termination or advance notice. Results indicated that forewarning significantly reduced the sense of loss (ANCOVA: p =.011, (eta ^2_p=.252)), with the most pronounced effects observed in levels of sadness. Crucially, this intervention had no negative commercial impact, validating it as a low-cost ethical solution with no apparent short-term drawbacks. Users with high levels of parasocial interaction were particularly vulnerable to sudden termination but benefited most from forewarning (d = −1.55 vs. − 0.81). We propose an “Ethical Offboarding Framework” based on these findings. Our study demonstrates that AI ethics must encompass relational endings; “compassionate AI” requires designing not just how agents arrive, but how they leave.},
booktitle = {Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems},
articleno = {486},
numpages = {5},
keywords = {AI companions, parasocial relationships, smart glasses, termination design, digital grief, ethical AI, forewarning, immersive interaction, emotional attachment},
location = {
},
series = {CHI EA '26}
}

키워드

AI companionsparasocial relationshipssmart glassesterminationdesigndigital griefethical AIforewarningimmersive interactionemotional attachment

논문 정보

유형
Conference
게재지
CHI EA '26: Proceedings of the Extended Abstracts of the 2026 CHI Conference on Human Factors in Computing Systems
Pages
1-5
출판 연도
2026
교신저자
Changjun Lee