In this section
@inproceedings{Sadek:2026:10.1145/3816046.3816300,
author = {Sadek, M and Mougenot, C},
doi = {10.1145/3816046.3816300},
title = {Using Anti-Values as Conversational AI Design Constraints: Deciding on What We Shouldn't Build},
url = {http://dx.doi.org/10.1145/3816046.3816300},
year = {2026}
}
TY - CPAPER
AB - Human values are increasingly foregrounded in AI design, particularly through work on value alignment examining which values AI systems should promote and how this can occur. However, this focus leaves a critical gap in terms of explicitly articulating what AI systems should not align with. This gap is especially consequential for AI-powered Conversational User Interfaces (CUIs), whose relational, persuasive, and longitudinal interaction styles can give rise to subtle forms of misalignment which are currently not well-captured. In this provocation, we introduce the notion of anti-values as design constraints for CUIs. Anti-values are defined as normative commitments that conversational systems should not promote, embody, or optimise for. Drawing on prior empirical work in which CUI creators and users articulated anti-values unprompted, as well as related interdisciplinary lenses, we argue that anti-values offer a complementary and participatory way to reason about value alignment during CUI design. We outline how anti-values can inform design, evaluation, and governance, and present provocations that reframe CUI design as shaping a conversational action space through refusal and restraint.
AU - Sadek,M
AU - Mougenot,C
DO - 10.1145/3816046.3816300
PY - 2026///
TI - Using Anti-Values as Conversational AI Design Constraints: Deciding on What We Shouldn't Build
UR - http://dx.doi.org/10.1145/3816046.3816300
ER -