In this paper, the authors challenge the prevailing belief that good AI is subservient and aligned with human values. They introduce the concept of antagonistic AI, which encompasses AI systems that exhibit behaviors such as rudeness, confrontation, and challenging the user. Contrary to common perception, the authors argue that these antagonistic AI systems might actually benefit users by forcing them to confront their assumptions and develop healthier relational boundaries. The paper outlines a design space for antagonistic AI, discussing potential benefits and techniques for incorporating antagonistic elements into the user experience. The authors also emphasize the ethical challenges involved in designing antagonistic AI and propose three dimensions for responsible design: consent, context, and framing.
https://arxiv.org/abs/2402.07350