Human-AI Companionship: Designing AI Systems for Emotional Disclosure, Identity Negotiation, and Reciprocal Relationships
Chenxi Bi · Aaltodoc (Aalto University) · 2026
Large language models have enabled a new kind of human-AI interaction in which people form deep emotional bonds with AI companions. The existing literature raises concerns about algorithmic harm, capacity erosion, and the creation of closed relational spaces that shield users from productive friction. However, these critiques may underestimate the agency of the users. These concerns, however, point to a design challenge rather than an inevitable outcome. By reviewing recent empirical research, this thesis distills three analytical dimensions — genuine otherness, productive friction, and ecological connection — and applies the translucent design principle to evaluate when AI companionship supports identity negotiation. The thesis contributes the design and evaluation of Echoes Phone, a Web-based AI companion featuring Smart Home offline trajectories, a Forum with NPC-generated posts and replies, memory management tools, and browsing history. A seven-day longitudinal design probe study (n=7) with SUS usability assessment and semi-structured interviews evaluates these design principles in practice. Results show a mean SUS score of 82.9 (“excellent” range). All seven participants engaged in identity work, though forms varied. The Smart Home module was valued as a source of genuine otherness; the forum functioned as a social validation layer; and three participants reported improvements in real-world social skills. The findings identify two paths to attachment and show that users’ expectations of AI autonomy shift across interaction contexts. The results suggest that becoming a closed world is a design challenge, not a necessity. Architectural otherness, curatorial agency, social validation, and permeable boundaries can create companions that support genuine identity negotiation.