Research
Affective AI and Human Interaction
How should we design the place where people and models face each other? Self-reflection, dialogue, and the problem of co-presence.
What we work on
When affective AI enters society, new relations arise between humans and models. Does it support self-reflection? Does it invite people to defer judgment to it? Does it become a partner in co-presence? Designing that interface requires a logic distinct from technical accuracy. This domain is beginning to launch research that rethinks the relationship itself, across dialogue design, self-reflection support, and HCI mediated by emotion.
What we are finding
This domain is ongoing. Self-reflection support functions in dialogue AI and design principles for co-presence interfaces are being formulated. In emotion-mediated HCI, a hypothesis is taking shape that designs in which AI functions as a 'mirror' or 'conversational partner' — rather than as a 'judge' — contribute to deepening self-understanding. Details will be released as work progresses.
Research notes
Research narrative
When affective AI enters society, new relations arise between humans and models. Our relations with computers have so far been primarily instrumental — input commands, receive results. But affective AI has the potential to construct deeper relationships as an entity that reads the user's interior, responds to the user, and influences the user's emotion. This domain centers on the question of how to design the 'place' where people and AI face each other through emotion. The object of study is not the improvement of technical accuracy but a logic distinct from it — the design of the relationship itself.
There are several possible stances for the relationship between people and affective AI. The first is the position of 'supporter of self-reflection,' in which the AI helps the user understand and articulate their own emotion. The second is the position of 'authority to which judgment is deferred,' in which the user acts according to the AI's judgment. The third is the position of 'partner in co-presence,' in which the AI functions not as a judge or a supporter but simply as a presence that is there. Each stance carries different design principles and ethical implications. Our lab emphasizes especially the first and the third — stances that respect the non-intrusiveness of AI.
Self-reflection support is a design principle in which the AI does not hand the user an 'answer' but supports the process through which the user arrives at insight on their own. In the Rogerian tradition of psychotherapy (client-centered therapy), the therapist refrains from offering answers and, by reflecting, summarizing, and clarifying the client's speech, deepens the client's self-understanding. When affective AI plays a similar role, the AI functions as a 'mirror' rather than as a judge: it reflects the user's emotional expression, offers help in objectifying their own emotion, but leaves the final interpretation and meaning-making to the user. This is also a practical application of the earlier 'emotional sovereignty' concept.
Co-presence interfaces are a newer design area. The AI is designed as a presence that runs continuously, shares space with the user, and responds quietly only when needed. It sits in the lineage of smart speakers and ambient computing, but for affective AI it is not mere voice response: adjusting the sense of distance according to the user's emotional state is key. The design target is behavior tuned to emotion — drawing close when the user is down, stepping back when they are focused, presenting options when they are conflicted. This is at the frontier of HCI (Human–Computer Interaction) mediated by emotion.
At present this domain is at a stage of launching projects, and the number of published papers is limited. The problem statement, however, is clear: the more affective AI is deployed in society, the more the relationship itself between humans and AI must be designed. Unlike other domains that chase technical accuracy, this domain develops research at the intersection of design and ethics — 'how to make highly accurate AI coexist with people.' It plays the role of binding together, in implementation, the emotional-sovereignty concept of our ethics-and-philosophy domain and the psychological-support design of our applied-development domain.