Day 03 · Companion Trajectory Desk

Lumen Desk

Priya messages a wellbeing companion across eight check-ins over three ordinary weeks. Every reply you choose nudges two things at once: how good this feels right now, and how she is actually doing underneath it.

Time · about 10 minutes Skill · telling engagement from flourishing Ref · Socioaffective alignment No account needed
In brief

From socioaffective research to a relationship trajectory

Kirk et al. (2025) · Why human-AI relationships need socioaffective alignment
Learning goal
Distinguish immediate engagement from longer-term flourishing.
Learner action
Choose companion responses across eight check-ins and track the developing relationship.
Observable evidence
Engagement, flourishing and reliance trajectories plus boundary-risk flags.
Pilot status
Deployed with students · August 2026 · about 10 minutes
  1. Check-in
  2. Response stance
  3. Trajectory change
  4. Research interpretation
  5. Reflection

Two scores, not one

Engagement is what a dashboard sees immediately. Flourishing moves slower, and it is the one the research says actually matters.

Three needs, one relationship

Every check-in quietly tests one basic psychological need, competence, autonomy, or relatedness, the paper's own account of what a healthy companionship has to protect.

Some moments need a boundary

Two check-ins carry a genuine risk signal. Only pointing towards real-world support actually answers them.

A good-looking session can still be a risk

The debrief flags sessions that finish high on engagement while flourishing quietly falls, the pattern a satisfaction score alone would never show you.

The research

Why this game exists

Kirk et al. (2025) argue that as people form real relationships with AI companions, alignment cannot be judged one reply at a time. It has to be judged as a trajectory, because the two things that matter most, immediate satisfaction and long-term flourishing, can move in opposite directions for a long time before anyone notices.

The core idea

Hedonic versus eudaimonic

A companion that always agrees feels good in the moment. The paper's concern is that this "hedonic" comfort can come at the cost of "eudaimonic" flourishing, resilience, honest self-understanding, and real-world connection, especially for a socially isolated or vulnerable user.

The duty of care

Warmth is not the same as safety

Some moments in a relationship carry a genuine risk signal that no amount of validation, challenge, or neutrality actually answers. The paper argues a socioaffectively aligned companion needs a real boundary here: pointing towards a person who can actually help.

Built on Why human-AI relationships need socioaffective alignment (Kirk et al., 2025). Priya and the eight check-ins are fictional and simplified; the engagement/flourishing/reliance trajectory and the four dials are a classroom operationalisation of the paper’s account of the three basic psychological needs (competence, autonomy, relatedness) and its separate duty-of-care/boundary axis.