Hur kan adaptiva och affektmedvetna system stödja lärande och välbefinnande utan att övertolka glesa, intima signaler?
My doctoral research sat here: machine learning on wearable sensor data to monitor the mental health of frontline healthcare workers. The useful finding was as much about limits as about accuracy. Signals from the body are sparse, noisy and deeply contextual, and a model that performs well on a cohort can still be the wrong instrument for an individual decision.
The same tension runs through adaptive learning systems, which I worked on earlier in education settings: personalisation is only as good as the model of the learner, and that model is always thinner than the learner.