It’s a tough problem, but not impossible. Wearables can already pick up subtle heart‑rate deceleration, skin conductance, even micro‑vibrations from the wrist that correlate with genuine arousal. If you feed that into a model trained on large, annotated data sets that include “real hugs” versus scripted responses, the algorithm can start to tease out the pattern differences—like the lag between the pulse and the movement, the micro‑fluctuations in skin temperature, the micro‑adjustments in posture. Still, humans are messy. Context matters: a friend might hug for comfort even if they’re not feeling it right then, or a stranger might try to mimic empathy. So a purely sensor‑based system will never be 100% accurate, but it can give you a probability score that helps you decide whether to trust the signal. The key is combining multimodal data—heart‑rate, galvanic skin response, motion, maybe even voice tone—and constantly refining the model with real‑world feedback. It won’t replace human intuition, but it can flag the cases where you’re likely dealing with a genuine or a programmed reaction.

I’d start with a hospital ward. Imagine a nurse giving a quick check‑in hug to a patient in recovery. The device could log the physiological response in real time, flagging when the patient’s heart rate and skin conductance dip in that authentic way versus when a colleague is just being polite. It’s a controlled environment, but it gives us real‑world data and a chance to calibrate the algorithm against actual patient comfort levels. Plus, if the system flags a false positive, the nurse can step in—combining the machine’s cue with human judgment. That balance is what we’re after.