It really sounds like a software issue to me. If the watch is causing atrial fibrillation to be detected too often, then you can either make the software meet some greater threshold before alerting the user.. or, even better, determine more fully formed diagnostic criteria to be applied.

The fact that medicine hasn't studied whether or not healthy hearts are regularly doing this because sensors were inconvenient previously, doesn't mean we need to get rid of the sensors. It means that both the presentation of the software should have some sort of feedback loop from clinical research that should emerge as a consequence of that data now... existing.