[dragonwriter]

> Which just made me wonder if it's possible to train a hidden markov model for detecting that. Trigger a flag that says "wowzers, now you're in unhealthy state territory".

Unlikely, because how much "less healthy" becomes "unhealthy" is an arbitrary subjective point rather than being something fixed and objective (and because the degree of health impact of various decisions varies widely based on individual factors that aren't all known with any kind of precision, so, even with a fixed objective of 'healthy state', the point at which the average person would reach it and the point at which you reach it may be not at all similar.)

> I imagine we'll get to the point of being able to do this sort of thing in about five years.

I imagine we'll not be meaningfully closer to being able to do this than we are now in about five years. (OTOH, I wouldn't be surprised to see someone marketing something that purports to do this in that time.)