It came from economists asserting the difference between modeling human behavior vs. the behavior of physical objects. In that sense it is not meaningless. It's an important distinction that in order to model human behavior, we have to accept that people's goals are myriad and constantly shifting, and that we can only come to know their goals or preferences by them being revealed in action.
Ah, but that's assuming something very much in debate. Humans are actually quite predictable in many ways, though not as well as springs (currently). Aggregate behavior can often be predicted to reasonable accuracy, at least as well as with other complex, non-human systems like the weather. And even individual behavior can be predicted with enough information (and neuroscientists can even predict some specific decisions seconds in advance, given the right instrumentation). Just knowing some demographic and contextual information about a person hugely reduces your error on predicting what their "choices" will be.
It's a little different because the weatherman isn't forcing his predictions on you.
The debate takes on a different dimension when you're talking about political policy, when the weatherman passes a law forcing everyone to wear raincoats on days of high chances of precipitation.
