It's not just a question of what you can observe, and it has nothing to do with chaos. If there is a hidden but definite state of the particle, you get different predicted results than if the state is genuinely indeterminate, as long as the hidden state is local to the particle, that is, that is does not get to engage in faster-than-light communication.
This kind of problem is usually discussed using the example of particle spins, rather than momentum and position. Take a pair of electrons (which each have a fixed amount of spin), emitted back to back so that their spins have to be oriented in opposite directions, to conserve angular momentum. You can measure the spin of one electron in the vertical direction or in the horizontal direction, but whichever way you measure it, if you measure the other electron in the same axis, it will have the opposite direction (up vs down, or right vs left).
Quantum mechanically, the two axes (vertical and horizontal) have a Heisenberg Uncertainty Relation between them: if you measure along one axis, you cannot know anything about spin along the other axis. (If you measure the first electron vertically and the second horizontally, you have a 50% chance of getting a spin pointing right on the second measurement and a 50% chance of getting a spin pointing left.) In QM, the spin along the second axis is indeterminate (really random) once a measurement has been made along the first axis (even if the two measurements are made on different electrons in the pair).
Suppose there is actually a real, hidden spin direction that we cannot directly measure, and that this is why the two always point in opposite directions. This would explain the correlation between the two spin measurements, and give the same answers as the "really random" QM formulation -- as long as you only measure vertically and horizontally. If instead you measure one of the spins at 45 degrees, you will get a different predicted correlation between the two measurements for the hidden, definite spin than for QM. (This is the essence of Bell's Inequality.) So you can perform an experimental test to determine which prediction is correct. That's been done, and the QM prediction is correct and the hidden variable prediction wrong.
This conclusion is valid unless the first particle can communicate with the second (regardless of how far apart they are, and how close together the measurements are in time) to tell it what axis the experimenter chose to measure along, i.e. unless the hidden variable is non-local.