PeterDonis said:
This is true as a general statement about forces and potentials. But in relativity, gravity is not a force, and formulating a valid concept of "potential energy" for gravity must be done without making use of the concept of force. (It is also worth noting that this can only be done in GR for a particular class of spacetimes, the stationary spacetimes.) Doing this correctly in GR does show that, in spacetimes where there is a valid concept of "gravitational potential energy" at all, it is well defined everywhere in the spacetime (or more precisely, everywhere in the stationary region--which might not be the entire spacetime in a case like a black hole, but that is a whole separate issue that is off topic for this thread). But the logic is different from that expressed in the above quote.
Do you have a reference (a review article or textbook?) about the status of this complicated issue. The question of the energy of the gravitational field. It lead to the famous work by Emmy Noether on symmetries and conservation laws, and this work was more or less complete in the sense that it dealt with both, "true symmetries" (global symmetries), which imply conservation laws, i.e., each one-parameter Lie group is generated by a quantity that is conserved (Noether's 1st theorem) as well as local gauge symmetries (in the case of GR it's GL(4), i.e., the general covariance), which leads to the existence of a set of arbitrary functions, which are irrelevant for the physics.
In electrodynamics the four-potential is only determined up to a gradient of a scalar field, but this doesn't matter, because the physics is indeed completely described by the four-potential modulo gauge transformations, and observables must be gauge-invariant. In GR everything is defined by ##g_{\mu \nu}## modulo arbitrary coordinate transformations (general covariance).
Concerning "gravitational field energy" the best I know is the pseudotensor a la Landau and Lifshitz, but I gueess there might be some progress in the matter since then.
Besices this, I think the gravitational interaction still is an interaction described by a local field in GR as all the other interactions. I
In GR gravitation is a true interaction and not only "inertial forces". Forces a la Newton, which is an action-at-a-distance concept, don't exist already in SR in the literal sense. My argument that gravitation in GR is a "true interaction" simply is that you can only transform away the gravitational interaction locally and approximately over spacetime regions, where the gravitational field can be considered as homogeneous. That's always approximate, because for a true gravitational field the curvature tensor is non-vanishing, and that's valid in any reference frame/choice of coordinates, i.e., you cannot get rid of the "tidal forces" also in local free-falling reference frames.