I could see this very soon if someone decides it's worth it, though it would require some entirely new format.
Programs like 3DOwl, although still very beta and filled with early-phase AI artifact fuckery flaws, already does a a similar job converting 2d footage to 3d.
To take existing 3d footage which already has some depth properties to it for AI to create a least a little 6dof effect seems even less difficult than converting a 2d source to stereo. Has more to work with. Maybe not enough to fake a full 360 panoramic walkaround, but about 15 degrees? Enough to tilt your head around a bit from normal, locked POV seems like a reasonable expectation.
The real problem would be creating that video format, essentially multiple videos layered over each other at the same time, which would be a real decoding bandwidth challenge unless each stored angle of perspective was really low-quality. Either that or it's gotta be rendered live.
