So my argument for this is that the expansion of knowledge in any field of physics depends on what is previously known, and what we physicists express or knowledge as equations. So here's what I imagine it would look like:
For example, lets the field of quantum mechanics:
Express all the governing laws and theories of quantum mechanics (from uncertainty to Klein-Gordon relations to conservation laws). Each variable must explicitly be expressed as it appears in every other equation you use, and it would obviously need to be in a language capable of solving differentials and other complex mathematical methods.
For each solved variable, use substitution from other equations, and perhaps by this, new equations are formed.
Since we know not all equations are distinctly general, "new" equations need to be checked; the computation to get to the new solution must be valid.
This must obviously be done via supercomputer since there is a large body of equations, and rules for upper level mathematical methods would have to be expressed as well. But basically, by brute force, we could program to solve all the entangled equations as a system, thereby, leading to possibly new theories based on existing knowledge.