These days, it does tend to mean doing global illumination using ray tracing.
Originally, the term “path tracing” was used to refer to taking a single random scatter at every step along a path of connected segments, as opposed to the idea of taking multiple random scatters at a point and averaging them, recursively. It’s a way of thinking of Monte Carlo rendering as taking one sample in a very high dimensional space - where a chain of ray segments, or “path” is a single sample - rather than thinking of each ray segment separately as a sample.
Ray tracing, FWIW, can sometimes refer to situations where you’re not even rendering. It has a more general meaning of doing line based visibility queries, which you can use for lots of things.
Ray tracing = initiate rays from light sources, bounce them around, accumulate them on camera. This is accurate but incredibly inefficient. This is how pre-baked light mapping and GI works.
Path tracing = initiate rays from camera, induce bounce rays toward light sources. Not quite as accurate but orders of magnitude mode efficient in terms of producing an acceptable image.