If one just sends photons through a narrow single slit, then the pattern that builds up on the screen (if you send multiple photons, and record their positions) will be a banded diffraction pattern.

If you have two slits, with a detector to determine which slit the photon went thru, then it'll behave as if it only went thru one of the two slits, at random, and what'll build up on the screen will be the two (slit A + slit B) overlayed diffraction patterns.

Finally, if you have two slits with NO detector, then what will build up on the screen is the interference pattern as if the photon had gone thru both slits simultaneously and the two resulting banded diffraction patterns interfered with each other. So, what SEEMS to be happening in this case is that the quantum state of the system post-slit is that of the photon simultaneously having gone thru both slits, each slit having diverted it per diffraction, and then these diffraction patterns (probabilities) interferering. Wave collapse can only be happening after this interference (if it was before then there would only be one diffraction pattern and no interference), presumably when quantum state interacts with the screen.

So, yeah, it seems that the "photon" does "go" through both slits, but this is a quantum representation, not a classical one.