In e.g. the Copenhagen interpretation, observation plays a special role, because it's the moment of wave function collapse.
In MWI, there is no wave function collapse, so there's no need to identify certain special events called "observations" which are somehow different from the rest of physics. It's just our eyes, and brains, and instruments, interacting and becoming entangled with the other things in the world.
When we "observe" Schrödinger's cat, we entangle ourselves with the cat just as the cat was already entangled with the cesium atom. Now the world is in a superposition of two states: in one, the cat is alive and we see it as alive; in the other, the cat is dead and we see it as dead. This is exactly what the equations say should happen. Nothing special.
Even more interestingly, say you want to predict the outcome of a quantum experiment. In the MWI, that outcome is deterministic, and it is a superposition in Hilbert space. However, in your own space, you still need to compute a probability for that outcome.
What is this a probability of in the MWI? The concept of probability across worlds is not really rigorously defined, especially since you need to arrive at very specific results to be consistent with observations.
I agree that the meaning of probability in the MWI context is an interesting open question. Mathematically, it's the squared amplitude of part of the wave function. But why this corresponds to our personal experience of frequency is something I'm unclear on.
That's just one of the many problems that MWI fails to solve.