Precisely, if we restrict fMRI to investigating phenomena and theories of cognition and the mind that are plausibly measurable at the appropriate temporal resolution, it will potentially start yielding some fruit.

It will also require fMRI researchers to think more carefully about their theories as well (e.g. noting the speed of the mind / amount and kinds of thinking involved in certain tasks, and being realistic about whether or not fMRI could actually capture something meaningful there). Too often there is no theory, and too many studies are just correlating patterns with some task without actually carefully thinking about the task and deconstructing the components, testing activations in those (e.g. ablation studies in AI research) and etc.