Philosophers are bad at formulating questions and theories

Philosophers often claim that although they may be bad at finding answers, they are good at asking questions. Well, if a (non-mathematical) question cannot be answered empirically, then it is a badly formulated question. The vagueness and untestability of philosophical theories reflects this. There is nothing rigorous, precise, or clear about analytic philosophy.

Edward O. Wilson wrote that his goal is to turn "as much philosophy as possible into science" (Consilience, 1998). For a naturalist, there is no other option. The only way to justify a philosophical thesis is to turn it into a scientific hypothesis and provide empirical evidence for it. However, this requires heavy modification, often rendering the original thesis unrecognizable. Perhaps some philosophical theses are already more compatible with science than others, but they are still too vague and slippery to be decisively confirmed or falsified by empirical evidence. As Timothy Williamson notes here, there is always some wiggle room in philosophy:

"Some metaphysical theories so informal that it is quite unclear what they entail. Whenever an opponent claims to draw false consequences from them, a proponent has the option of denying that they really follow. Sometimes metaphysicians seem to reserve the right to make up their theory's consequence relation as they go along. That has the advantage of rendering their theory hard to refute, but the disadvantage of undermining its 'predictive' power." (Modal Logic as Metaphysics, 2013)

This is why scientists don't like it when philosophical theses and dichotomies find their way to science:

"the dichotomy between associationist versus cognitive models (or 'statistics versus rules'), along with the 'learned' versus 'innate' dichotomy, are pre-scientific intuitive oppositions that are ill-defined, and thus ultimately untestable, and pernicious to empirical inquiry.

...

the history of science provides abundant evidence that science progresses faster when embracing crisp, clear models, even incorrect ones, than by accepting insightful but fuzzy metaphors that include partial truths, but are resistant to disproof.

...

debates employing these traditional dichotomies foreground irrelevant and ultimately philosophical issues, while downplaying or ignoring the empirical issues of central relevance." (W. Tecumseh Fitch: Toward a computational framework for cognitive biology: Unifying approaches from cognitive neuroscience and comparative cognition, 2014)