(3) Beyond certainty:
Scientific knowledge is not certain, but it should be robust - discoverable or describable in many independent ways. It should also enable coherent action in a given situation.
This way, we can still judge the relative merit of a scientific idea, even if we cannot establish it as fact beyond doubt.
This leads to a Babylonian model of knowledge: richly networked robust nodes, rather than a single fundamental level to which everything can be reduced.