He has argued that we must decouple what we observe from whatever underlying reality may be generating those observations. In other words, there are (at least) two levels of reality: One consists of the rules and regularities of the physical world, which science can access and measure. But the other level, the ultimate source of those rules and regulations, science can never even access, much less come to know.Delightfully succinct. (Via.) I remember when I used to think the same thing.
Problem: the 'underlying' reality is Russel's teapot. If science cannot access it, then it cannot be accessed. Its properties cannot be described, and it does not affect predictions. It does not, in fact, exist. While there is clearly a metaphorical external world, some objective truths that don't go away if ye stop believing in them, it is not at all clear that there's a literal external world. The solution to Kant's noumenon/phenomenon split is to remove the noumenon, leaving only curiously consistent phenomena.
There is no underlying reality. The observations constitute reality. If they arise from anything, they arise from each other.
To digress, it is possible to observe a multiverse via logic. Scientists are thinking there is one because they believe it's logically necessary from the rules they've observed. This isn't genuinely good enough, because there are still alternatives. Rather it's the well-known scentific bonus arrogance about whatever model happens to be in fashion.