I disagree with OP. I think the phrase "rocket science" is still valid as a stand-in for the performance of a complicated task with high accuracy. In terms of the effort required to create a useful prediction, I agree that many things are more complicated than rocket science (often by virtue of being intractable). However, in almost all of these things success is not as highly correlated with the performance of some mental/engineering feat. If a rocket correctly performs a non-trivial task (i.e. gets to a specific orbit/planet/location), we know with certainty that the rocket scientist created a valid physical model and performed difficult calculations on it with high accuracy (ditto for other parts of the engineering). Testing rockets is expensive and the likelihood of making one that works by trial-and-error is zero (I'm referring to rockets on the scale that put satellites into orbit, launch probes thorough the solar system, etc). In comparison, common wisdom holds that startups usually launch with poor product/market fit and pivot until they find something that works. Eventual success doesn't correspond nearly as well to the performance of mental acrobatics.
Don't get me wrong, both tasks are difficult, require "guts", and so on. But I think a direct comparison of the difficulty is disingenuous. To me, "This isn't rocket science" doesn't mean "this is easy", it means "we should be able to eventually succeed by trial&error".
@pja: your observation that "not rocket science" only became popular in the 90s jibes well with associating "rocket science" with "low margin for error", as challenger exploded in 1986.