A compiler can only catch syntax errors (ie grammar). But just as a well-written essay can be based on false-premises, so a syntactically correct program can be

A compiler can only catch syntax errors (ie grammar). But just as a well-written essay can be based on false-premises, so a syntactically correct program can be full of bugs, particularly if the programmer had a faulty idea of what the problem was to begin with. Pure logic can only work correctly with reasonable premises. But it can't provide us with those premises. We can only construct, refine, and share those through natural language. To get back to the point of the OP, I think the underlying assumption that systematic modelling is the new literacy only makes sense if we assume the "old" literacy. In other words, the ability to model the real world only works if we can actually understand and communicate about that real world outside the models. Without the underlying meaning, the model is just a toy.