Not quite, I think. Some kinds of redundancy are good, and some are bad. Good redundancy tends to reduce mistakes rather than introduce them. E.g. there's lots of redundancy in natural languages, and it helps resolve ambiguity and fill in blanks or corruption if you didn't hear something properly. Similarly, a lot of "entropy" in code could be reduced by shortening names, deleting types, etc., but all those things were helping to clarify intent to other humans, thereby reducing mistakes. But some is copy+paste of rules that should be enforce in one place. Teaching a computer to understand the difference is... hard.
Although, if we were to ignore all this for a second, you could also make similar estimates with, e.g., gzip: the higher the compression ratio attained, the more "verbose"/"fluffy" the code is.
Fun tangent: there are a lot of researchers who believe that compression and intelligence are equivalent or at least very tightly linked.
