Even if you can predict halting (which for the large class of decidable problems you can) you still have to run the computation to see the results.

To me it's more about the expressiveness of languages.