Here's the way I see it. Part of interpreting a contract is deciding whether certain facts are true or not. This part seems hardest for a computer to do (for now), so this part can still be delegated to humans. In the law these are called "questions of fact". On the other hand, "questions of law" are the part that I think can be executed by a smart contract (once the facts are input).
You're right that many contracts are not well specified. These are "bugs" in contracts. People will have to write bug-free smart contracts. There will be a library of contract templates (github-style). This stuff already exists with contracts - lawyers have software that spit out standard forms and fills in the name of their party. And we pay them $500/hour to do it.