There are too many variables thrown into each design for an automated solution to be of any practical usage. Also, except for the most minimalistic designs, the amount of options - and thus the amount of comparisons needed - will quickly become unmanageable to review (as maths has it).
What's more, many design decisions are made based on experience, knowledge (say, of cognition), familiarity with the specifics of the problem and the target users. It is impossible to replace all of these by an algorithm. Just consider the elements on the black stripe at the top of this page. There has been some thinking behind how these are organised, aligned, spaced, the colours, whether or not to include text or icons, arrows, borders, placeholders, etc.
There are, however, many aids that can help reach decisions and being manual they leave the control in the hand of the designer, who by the very virtue of attempts (conceptual designs) learns what works and what doesn't - the latter dramatically reduces the number of options.
One example for this is task models, where you create associations between content and actions and then simply have to segment the whole thing into containers, whether pages, panels, menus, etc. You can then use analytics data to further determine the size and position of various elements.
Network analysis is another tool that is largely ignored at the moment with relation to interface design, although I'm convinced that in years to come it will gain dramatic popularity. Although it would not output the actual design, it can provide highly valuable data on its performance. If anything is to become semi-automatic in the future, my bet would be on this square; but to get good results from network analysis you have to feed in good data - and there's plenty of it when it comes to UI design.