The ability to identify a problem spontaneously (for instance, identifying and understanding that there is a problem in politics), having the necessary logic to look at the problem from multiple points of view, realizing things about a variety of issues (scarcity v private ownership, self interest v selflessness, freedom v the common good, etc.), realizing that maybe there is no 'right answer', but trying to do a better job than most media folks of identifying the issues, and articulating some ways that we could adjust our politics that would make things better.
But the computer must be intelligent enough to identify the problem (self directed instead of programmed) and come to it's own conclusions. An AI should be capable of working in the abstract, thinking philosophically, and understanding all sorts of things. Just because I can calculate doesn't mean that a device that calculates or simulates things is intelligent. It has no understanding, and therefore has no actual intelligence. Frankly, I don't think this is something computer science will ever overcome, and I'm OK with that.