A perfect political AI would run a government perfectly. The problem is that perfect is really hard to define, and the stakes are quite high.
One approach might be a robot that follows rules. For example, it may be told to enforce the opinion of the majority. However, this is not as ideal as it may seem. The knowledge with which to make good decisions is not universal. The public is NOT educated enough to decide a location for a nuclear powerplant. The public is not educated enough to decide whether we should go to war. There isn't enough time in the day to be educated enough on all issues, this is why there is division of labor. Plus, in the case of war, some of the essential information may be classified, and not safe to dissiminate.
In general, rule based systems are hard to put in place well because you must ceede power to the rules. You lose the ultimate power the instant you engage the rules. If you made a mistake, well, rules are rules. You can add rules to change the rules but that only defers the problem. There is an infinite regression to deal with there.
On the other extreme, we can give the power to the AI to make decisions. However, how do we know those decisions will be in our interest before we enact the AI? There is more than one movie about a AI bound by the wrong rules. Some of them have even refused to open the pod bay doors. This ability to decide also opens the door for corruption. It will be in a slightly different form, rargetting an AId decisions rather than a human 's, but the effect would be the same
So what you want is a balance. The AI needs to have commitments to some rules, but the freedom to avoid catestrophic misinterpretations. You have to decide which things you want you AI to hold to while it is in office. You also have to get some agrements between all of the citizens. Everyone will have to have some say in what the AI will hold to.
Which is starting to sound a lot like campaign promises , don't you think?
Edit: from the comments, I mentioned Isaac Asimov's Three Laws of Robotics. 70 years after he wrote them, they're still considered some of the best rules ever written for a robot to follow when working with humans. For those who are not familiar with them:
- A robot may not injure a human being or, through inaction, allow a human being to come to harm.
- A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
- A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws
Asimov then spent an entire career writing books about the loopholes in these laws, and how each one could play out. This just goes to show how nuanced such a program would have to be.