How to control AI systems much smarter than us
I think this is a very important question that we should address. For example, what does it mean that AI system is much smarter than us?
There are a lot of ways to be "much smarter". Sometimes a person comes up with a solution that we see as "genius". We saw this with AlphaGo when it beat Lee Sydol. AI was able to figure out new strategies in Go by playing millions of simulated games and studying all the other games. We feed it enormous amounts of Data and it is able to spot patterns. We can do the same thing but can't digest as much data as AI can.
But there are other ways. Sometimes being "much smarter" is simply more observant to the little things around us. This kind of smarts is illustrated in the TV show Monk where he notices small little things. It is also illustrated in shows like Eagle Eye and Person of Interest about the overwhelming superiority that AI has when it comes to observing every single person on the planet. In Eagle Eye they go into a secure room to try and keep AI from listening and it looks at the ripples in the coffee cup to determine what is being said. AI can observe every email, text, phone message, convert every phone to a bug, track people by their phones or cars, and tap into every camera feed.
Sometimes being smarter is simply being more diligent. If you have a million lines of code you need to go through to find something it requires a great deal of diligence and persistence to do that. AI has the computing power to do that.
So then what we need is a system of checks and balances. Have three different AI's that do not communicate with each other look at a problem. Suppose one of them comes up with a move that appears to be "genius". Ask the other two AI systems to evaluate the move and determine possible faults and negative outcomes. Then if they do come up with something run that by the first AI to see if it had considered that and if there is a solution.
If one AI notices some small little features or observations that might be critical, ask the other two AI systems if it agrees.
Also evaluate the lines of code of each AI using different AI systems. Yes, it AI becomes very smart let's use different and independent AI systems as a check and balance to the power.
You don't want to be beholden to any one system. If an AI appears to be manipulating people for some nefarious end you want to be able to immediately shut it down, cut it off, and investigate. Also, you don't want the AI to simply be calling the shots. It suggests a path to take you want to be able to come back with relevant and well thought out considerations and concerns.