Prioritize the use of AIs to empower humans to communicate and decide together, and it won't take over the world.
If we use the AI systems to primarily empower humans to decide together, then likely the AI won't pose the control issue, because they will be a tool that people use to empower the collective of people to decide, rather than an independent agent, that decides purely on its own. (See, agency vs tool.)
Among the examples of use of AI to empower human collective decision-making, are -- machine translation. Another example is the use of it to compress information for communication purposes (like in the idea of reflectoputer). The idea is basically to think of all other ways in which AI can empower us to make decisions together, rather than completely automating decisions.
This idea also integrates the ideas (video) by Louis Rosenberg, on Swarm Intelligence, and the idea that extremely powerful communication technologies must precede extremely powerful computation technologies, if we want to stay in control of AI.