How can humans maintain control over AI — forever?

I wrote my first real artificial intelligence program on punched cards 45 years ago. Since then, I have worked mainly on improving the capabilities of AI systems. My goal, like that of the founders of the field, was to realize general-purpose AI — that is, AI systems that match or exceed human capabilities across the full range of tasks to which the human mind applies itself.

Like anyone even casually acquainted with science fiction, I have also been aware of the possibility that AI systems could threaten human supremacy. My textbook “Artificial Intelligence: A Modern Approach,” first published in 1994, even included a section called “What if we do succeed?” I was far from the first AI researcher to consider the possibility that we might regret success in AI. Alan Turing, the founder of computer science, wrote in 1951, “It seems probable that once the machine thinking method had started, it would not take long to outstrip our feeble powers. … At some stage therefore we should have to expect the machines to take control.” This warning was largely ignored because until recently, success in AI seemed a very distant prospect.

By around 2013, I became convinced that success was less distant and that neither the AI community nor society at large were paying enough attention to its consequences. In fact, the issue was possibly the most important question facing humanity. I began giving talks in which I explained that the arrival of general-purpose, superintelligent AI is in many ways analogous to the arrival of a superior alien civilization but much more likely to occur. The messages of impending arrival were piling up in humanity’s inbox from the alien civilization and humanity was sending back an “out of the office” autoreply, with a smiley face attached.