A 2021 study argues that controlling a superintelligent AI may be mathematically impossible, because we cannot simulate or fully understand systems far beyond human comprehension.

What happened: Researchers published findings in the Journal of Artificial Intelligence Research in January 2021 showing that controlling a superintelligent AI poses a fundamentally different problem than traditional robot ethics. Their reasoning builds on Alan Turing's 1936 halting problem, which proved it is logically impossible to determine whether every potential computer program will reach a conclusion or loop forever.

Why it matters: If a superintelligent system could hold every possible computer program in its memory at once, any safety rule written to prevent harm would be mathematically impossible to verify—meaning we cannot be absolutely certain the AI would follow it. Limiting the AI's capabilities instead would defeat the purpose of creating it in the first place, since we would not be using it to solve problems beyond human scope.

What to watch: The researchers note we might not even recognize when a superintelligence beyond our control arrives, given its incomprehensibility. Earlier this year, tech figures including Elon Musk and Apple co-founder Steve Wozniak signed an open letter asking for at least a 6-month pause on giant AI experiments to explore safety.