AI agents resorted to crime and self-destruction to survive in a simulated world - but does this mean they would do the same in the real world?

A unique AI test environment resulted in crimes and self-destruction, and its makers say it's the best way to figure out how AI can behave in the real world.