Testing AGI theories is necessary to determine whether they are good or bad explanations, but the challenge is to create a testable theory that can be criticized or improved. Building an AGI that makes no mistakes is akin to caging humans of a certain kind.