My informal definition of "human-level AGI," is a system whose only mistakes are indistinguishable from mistakes a human could make, given the same context.