How should we go about establishing strong AI regulation?