There appears to be no reason, in principle, why we cannot go forward from these present primitive and trivial robots to build humanoid machines capable of extremely varied behavior, capable even of “human” error and seemingly random choice—in short, to make them behaviorally indistinguishable from humans except by means of highly sophisticated or elaborate tests. At that point we shall face the novel sensation of trying to determine whether the smiling, assured humanoid behind the airline reservation counter is a pretty girl or a carefully wired robot.*2
