On the theoretical limits of machine creativity

We confidently ascribe cognitive powers to others without inspecting the neural substrates on which they depend. Yet those sceptical of the powers of machines focus their criticism on the underlying mechanism, not the manifest behaviour. It is argued the most complex forms of cognition, such as creative imagination, are outside the reach of any machine, for they definitionally lack a mechanistic constraint on the space of possible solutions that a machine supposedly needs. Here I show this criticism misconstrues both the nature of human beings and the nature of machines. Since the human brain cannot operate without conditioning, it is not free from constraint; nor can its constraint be guaranteed to be less than that of an equivalently trained machine. Equally, contemporary machine agents operate under constraints of far greater complexity than any human being could intuitively grasp, vitiating the claim they are simply following intelligible mechanical rules. If no theoretical limit is sustainable then the question of whether or not a machine could match or exceed human creative powers must remain open and empirical, to be settled by behaviour alone.