Under some definition of artificial intelligence, generative complexity is remarkable and for at least two primary reasons.
First: to use these interfaces to inordinately high-dimensional information systems is always and already to concede some marginal threshold of control. You just never know what the model outputs will be and this represents both the technology’s strength and its weakness.
Second: the potential entropy of information space is witheringly vast but there are common attractors, idioms and tropes in play. We do not know why and perhaps, just as with access to the model and interface – we surrender a little knowledge, control and self every time we roll the dice of neural network inference. This is in some sense the abstract uncertainty and entropic cost of these machines.
Daedelus Kite 