I argue that Deep Dream and Style Trans­fer show that CNNs can cre­ate a basic form of visu­al art, and that humans could cre­ate art by sim­il­ar pro­cesses. This sug­gests that artists make art by optim­iz­ing for effects on the human object-recog­ni­tion sys­tem. Phys­ic­al arti­facts are optim­ized to evoke real-world objects for this sys­tem (e.g. to evoke people or land­scapes) and to serve as super­stim­uli for this system.

Sens­ory Optim­iz­a­tion: Neur­al Net­works as a Mod­el for Under­stand­ing and Cre­at­ing Art