this icon 
is arguably an abstraction of an airplane, but tells us little about how airplanes work
this icon 
doesn’t tell us how brains work & is not constrained by how brains work
deep nets may prove to be more like icons than bona fide, empirically useful neural modelshttps://twitter.com/timos_m/status/1200429999251169283 …
-
They use a deep net to predict neural responses in the visual cortex. (Yamins et al., 2014)https://www.pnas.org/content/111/23/8619 …
and the same team has also noted differences from human performance in later work and eg silhouettes, line drawings etc not considered one thing to show that CNNs moderately correlate on some task, another to show they are true or even useful model of actual underlying process