The use of artificial intelligence and more precisely of Convolutional Neural Networks for
the morphological classification of galaxies has been shown to be extremely reliable. In the
last years, detailed kinematic studies of local early-type galaxies have revealed that a significant fraction of them seem to be rotationally supported. However, they seem to present no
morphological signatures in their stellar morphology. In this work we apply machine learning
to investigate the relation between kinematics and morphology of early-type galaxies. We use
to that purpose SDSS imaging and spectroscopy from Manga together with simulated galaxies
from the TNG100 cosmological simulation. We show that Neural Networks can distinguish
fast and slow rotating early-type galaxies with 78% accuracy based exclusively on their stellar
morphology even if no apparent signatures are visible to the human eye
