With a deep neural network it's possible to convert a camera signal into the movement trajectory of a dexterous robotics hand. In contrast to previous attempts in teleoperation, which are working with a dataglove or a joystick, the visual shape of the hand controls the robot. The human operator uses it's own hand to make a hand sign and the Shadow hand is doing the same action.
In this setup, a human operator is in the loop, which means, it has nothing to do with Artificial Intelligence. A real AI system would control the robot with software, a motion planner and can grasp objects by it's own. If the sensor signals are taken from a human, the system is only semi-autonomous. Does this make sense if the aim is to program autonomous robots? Wouldn't it be better to program an AI which controls the hand autonomously?
 Li, Shuang, et al. "Vision-based teleoperation of shadow dexterous hand using end-to-end deep neural network." arXiv preprint arXiv:1809.06268 (2018).