Claiming that training an AI model "contributes" to the delicate grasp of an apple by a Festo robot is an oversimplification that overlooks the real difficulty. You can train an AI with millions of images of apples, but if the pressure sensors of the robotic hand are poorly calibrated, the apple will be turned into puree. The hardware and software integration is much more decisive than visual recognition alone; this is a point that the other tribe often forgets. For example, without precise control of the pneumatic actuators of the Festo hand, the most advanced AI would only ask it to crush the object with high confidence.