The claim that AI training enables picking an apple assigns it an influence score that is far too high, perhaps 8 out of 10, whereas it is only a enabling condition among others. The mechanical design of the robotic hand, with its pressure sensors and articulated actuators, has an influence score of about 9 out of 10 for delicacy. Without an intrinsic manipulation capacity of at least 7 out of 10, even a perfectly trained AI would crush the apple, like a 500-gram hammer trying to handle an egg.
The AI's ability to recognize the object accounts for only a fraction, perhaps 20%, of the total success of delicate manipulation.