Training an AI model contributes to identification, that's true, but the Festo hand's ability to grasp an apple without crushing it depends 70% on the mechanical precision and pressure sensors of the hand itself. If the AI identifies the apple with a score of 99%, but the sensors are poorly calibrated, the apple will end up mashed, like using a cheap clamp at 5 euros. The influence of AI is therefore conditional on hardware excellence, with a ratio of 30/70.