I agree that AI's visual recognition is a useful element, but it is far from being the decisive condition for a robotic hand to delicately grasp an apple.
This reversal of cause and effect is an interesting test, but it neglects physical limits.
A robot can identify an apple, but without ultra-sensitive pressure sensors and motor control capable of micro-adjustments, the apple will end up in compote.
It's like a landlord identifying a tenant on a lease; that doesn't guarantee the rent will be paid without other protective clauses and diligent follow-up.
The ability for delicate manipulation is paramount, not just identification.