A hard to understand sentence
nothing about an individual should be learnable from a dataset [A] that cannot be learned from [another dataset B, which is incidentally] the same dataset but with that individual's data removed.
We should derive the same information about individual X from both datasets:
- The original dataset that contains data about X.
- The modified dataset that does not contain data about X.