There are cryptographic ways to compute a set intersection between two sets A and B held by two different parties without A and B revealing the membership of elements of the set.
If set A is the set of users shown ads for Merchant M, and set B is the set of customers at merchant M who purchased something, then this intersection can tell you what percentage of the ads were shown to people who made a purchase without revealing who made the purchase to A or revealing who was shown an ad to B.
There’s probably a way to do it with differential privacy as well but it might be less efficient.
This is like the fourth time this story has hit the media. Google really needs to publish the full transparent details of their protocol and let the academic community analyze the, if only to avoid misinformation and speculation.
If set A is the set of users shown ads for Merchant M, and set B is the set of customers at merchant M who purchased something, then this intersection can tell you what percentage of the ads were shown to people who made a purchase without revealing who made the purchase to A or revealing who was shown an ad to B.
There’s probably a way to do it with differential privacy as well but it might be less efficient.
This is like the fourth time this story has hit the media. Google really needs to publish the full transparent details of their protocol and let the academic community analyze the, if only to avoid misinformation and speculation.