Walt Mossberg, writing for The Verge, on Apple’s push for AI with privacy and their continued role as a hardware-first company.

On “differencial privacy”:

This works by randomizing local data from each device by injecting it with meaningless “noise,” or nonsense data, so it can’t be traced back to the device’s user. Then, this information is mingled on Apple servers with very large amounts of similarly randomized data from many, many other devices. To further protect individuals, Apple is also imposing a “privacy budget” that will prevent too much data — even when randomized — from getting to the server from any one user.

Even though the data is no longer precise due to the randomization, Apple says it still displays a “slight bias” toward trends that would be valuable for the software to learn, such as new slang words.

You may wonder why Apple didn’t simply use aggregated anonymous data, without jumping through these hoops. But, in fact, anonymous data can, and has, been cracked to reveal certain individual information.

On “hardware first”:

The best example of this was that it again declined to extend its much-loved and much-used iMessage messaging system to Android, even though Google still seems vulnerable in this area. Apple did announce a clutch of new features for iMessage, like giant emojis, and handwritten texts. And it’s turning the service into a true platform that can host third-party apps like cash transfer services, stickers, photo editing, and restaurant reservation apps. But all of this seemed more about keeping people on Apple hardware than about building the biggest possible services.

When I asked a senior Apple executive why iMessage wasn’t being expanded to other platforms, he gave two answers. First, he said, Apple considers its own user base of 1 billion active devices to provide a large enough data set for any possible AI learning the company is working on. And, second, having a superior messaging platform that only worked on Apple devices would help sales of those devices — the company’s classic (and successful) rationale for years.

To me, Apple’s strategy is a tiered class system. SMS text messages are the lowest tier, a common currency. Everyone can send and receive text messages, no matter the platform. SMS is vanilla, no bells and whistles. SMS messages appear in green in the Messages app, a clear signal that they are outsiders in the soothing blue sea of iMessage uniformity.

The iOS 10 beta brings a new tier of messages, with new looks and behaviors. As an example, “bubble effects” allow you to customize the actual delivery of your message with animations that convey a specific mood or meaning. But just as official blue iMessages can’t be sent to Android users, those bubble effects’ messages can only be sent to iOS 10 users.

[As a side note, with this first beta, trying to send a bubble effect to a non iOS 10 device causes the send to fail. I suspect that is purely early beta behavior and that, ultimately, the effect will be stripped and the text itself sent on.]

As Walt points out, this only helps Apple to sell hardware. Want bubble text messages? Buy an iPhone, iPad, or Mac.