Apple and Privacy

Apple’s privacy stance has been in the news again this week, mostly because of a speech Tim Cook gave to an event in Washington, which honored him for his (and Apple’s) commitment to privacy and encryption.

Apple’s admirable but over-played privacy stance

My reaction to the speech has been somewhat mixed, as these two tweets indicate:

As I see it, Apple’s commitment to privacy is an admirable one, and one that provides a useful competitive vector as it seeks to differentiate itself against companies like Google and Facebook. But I feel that here, as when Tim Cook repeats the now-hackneyed phrase “when you’re not paying, you’re the product”, he’s overplaying Apple’s hand 1. There’s an underlying truth to both of these claims, but it’s not as cut and dried as Tim Cook makes it seem to be. And I believe that’s likely just positioning, or in other words making the point in the strongest possible black-and-white terms, even though Tim Cook (and the rest of Apple’s leadership) clearly understands that there’s more nuance to this in reality. It’s obvious that many people do value these free services, and are willing to make the tradeoffs inherent in them (if they understand them at all). If Apple really believed everything Tim Cook said in literal terms, I’d be worried about the company, but I don’t actually believe it for a minute, though I absolutely buy Apple’s commitment to privacy and its intention to continue to differentiate on this basis.

Privacy and machine learning

Ben Thompson, in his excellent Stratechery daily email (subscribe here) makes to some extent this same point in his email today, and I agree with those thoughts pretty completely. But he and others have also taken this point further and talked about a supposed downside to all this, which is that by refusing to collect personal data about users, Apple risks not being very good at machine learning. I actually think this point is false, because it conflates three different kinds of data collection and analysis:

  • Data collected on an aggregate basis to allow computers to determine broad trends, better understand text and speech across the entire base, glean information about searches and the best responses to them, and so on. There is nothing user-identifiable about this form of data collection.
  • Data collected about individual users to better customize services and products to their individual needs – i.e. learning favorite places, home and work locations, building patterns of searches to better interpret future searches, and so on. This kind of data collection exists on a spectrum, with some forms of data explicitly provided by the user and others easily inferred, with other data reliant on deeper analysis.
  • Data collected about individual users to build profiles which can be used to target advertising. The only benefit to the user from this form of data collection is making the ads they see more relevant, and the downside is a vague sense of creepiness that third parties are suddenly serving up ads which make use of quite private data from browsing, searches, or the contents of emails (and potentially photos).

The reality is that machine learning takes place across these three different types of data collection, but only the first two are primarily about creating a better experience for users, and Apple has shown itself to be perfectly willing to engage in the first kind with products like Siri and Spotlight, and quite comfortable with at least some forms of the second. It’s only the third category that Apple eschews and which Tim Cook appears to be criticizing other companies for in his public remarks.

Apple’s privacy stance

The table below summarizes my inferences of Apple’s privacy stance on these three categories:Screenshot 2015-06-05 09.37.14If this is accurate, and I believe it is, then Apple isn’t constrained at all when it comes to broad improvement of its services on an aggregated basis, because it’s clearly entirely comfortable with this form of data collection and analysis, and has even recently started crawling websites itself to further this effort. In the second category, it seems very comfortable with building basic profiles of its users through a combination of data users actively pass to its systems and a small amount of data inferred from behavior 2. As such, it’s capable of customizing certain of its services and willing to do so. One good example of this is the customization of its QuickType keyboard on a per-user basis, although Apple is careful to point out that “Your conversation data is kept only on your device, so it’s always private.” When Apple does customize services on a personal basis, it often keeps the information on the device, unlike competitors who use cloud services to make these customizations available across devices, which really does constitute a compromise on Apple’s part. 

However, it’s the third form of data where Apple really seems unwilling to engage in broad data collection and analysis, and that really doesn’t affect its ability to provide its users with compelling services. As such, I think it’s a stretch to suggest that Apple will somehow always be inferior at machine learning because it eschews targeted advertising – the two are for the most part separate, and it’s entirely possible for a company like Apple to engage in both broad based aggregate machine learning and machine learning on an individual user basis without either compromising privacy or engaging in the type of behavior it’s criticizing in others.

Apple still has work to do in machine learning

None of this is to say that Apple is just as good at machine learning as competitors. I honestly believe this is the core of Google’s differentiation as a company and Facebook seems to be becoming increasingly strong in this area too. Apple is currently weaker in these areas, but I don’t believe that its privacy stance is the reason – I just think it hasn’t chosen to invest in these areas as heavily, and to the extent that it is doing so now, it has some catching up to do (and it seems that at WWDC Apple may announce some advancements in this area relating to Siri).

The same applies to Apple’s cloud services in general – this simply hasn’t been a major focus for Apple so far, for a variety of reasons, and its cloud services simply aren’t as strong as competitors’. Google Photos is a great example of that – its seems to do certain things much better than Apple’s Photos product, and a large part of that is about machine learning. And yet Google Photos is also a great example of exactly what Tim Cook is talking about – the inherent unease about sharing such personal data with a company you know would like to use it to target advertising to you. If Apple produced a similarly compelling product, there would be none of that unease, but you’d likely pay for the privilege of using it directly. Therein lies the real difference between Apple and its competitors.


  1. For a more nuanced analysis of the latter claim, please see this post
  2. See this earlier post for an example of this, which Apple hasn’t shouted about much to date
  • Mark Jones

    Thanks for your post. I agree with your distinctions. I do agree with Ben Thompson to the degree that Apple’s privacy stance can slow or degrade Apple’s services – such as in the example you provide. Apple will have to figure out other ways to enable my devices to smartly share personal customization info with each other.

    One question: What is the future of Apple’s iAds product with this stance? Does Apple only use aggregate data for it?

    • Thanks for your comment, Mark. iAds are the one exception to Apple’s general aversion to ads, and Tim Cook referred to these specifically in his public privacy letter:

      “One very small part of our business does serve advertisers, and that’s iAd. We built an advertising network because some app developers depend on that business model, and we want to support them as well as a free iTunes Radio service. iAd sticks to the same privacy policy that applies to every other Apple product. It doesn’t get data from Health and HomeKit, Maps, Siri, iMessage, your call history, or any iCloud service like Contacts or Mail, and you can always just opt out altogether.”