Last Week on My Mac: Trial ethics

Last week we celebrated the fifth anniversary of APFS, the file system which Apple secretly rolled out to millions of iPhones and iPads on 27 March 2017. It was also the week when we discovered that services introduced over a year ago in macOS Big Sur appear to conduct ‘trials’ and ‘experiments’ on our Macs. While Apple came clean about APFS just over two months later, it shows no sign of explaining what’s been happening on our Macs. Should it?

One big difference is the platform. iPhones and iPads, no matter how far you might push them, are consumer devices. They’re intentionally constrained to do the things that Apple deems they should. You can of course jailbreak them to run your own code, but the moment you do, you’re out on your own, no longer a consumer but a hacker.

Macs, on the other hand, are computers whose primary purpose is to enable, not constrain. Apple even encourages us by providing its pro development environment, Xcode, completely free through its own App Store. Terminal provides ready access to a huge suite of powerful command tools which far exceed anything needed by consumers. We’re encouraged to take our Macs to the limits, and those limits normally aren’t set by Apple.

If you’re still not convinced of the difference, consider what would have happened had Apple secretly rolled out the first release of APFS not on iOS devices, but on Macs instead, as part of macOS Sierra 10.12.4, which was released on the same day that the iOS 10.3 installer was busy converting all those iPhones to APFS. Instead of a gasp of breath and admiration of Apple’s boldness, as happened, there would have been outcry.

triald is different from the APFS rollout, but some of the same principles apply, as they do in much of life, because they’re part of ethics. The principle here is treating adults as adults, with respect. When you’re going to change something which you suspect could affect them, you communicate, inform, and obtain their consent.

There’s no legal obligation, of course. Carefully worded licence terms, privacy statements, and other generic information work around that. But those conditions which are so often thrust at us and clicked through, only highlight the disparity here. Every time I want to update macOS, I’m presented with a long and impenetrable document detailing Apple’s conditions, and have to give my consent. Apple considers it’s necessary to obtain that agreement before I can benefit from fixes to its own bugs, yet it can add services which conduct ‘trials’ and ‘experiments’ without so much as a passing mention.

Others have made it clear that, whatever triald does, it isn’t covered by existing codes of ethics. I’m sure that’s true, but that ignores ongoing debate about the ethics of artificial intelligence generally, including machine learning, and the codes themselves. Whether you’re evaluating novel surgical procedures, testing a new vaccine, or surveying people’s attitudes to immunisation, a fundamental principle is that you obtain the informed consent of those taking part.

Informed consent has a relatively recent history. Although but a small part of the overall Nuremberg Trials in 1945-47, it was embodied for the first time in a memorandum known as the Nuremberg Code, but wasn’t really incorporated fully into non-experimental medical practice until late in the twentieth century. I can still remember doctors who, when examining patients, would instruct them to assume a position for invasive examinations and proceed without explanation or any attempt to obtain informal consent. (If any physician still does that to you, simply remind them of the law on assault.)

Since then, the introduction of ethical codes to other professions and areas of research has invariably been built on a foundation of informed consent. One notable bone of contention has been deceptive research in psychology, which remains highly controversial.

Privacy is also founded on informed consent, as Apple repeatedly emphasises. For example, App Tracking Transparency requires that apps obtain the user’s informed consent when they want to track that user across apps and websites owned by other companies. What Apple terms Privacy Nutrition Labels on product pages in the App Store provide the information allowing the user to decide whether to buy and use an app, making that purchase subject to informed consent. As Apple’s slogan puts it “you have control over what you share”.

As we don’t know what triald does collect, and what it might share by way of results, it’s impossible for us to know how Apple considers its activities relate to its long and detailed statements on privacy. Although I’m sure those who already accuse Apple of not adhering to its high ideals will be as sceptical as ever, I’m confident that triald does comply with its stated policies in all but one respect: nowhere in those pages do I see any reference to trials or experiments, except in the very different context of health studies.

Whether Apple is performing “A/B testing” of products, or trialling different ML techniques, doesn’t matter. The silent introduction of triald doesn’t appear to conform to Apple’s high ideals. I think it should, and that it should do so explicitly and transparently.