Best Thing About Siri On-Device Learning May Be Ability to Disable It

Apple is on a roll.

By Kevin RR Williams

TECHNOLOGY The masses go gaga over each new Apple product revelation. Another group takes a step back and asks, “Deja Vu. Haven’t I seen this somewhere before?” The Worldwide Developer Conference (WWDC) 2017 did not disappoint either audience. People immediately began poking fun at the shape ▿ of the HomePod Siri enabled speaker.

It seems Apple has been on a roll lately. All puns aside, there is something ominous about the new Siri available on HomePod, iOS 11, and the next Mac OS. It is something called “On-Device Learning.” As the name implies, data is stored on the device, which is good. But system-wide predictive logic can be problematic when the “intelligence” is artificial.

Apple ran into this issue with the introduction of its Messages app. Because predictive text reduces data entry times, it was initially hardwired with no opt-out. The iOS corrected intentional abbreviations like “LOL.” This drove serial texters mad. Learning the error of its ways, Apple provided methods to disable predictive text ▿. Disaster averted. In iOS 11, Siri is doing it again.


Siri waveform

On-Device Learning

You know how when you shop at Amazon, everything you browse seems to follow you — all over the Internet? It is even worse with Amazon suggestions and emails. If you purchase an iPad cover, you are bombarded with dozens of other after-purchase covers and sleeves. (Since you just purchased one cover, you must need 6 or 10 more.) When you stream a movie, other films are suggested “because you watched…” I used Priceline to book an emergency flight for a funeral. Priceline now thinks I am a globetrotter, emailing me numerous travel offers. This algorithm is similar to On-Device Learning. Ironically, the new Safari disables browser data retrieval for predictions from other websites.

Here is the idea: With On-Device Learning, when you search for eyeglasses, Apple assumes you are nearly blind. So local eye exam locations show up on Maps. If you lookup photos of a different country, Apple assumes you are planning a trip there and curates news about the destination in Apple News. If you read an article about bombings or politics, the same principle applies.

My problem is this. As a clinical researcher, my interests are diverse. It is virtually impossible for Apple to predict what may interest me. When writing a blog article about cancer, significant research on the subject may be gathered for several days. After the article is complete, it is time for the next topic. It may be blueberries, longevity, human anatomy, dieting, vitamins, programming, emotional disorders, or women’s health. Interspersed with all this I check the latest news headlines, read the Bible online, skim Twitter feeds, and pop into Pinterest. Mix it all together and my Siri profile might be that of a depressed woman, desperately seeking (super)natural remedies. While reading a book, I may lookup words and places for better comprehension. Interest associated with an article is temporary and should not become an enduring symbol of my inter-application preferences.

Apple mobile devices have yet to offer multi-user logins. Therefore someone else who uses your tablet adds to the mayhem.

The only solace in all this is, basing my believe on Apple history, On-Device Learning will be optional. I can hardly wait for iOS 11 to roll out so I can disable On-Device Learning and then enjoy all the other new features like more natural sounding Siri voices, drag-and-drop, improved app switching, expanded iPad dock, and easier keyboard functionality.

Protected by Copyscape

Tags: learning, mobile devices, programming, technology