Tuesday night at WWDC 2016, John Gruber had his weekly podcast The Talk Show with Phil Schiller and Craig Federighi. A recording of it is available on his blog. It’s the best 78 minutes of video you’ll spend this weekend.
My favorite highlight of the keynote, aside from Apple’s intriguing implementation of differential privacy, is the fact it is doing data analysis on-device instead of in the cloud the way Facebook and Google do it. A little past the 51 minute mark of the episode, John wonders whether Apple’s securer strategy is going to win in the long run. In Apple’s favor, he points out that a distributed on-device system implies that, in Craig’s words, “There are a billion phones to throw at this problem.” A billion iPhones could easily match the computational power of a robust cluster of cloud servers. Craig then goes on to mention a second important point:
There’s this idea that, well, if you don’t have the data, how would you ever learn? Well it turns out if you want to get pictures of mountains, you don’t need to get it out of people’s personal photo libraries. Like—we found out we could find some pictures of some mountains; we did some tough detective work and we found them.
It will be interesting to see how this plays out in the next 5-10 years. It’s not my area of expertise but at a gut level I can’t help but worry that it’s going to be hard having AI1 as sophisticated as Google’s without eventually moving to the cloud. I see Apple as always being stronger at privacy than competitors, and always weaker in data analysis. I’d love to be wrong. If anyone can do both better than anyone else, it would be Apple. But neither feature one can happen overnight, or in just one year, and quite frankly Apple’s got some catch up to do.2
It’s worrisome to think that most consumers might prefer the convenience of sophisticated data analysis over privacy if they were forced to choose between the two. This isn’t really a decision people have to make yet because the tech industry as a whole is still just on the cusp of AI. In other words, it’s not going to be clear until a few more years when AI plays a much more significant role than it currently does and when the companies that don’t have AI or don’t have it as thoroughly are more painfully noticeable.3
Here’s to Apple proving that these two ideas, privacy and deep learning, can be symbiotic instead of mutually exclusive.
- I’m throwing around deep learning, data analysis, and AI as synonyms here. ↩︎
- I can’t help but wonder what sort of “detective work” Craig had in mind. Where did the photos come from? Was some sort of Google Custom Search API integration involved? If Apple doesn’t want its fingers in people’s personal photo libraries but at the same time wants the same level of vendor-independent intelligence as Google, I don’t see how it can avoid eventually building its own robust search engine crawler. That’s some serious catch-up required indeed. ↩︎
- Unless major catastrophe strikes, civilization will be truly unrecognizable in the next 50-100 years because of AI. It’s coming. ↩︎