It's been almost two years now since I picked up an Android phone for the first time, during which time it has gone from a generally unloved, nerdy thing to the soon-to-be dominant smartphone platform. This is a remarkable and sudden development--when people start fretting about the state of Android as an OS (fragmentation, competing app stores, etc.), they tend to forget that it is still rapidly mutating and absorbing the most successful parts of a pretty woolly ecosystem. To have kept a high level of stability and compatibility, while adding features and going through major versions so quickly, is no small feat.
Even back in v1.0, there were obvious clever touches in Android--the notification bar, for instance, or the permission system. And now that I'm more used to them, the architectural decisions in the OS seem like "of course" kind of ideas. But when it first came out, a lot of the basic patterns Google used to build Android appeared genuinely bizarre to me. It has taken a few years to prove just how foresighted (or possibly just lucky) they actually were.
Take, for example, the back button. That's a weird concept at the OS level--sure, your browser has a one, as does the post-XP Explorer, but it's only used inside each program on the desktop, not to move between them. No previous mobile platform, from PalmOS to Windows Mobile to the iPhone, used a back button as part of the dominant navigation paradigm. It seemed like a case of Google, being a web company, wanting everything to resemble the web for no good reason.
And yet it turns out that being able to navigate "back" is a really good match for mobile, and it probably is important enough to make it a top-level concept. Android takes the UNIX idea of small utilities chained together, and applies it to small screen interaction. So it's easy to link from your Twitter feed to a web page to a map to your e-mail , and then jump partway back up the chain to continue from there (this is not an crazy usage pattern even before notifications get involved --imagine discovering a new restaurant from a friend, and then sending a lunch invitation before returning to Twitter). Without the back button, you'd have to go all the way back to the homescreen and the application list, losing track of where you had been in the process.
The process of composing this kind of "attention chain" is made possible by another one of Android's most underrated features: Intents. These are just ways of calling between one application and another, but with the advantage that the caller doesn't have to know what the callee is--Android applications register to handle certain MIME types or URIs on installation, and then they instantly become available to handle those actions. Far from being sandboxed, it's possible to pass all kinds of data around between different applications--or individual parts of an application. In a lot of ways, they resemble HTTP requests as much as anything else.
So, for example, if you take a picture and want to share it with your friends, pressing the "share" button in the Camera application will bring up a list of all installed programs that can share photos, even if they didn't exist when Camera was first written. Even better, Intents provide an extensible mechanism allowing applications to borrow functionality from other programs--if they want to use get an image via the camera, instead of duplicating the capture code, they can toss out the corresponding Intent, and any camera application can respond, including user replacements for the stock Camera. This is smart enough that other platforms have adopted something similar--Windows Mobile 7 will soon gain URIs for deep linking between applications, and iPhone has the clumsy, unofficial x-callback-url protocol--but Android still does this better than any other platform I've seen.
Finally, perhaps the choice that seemed oddest to me when Google announced Android was the Dalvik virtual machine. VMs are, after all, slow. Why saddle a mobile CPU with the extra burden of interpreting bytecode instead of using native applications? And indeed, the initial versions of Android were relatively sluggish. But two things changed: chips got much faster, and Google added just-in-time compilation in Android 2.2, turning the interpreted code into native binaries at runtime. Meanwhile, because Dalvik provides a platform independent from hardware, Android has been able to spread to all kinds of devices on different processor architectures, from ARM variants to Tegra to x86, and third-party developers never need to recompile.
(Speaking of VMs, Android's promise--and eventual delivery--of Flash on mobile has been mocked roundly. But when I wanted to show a friend footage of Juste Debout the other week, I'd have been out of luck without it. If I want to test my CQ interactives from home, it's incredibly handy. And of course, there are the ever-present restaurant websites. 99% of the time, I have Flash turned off--but when I need it, it's there, and it works surprisingly well. Anecdotal, I know, but there it is. I'd rather have the option than be completely helpless.)
Why are these unique features of Android's design interesting? Simple: they're the result of lessons successfully being adopted from web interaction models, not the other way around. That's a real shift from the conventional wisdom, which has been (and certainly I've always thought) that the kind of user interface and application design found on even the best web applications would never be as clean or intuitive as their native counterparts. For many things, that may still be true. But clearly there are some ideas that the web got right, even if entirely by chance: a stack-based navigation model, hardware-independent program representation, and a simple method of communicating between stateless "pages" of functionality. It figures that if anyone would recognize these lessons, Google would. Over the next few years, it'll be interesting to see if these and other web-inspired technologies make their way to mainstream operating systems as well.