Google finalizes Android P as Android 9 “Pie,” launching today (ars technica)

You're right, but I wouldn't be surprised if this, even if not implemented as bluntly as oom-killer-as-a-web-service, still ends up tied to the Play Services blob so they can push new heuristics after the first wave of complaints.

> Based on the app usage patterns, each app is placed in one of five priority buckets. The system limits the device resources available to each app based on which bucket the app is in.
> [...]
> The system dynamically assigns each app to a priority bucket, and reassigns the apps as needed. The system may rely on a preloaded app that uses machine learning to determine how likely each app is to be used, and assigns apps to the appropriate buckets. If the system app is not present on a device, the system defaults to sorting apps based on how recently they were used.

I guess that preloaded app is part of the proprietary Google services, so non-Googlified AOSP devices can either use the default recently-used behaviour or implement their own equivalent app. (I'm not certain but I don't see why else they'd design it as an optional app.)

https://deepmind.com/blog/deepmind-meet-android/ gives a tiny bit more detail on the algorithm:

> we’ve partnered with the Android team to develop a feature called Adaptive Battery that uses a deep convolutional neural net to predict which apps you’ll use in the next few hours and which you probably won’t use until later.

The pixel 2 takes some of the best low light pictures I've ever seen, bar none and I'm including SLR's in that list. They've got some seriously cool software running with the "AI" chip that creates pictures that the camera hardware shouldn't be capable of providing. As an example my profession requires that I sometime stick my camera into a culvert, pipe or manhole, etc to take a picture because it's impossible to see anything in the dark tunnel while in bright sunlight. Most camera are incapable of resolving the brightness difference between the inside of the pipe and the brightness at the entrance and exit and will either overexpose the bright areas or underexpose the dark areas. The pixel balances the brightness across the image and provides an image where you have perfect exposure throughout the image. It does this by taking images at multiple exposures and using its AI software to blend an image with near perfect exposure throughout, it's frankly amazing.

The AI silicon also recognizes multiple pictures taken the same area and submits the pictures to the cloud to be combined, all done without any user intervention. I believe they have intentions to expand the use of the chips including using it for on phone speech recognition and other features. IIRC the Pixel 2 includes this functionality. Google has indicated they intend to expand the use of these AI chips into every area they can where it make sense to use it.

Whose "need"? Your's or Google's? And which one does count, in the end?