Google: Our new Android P beta shows how AI gives you more battery life

DeepMind takes a first crack at using AI to solve Google's challenges on mobile devices.
Written by Liam Tung, Contributing Writer

Google's Android developers have teamed up with Alphabet's DeepMind researchers to bring deep neural-net enhancements to Android P.

The enhancements, which target battery life and screen performance, have been unveiled alongside the launch of the first Android P beta, which is now open to six more handset models beyond Google's Pixel devices. The beta follows the first Android P developer preview in March.

Owners with a Sony Xperia XZ2, Xiaomi Mi Mix 2S, Nokia 7 Plus, Oppo R15 Pro, Vivo X21UD and X21, and Essential PH‑1 can now join the Android P beta program.

This beta program will be the fastest route to testing Android P's new Adaptive Battery feature, which was developed with researchers from the unit at Google parent Alphabet's AI firm, DeepMind, dedicated to delivering AI breakthroughs just for Google.

See: Special report: How to implement AI and machine learning (free PDF)

Historically, these AI projects have addressed large-scale infrastructure, such as saving energy at Google's data centers and improving Google Assistant.

Adaptive Battery is the first project that scales down DeepMind's machine-learning applications for running on mobile devices. The feature prioritizes a user's favorite apps and places more restrictions on access to system resources for less frequently used apps.

Adaptive Battery adds to past optimizations for battery-life improvements such as Doze, App Standby, and Background Limits.

DeepMind's work helps the device predict apps that will be used in the next few hours and those unlikely to be used until later. DeepMind says early testing has produced "very promising" results and should help improve battery life by reducing background activity.

See: What is AI? Everything you need to know about Artificial Intelligence

The other smart feature aided by DeepMind is Adaptive Brightness, which uses machine learning to automatically adjust screen brightness after learning a user's preferences in different light conditions. The aim is to reduce the number of occasions users need to manually adjust brightness.

A new machine-learning feature aimed at developers is App Actions, which can recommend apps based on predictions about what the user is attempting to achieve.

This technique can help developers bring an app's relevant features to the fore and is aimed at boosting app engagement. It goes hand in hand with Slices, a method for developers to provide content in Google Search and Google Assistant.

Google is targeting Q3 for the final release of Android P following three more beta builds between now and then.


Adaptive Battery is the first project that scales down DeepMind's machine-learning apps to run on mobile devices.

Image: Google

Previous and related coverage

Android P: New features, release date, and everything you need to know

Google's latest mobile operating system is here -- and we have all the details currently available.

Google's deep learning system aims to tame electronic health records

Google is using a deep learning system to navigate electronic health records without the data wrangling.

Google AI can predict your heart disease risk from eye scans

Google's deep-learning algorithm could offer a simpler way to identify factors that contribute to heart disease.

Google's DeepMind: What can these battling AIs tell us about human behavior?

Google-owned DeepMind is using artificial intelligence to study how cooperation and conflict emerge.

Google DeepMind's latest AI? So smart it can 'reason' its way around London's Tube

Google's DeepMind subsidiary has created an artificial-intelligence system that can store memories and use them to reason.

Editorial standards