One of the novelties of Android 12 that have gone a bit unnoticed was Private Compute Core, in principle an evolution of Device Personalization Services, present in stock Android mobiles and which takes care of functions such as intelligent responses, instant transcription and the that is being listened to. Now Google has given more details about what is Private Compute Core.
For the time being Private Compute Core will remain in charge of these three intelligent functions of machine learning processed directly on the phone itself, with the novelty of the arrival of Private Compute Services, with which this local processing on the mobile will be able to communicate with the cloud respecting the privacy of the users.
Machine learning on mobile
For some time now, Android has integrated several smart functions that are processed on the phone itself, without the need to communicate with Google servers. This is especially important since, for example, smart responses are based on the messages that reach you in the different messaging applications, so processing them on Google’s servers would mean sending all your messages.
This is the reason for the Device Personalization Services application, present on Google Pixels and mobiles with pure Android that include these functions. With the arrival of Android 12, Device Personalization Services has been renamed Private Compute Core, a system isolated from the rest of the operating system and applications in which these functions whose privacy is important are processed.
At the moment, Private Compute Core takes care of three smart functions that are processed on the phone itself: automatic transcription, recognition of the song currently playing on the lock screen, and automatic responses for messaging applications. Google has announced that more features will be added with each new version of Android.
Private Compute Core takes care of processing the information without sharing any data with the applications until the user takes an action (for example, they choose to send one of the response suggestions). While the bulk of the work happens on the phone itself, recognition models need updating from the Google cloud for more efficient operation.
That’s where it comes in Private Compute Services, the novelty that Google has announced, and that acts as a bridge between Private Compute Core and the Google cloud, implementing technologies such as federated learning, federated statistics and private information retrieval. Thus, Private Compute Core does not have direct access to the Internet, but only through Compute Services.
Android’s Private Compute is open source, and Google claims that will publish its source code soon so that security researchers can analyze and verify the security of the implementation. For users, at the moment, without many changes, except for the new name of the application and in the mobile settings.
More information | Google