Expecting anything less than an onslaught of artificial intelligence (AI) announcements by Google at their annual developer conference keynote would’ve been fallacious, to begin with. The tech giant needed to flex muscles, considering how the competition is fierce. There’s an updated Gemini 1.5 Pro model with logical reasoning, a new Gemini Live that’ll get more capabilities later this year, a new and lightweight Gemini 1.5 Flash, updates for Gemini Nano and more AI in Search. Yet it is the declaration of a new era for Android, wrapped in a blanket of AI, which will immediately reach out to a large subset within the 3 billion active Android users (and counting).
As Sameer Samat, who is vice president of Product Management for Android at Google told HT, it is “reimagining Android’s consumer experience and the way you interact with your phone with AI at the core, and that multi-year journey begins now.”
With Android’s AI chapter, Google potentially eclipses Microsoft’s Copilot AI assistant integration in Windows 11 in terms of influence. Copilot potentially reached 500 million PCs in a short period. A lot will depend on the speed at which these updates reach user devices. For Android’s AI rollout, tablets are very much a part of the journey too.
Android x AI, it’s time
Three features form the basis of AI for Android, which are Circle to Search, Gemini becoming the default AI assistant replacing Google Assistant, and the deployment of the updated Gemini Nano model for a variety of on-device tasks. Thus far, Android has had some AI functionality, but that was often limited to certain devices (such as Google’s own Pixel phones) or in scope. At the sharp end was Google’s Recorder app, which uses AI to transcribe recordings. On the other end, the example of AI integration in the Messages app retained limited scope without conversation integration – that meant AI capabilities remained limited to text prompts by a user to generate draft messages or plan events.
HT asked Samat whether the diversity of Android hardware, particularly with mid-range and budget phones, may prove to be a problem to tackle in the next couple of years.
“One of the strengths of the Android ecosystem is the diversity of devices available to consumers at different price points and with differing capabilities,” he said, before adding that while it’s been great to get more users to buy powerful computing devices, there are premium Android devices that do have more powerful hardware complete with a neural processing unit (NPU; important for on-device AI processing) and accelerated AI capabilities.
“Our strategy generally is to make sure that we have a hybrid model for execution. Some will be done on-device only for privacy and latency reasons. For most tasks it may be possible to execute them on adevice, but when not, go to the cloud. We believe this approach serves the ecosystem well,” said Samat.
“I think over time, you’ll see bespoke models or models to split into sort of more specific use cases. Right now, Gemini nano multimodal model is really state of the art that is pushing the edge of what’s possible,” pointed out Dave Burke, who is vice-president for Engineering at Google.
The when, the how, and the phone makers
How long would it take for users to get access to the new AI features on their Android devices? The answer, Google tells us, lies in a two-pronged approach.
“Android is always getting better and therefore many of our experiences are part of Google Play services and Play System Updates. We work closely with our partners to bring these experiences to their devices,” said Samat. That means these two sets of updates should enable some of the new AI functionality on Android phones and tablets, dependent on the specs.
Alongside, they’re working closely with SoC partners (read, chip makers such as Qualcomm and MediaTek) to make sure Gemini Nano can run effectively on flagship devices.
And then there would be a third element, where phone makers would be required to further optimise for Android’s underlying changes to work with their customisations (Samsung’s One UI, Xiaomi’s HyperOS, OnePlus’ OxygenOS and so on).
With this, it is also the end of a period of exclusivity for Samsung’s Galaxy S24 flagship phones, with Circle to Search. Samat said the broader integration of this search method has been optimised for tablets too, following feedback they got from students. He referred to the example of a physics problem, inwhich a student can circle on the screen to invoke search, which then details step-by-step instructions for solving it. “This is only possible because of the deep integration between search and what we’ve done with the operating system,” he said.
Android will use the larger Gemini Pro model, as well as a smaller Gemini Nano model, the latter focusing on tasks where on-device processing is an option.
For a while, Google gave users a choice, to switch Google Assistant to Gemini AI. Now that the transition is complete. Gemini’s contextual awareness capabilities, as Burke underlined, will be able to anticipate what a user is trying to do and that’ll help build context and relevance of suggestions.
For instance, in the Messages app, Gemini can be invoked by a user as a floating assistant window for search and image generation. Burke illustrated with an example of receiving a video link in Messages, which opens on YouTube, and Gemini in Messages is aware of the video and proactively shows suggestions for more information. Or how it can contextualise and answer questions from a mammoth PDF document someone may have shared.
On-device AI, which means tasks will be processed locally on a phone or tablet and not sent to the cloud, figures prominently in Google’s vision of AI for Android. At its core is the 3.8-billion parameter Gemini Nano model. “This means your phone can now start understanding the world the way you understand it, not just text input but actual sights and sounds and spoken language,” said Samat.
Thus far, the transcription capabilities of the Recorder app on Google Pixel phones are an example.
With Android’s AI vision at the cusp of reaching millions of users, would there be an option or method for users to opt out of sharing data for AI training and improvement? “You can choose how and where you want to use AI-powered features. For example, the Gemini app on Android is opt-in, and Circle to Search can be turned off in your phone’s Settings,” Samant confirmed. Google said these developments are tethered to their AI Principles, and therefore users will be given the choice.