Google I/O 2024 shed light on the new artificial intelligence (AI) initiatives the company has taken and how it plans to integrate them within its Android platform. During sessions organized for developers, the tech giant took a deeper dive into the development of artificial intelligence on a device that app makers will soon be able to take advantage of. It currently offers server-based AI features as well as on-device AI features, but these are mostly used by Google’s internal apps. Notably, the company also announced an enhanced Google TalkBack service powered by Gemini Nano with multimodality later this year.
The 19-minute video titled “Android on-device AI under the hood” was posted on YouTube on Google’s Android Developers channel. The developer-focused session opened with a discussion of the benefits and limitations of keeping AI processing localized within the device. Some of the highlighted benefits include local processing, offline availability, potentially reduced latency, and no overhead. The additional costs here refer to the cost of cloud computing.
Although not mentioned in the video, the AI on the device is also a good thing for users in terms of privacy. When the computing takes place inside the device, the data that the AI uses to generate the content never leaves the device and cannot be accessed even by Google or third-party developers. However, all these advantages come at a price. The biggest limitation is the computing capability of a device the size of a smartphone. Despite new chipsets being introduced with neural processing units (NPUs) and more powerful GPUs, they cannot compete with cloud-based infrastructure.
This means that AI features will be limited in what they can do. Highlighting the same, Google has classified the AI features on the device into three groups Consume, Create and Classify. Consume includes actions such as compressing a page or providing an email preview. Create includes text generation and its reformulation and rewriting, while Classify includes sentiment analysis and checking the tonality of the message. Namely, Google didn’t mention any multimodal AI features, probably because powering them locally on the device isn’t currently possible.
The session also focused on the Gemini Nano, which seems to be the tech giant’s preferred choice for Android features on the device. While Gemma has also been mentioned as a potential alternative, most of Google’s current on-device AI features, such as Magic Compose in Google Message, Summarize in Pixel Recorder, and Gboard Smart Replies, are all built on Nano. Furthermore, the company also confirmed that the Google Talkback service will also get an upgrade later this year and will be powered by the Gemini Nano.
Until now, the choice of apps that use AI on the device has been limited to Google’s internal apps. But the company hopes that through these interactions, it will be able to involve more developers in the attack.
