Google's Pixel phones are leaders in AI features. Except that nobody owns these.
AI and your cellphones
In a previous post, I reported how Google’s unveiling of Gemini LLM didn’t quite live up to the hype. Until Google releases its Gemini Ultra to the public, I must say OpenAI remains an industry leader in natural language AI.
However, Google is an industry leader when it comes to incorporating AI into their phones. Google Pixel 8 and Pixel 8 Pro introduced several proprietary AI applications including Magic Editor, Magic Audio Eraser, Best Take, and AI-enabled Recorder. Google’s AI media apps, however, do not run on the phone itself. They run on Google’s cloud servers. While this approach unburdens the phones from carrying out heavy computations, it also presents a number of issues.
The cloud features require an internet connection and also significant bandwidth to run as the media files can consist of a large amount of data.
Requiring the users to send raw data for processing over the internet can result in privacy and data integrity issues.
The service providers must maintain the necessary cloud infrastructure, which incurs significant resources.
The above drawbacks are present in the current AI mobile, apps including ChatGPT mobile apps.
Google’s announcement of Gemini Nano, however, indicates that a new stage is unfolding. Google’s Gemini Nano model is small and simple enough to run on high-powered mobile phones. Google Pixel 8 Pro can now summarize audio and write responses using Gemini Nano.
Gemini Nano doesn’t have the capability to process videos and images like the cloud models. Thus, for the foreseeable future, the heavy lifting of images and videos will need the cloud to run. Nonetheless, I have no doubt the tech companies eventually want to run multi-modal generative AI on mobile devices.
Google is not the only company looking into entering the mobile AI foray. In fact, Xiaomi, a Chinese smartphone manufacturer, introduced its own “on-device” large language model on its phones before Google. Xiaomi’s model reportedly has 1.3 billion parameters which makes it comparable to Google’s Gemini Nano 1 model. Xiaomi’s Chinese competitor, Oppo, is also set to launch its own device LLM called AndesGPT on its phones.
Samsung has announced that a mobile version of its Gauss LLM will be coming to Samsung phones in 2024. Gauss on mobile will be introduced with the upcoming Samsung S24 phones.
Apple of course is a wild card here. Before LLM, Apple was a pioneer in incorporating simpler AI features to smartphones. (Think of Siri.) Apple prefers to run as much computation on devices, instead of clouds, for the same reasons outlined above. Yet in the current wave of generative AI frenzy, they have started late. Apple, being late to the party, will spend enormous resources to catch up. According to this Bloomberg article, Apple plans to spend a billion dollars a year and plans to incorporate stronger AI into everything including Siri and Apple Music.
It’s a certainty that Apple will unveil its AI-related features and roadmap at next WWDC. The industry expects LLM powered Siri to be introduced for iPhone 16 running iOS 18. It is not yet clear how much of Apple’s iOS AI will run on the device or on the cloud.
In conclusion, AI technology is poised to transform the nature of smartphones and mobile devices in coming years. More complex AI operations will rely on cloud-based computing for time being. But many companies are rushing to implement on-device AI to improve user experience and privacy, not to mention decreased cost. As mobile phones have transformed our lifestyle, it remains to be seen how mobile devices running AI will transform our future.