Apple’s digital assistant Siri was a real revolution when it was launched in 2011 – but after that it lived in a niche for years, attracting attention with its unintelligent responses and sometimes blatant incompetence. Google’s Siri counterpart “Assistant” was only slightly better, responding to many commands with the sentence “I found that on the Internet”. The weakness of both has become even more apparent in the last two years, as chatbots like ChatGPT are now demonstrating how helpful assistants with artificial intelligence can be for their users.
In the past two weeks, Apple and Google have finally got the AI revolution on smartphones rolling: At its developer conference last week, Apple showed a new version of Siri for the first time, which enables real conversations with the user, driven by a large language model called “Apple Intelligence”. Google announced something similar in the spring for its Android operating system with its AI Gemini.
And while Apple users still have to wait until the fall for the AI version of iOS, Google brought artificial intelligence to its Pixel smartphones 8, 8 Pro and 8a with an update last week. These so-called “feature drops” are comprehensive this time – and they make the current generation Pixel phones the first real AI smartphones.
Google’s new Gemini functions will officially appear first on its own Pixel phones. However, the company has already agreed to a cooperation with Samsung and wants to bring Gemini to the South Korean company’s Galaxy phones as well. This has advantages for both sides: Samsung can market Google’s Gemini technology as its own Samsung AI, and Google can test its algorithms on a much larger customer base. As good as the Pixel phones are, they are not very widespread. Google’s flagship Pixel 8 Pro in particular is selling significantly worse than Samsung’s Galaxy devices.
Google’s Gemini is the first AI on smartphones to allow real conversations with context. This makes Google search by voice really useful for the first time. For example: When asked: “What will the weather be like?”, Google Assistant provides a weather summary for the current location today. When asked “And tomorrow in Cologne?”, the weather forecast for Cologne for the following day follows.
The next query, “What can I do there tomorrow?”, provides useful suggestions, including the events calendar, for Cologne the following day. The Gemini algorithm remembers previous queries and uses them to construct a context for the conversation – this was not previously possible on smartphones, also because the “assistant” had to send large amounts of data to Google’s servers.
With the latest version of Gemini, Google has built a smaller, less data-intensive model called “Nano” that can run directly on the smartphone. On the one hand, this ensures greater data protection, and on the other, Google’s servers are not flooded with requests. To make this work, Google has built a chip called “Tensor G3” into its current generation Pixel phones that is specifically designed to solve machine learning tasks.
This allows the company to save computing power in its own data centers, as well as battery power and data volume for its users. Apple also wants to take a similar approach, with a version of “Apple Intelligence” adapted so that it runs directly on the device, without computing assistance from the cloud.
Google’s search offensive on smartphones goes even further: Users have been able to search for things using the camera for some time now. In the new version of Google Search, Gemini now interprets images. The user can draw a circle around the object they are looking for with their finger, then the AI goes on the search, provides context and describes what it sees.
The voice application is also useful in other areas: Gemini now edits short messages and makes suggestions on how they can be rephrased – the user can specify the style. Gemini can also prepare suggestions for emails on the smartphone. However, anyone who wants this function integrated directly into Google’s office programs Gmail and Documents currently has to take out a paid premium subscription to Google’s One service and pays 22 euros a month for it.
The use of AI in the camera software of the Pixel smartphones is also noteworthy. The Pixel 8 Pro in particular uses three lenses simultaneously to put images together using an algorithm – a function that worked well in WELT’s test. The photos from Google’s smartphone flagship can then be edited using a “magic editor”.
Google’s AI can, for example, remove distracting elements from the image and generate suitable replacement images, or, in the case of group portraits, put together an image from several photos in which all the people in the photo have their eyes open and are smiling. Portraits can be enhanced afterwards with blurred backgrounds, and photos taken at night turn out much better thanks to AI.
The optimization of exposure and colors using AI also works remarkably well – this has been the case for a while, and Google has been providing its paying customers with the algorithms for several years. What is new here, too, is that the processing with the Tensor chip is largely done on the phone itself.
In the USA, Google also wants to use its AI on the phone to analyze phone calls in real time: If a fraudster calls and tries one of the well-known customer service scams, the AI should warn the user. Since the analysis of the conversation content takes place on the phone itself, Google says the user’s privacy should be protected. For now, the company is only trying this in the USA – whether the function would also be compatible with European AI and data protection rules has not yet been clarified.
