Generating and modifying images, erasing objects in photographs, writing texts, creating videos… The new generation of high-end mobile phones is coming loaded with artificial intelligence. And it won’t be long before this trend permeates other more modest models that are yet to come. At this point, the industry faces a dilemma: to process all these tasks in the cloud or on the device itself.
The case of Samsung and its latest smartphones The Galaxy Z Fold 6 and Flip 6 foldable devices illustrate the dilemma. The Korean firm presented its new terminals this summer and with them announced new tasks based on artificial intelligence. But not all of them work in the same way. Some are processed on the device itself, while others use the cloud. The first are solved with the resources of the device. smartphonebut in the latter the information travels through the Internet to a data center, where servers execute the task and send the result to the phone.
For these types of AI functions that are processed in the cloud, Samsung has specified a clause: they will be free until the end of 2025. Why? “We have drafted this provision for a simple reason. We are humble enough to recognize that we do not know what will happen,” says Patrick Chomet, Head of Customer Experience at Samsung Electronics.
Processing AI tasks in the cloud is by no means free. It requires significant computing resources, and this is energy intensive. In February 2023, when ChatGPT had barely emerged, the analyst firm SemiAnalysis published A study on the potential costs of AI. They calculated that a query to ChatGPT would cost around 36 cents. And here we are only talking about text, because the service did not yet work with images, the generation of which increases processing costs. At that time —OpenAI’s application had already announced that it had surpassed 100 million users— operating ChatGPT would cost its creator around $700,000 a day.
Whether smartphone users will compulsively use these AI features, with the associated costs, or whether these services can be made profitable in the future, Samsung is acting cautiously. “In the camera, there are many functions that run on the device, such as portrait (applying a 3D animation or watercolor style to a photo of a face) or erasing objects. If you want to do something advanced, such as going from sketch to image or editing videos, this is done in the cloud. And we don’t know how people are going to use all this. In two or three years, based on how users behave, we will know,” explains Chomet. The executive does not rule out the possibility that some generative AI functions will require a subscription in the future.
Promote local processing
AI task processing in data centers is growing steadily. In 2023, the cloud AI market was valued at $62.63 billion, According to the analysis firm Grand View ResearchBy 2030, it is estimated that it will reach 646.6 billion dollars, at an annual growth rate of 39.6%.
But smartphone manufacturers are aiming to run some of the AI tasks locally, using the device’s own hardware. This involves the main processor (CPU), the graphics processor (GPU) and a specialised unit for AI functions (NPU). This approach has its advantages. It offers greater privacy to the user, as data does not travel to external servers, and artificial intelligence works without an internet connection.
However, when AI is done in the cloud, it uses more powerful systems. Accuracy is higher for certain tasks, such as generating images or complex texts. From here, smartphone manufacturers have to decide which mode to use for each task.
For Samsung the criteria is clear. If the performance of the hardware of the smartphone That’s enough, the function is not transferred to the cloud. Thus, in the Galaxy Z Fold 6 and Flip 6, the text composer, to write an email or a post for social networks, works locally. But the generation of images from a photograph or the real-time translation of a call require online processing.
“The two elements we take into account are performance and privacy,” says Chomet. “Some tasks need to be very fast and others should work without connectivity, such as tasks that have to do with the private use of personal data. These two elements, performance and privacy, guide us to run the functions inside the terminal as much as possible.”
The other major manufacturer of smartphonescurrently second on the market, faces the same dilemma. When Apple announced its artificial intelligence, it stressed that there will be certain queries that will be resolved locally. Tasks that require more powerful AI systems will travel to the cloud. But only the data essential to obtaining the result will be sent, without the information being stored or used to train the algorithms. In addition, for some requests, the devices will redirect the user to ChatGPT to obtain a better response.
The hybrid approach between local and cloud processing seems to be the way forward. Manufacturers will differ, of course, in the weight they give to each of these two options. But also in how they present them. “We will make it clear to people exactly what is not running on the device so they can choose,” adds Chomet.
The enormous energy consumption of AI processing in the cloud also poses another dilemma. The more this technology is used, the more electricity consumption to power the servers and water to cool them increases. Companies such as Google and Microsoft have reported significant increases in their electricity consumption in 2023. The former even pointed out the energy voracity of artificial intelligence in its annual sustainability report.
You can follow THE COUNTRY Technology in Facebook and X or sign up here to receive our weekly newsletter.
