Artificial intelligence is the next battlefield. Special chips and software make smarter phones.
We have arrived in the year in which the appearance of the smartphone has almost crystallized out. Samsung kicked off with its virtually borderless Galaxy S8 and also LG, Xiaomi and Essential already have phones with hardly an edge. The new iPhone will also be added to that list, which means that the phone from outside will simply be a screen even more than before. This makes it even more important when choosing a smartphone what it does and how smart it is.
Manufacturers all focus on artificial intelligence (AI). In addition, you probably think of the smart assistants such as Siri and Google Assistant. Rightly so, but that’s just the tip of the iceberg. Because responding smartly to questions and commands will soon be just one of the many things artificial intelligence will do for us.
Proceeding predictably and almost imperceptibly may become the most important function of artificial intelligence. Face and object recognition will be much better for photo search even more accurately, predictive typing will become so smart that it will represent entire sentences and smartphones will soon know before you even know which app you are going to open. Artificial intelligence soon understands exactly what you take a photo or video of, and can adjust the settings better than you can. And because the phone will constantly learn from you, it can use the battery more efficiently.
Although artificial intelligence is, of course, software in the end, it greatly benefits from own hardware. For example, Huawei recently announced its latest chip, the Kirin 970 , at the tech fair IFA . The chip has a separate component that runs on artificial intelligence, and combines the cloud and local data processing. For example, AI can dynamically control and distribute tasks such as image recognition between all parts of the telephone and special servers elsewhere. Volens Huawei can use this chip to recognize images five times faster than the competition, in addition to numerous other functions. ‘The Kirin 970 is the first step in a series of new developments that bring powerful AI functions to our devices,’ said Huawei CEO Richard Yu about the chip.
Apple also reportedly takes such a first step, but as always in secret. Earlier this year, Bloomberg reported that Apple would work on a separate chip specifically for AI, the Apple Neural Engine. Just like all other phones without such a separate chip, AI tasks on the iPhone are controlled by the main processor and the graphics chip.
By shifting those tasks to a special chip, or a separate part of it, the AI could become more versatile, while the phone becomes more energy efficient. Apple has been using separate chips for some time: the iPhone has an M chip for keeping track of all the sensors of the device and a Secure Enclave chip that controls security with passwords and the fingerprint scanner.
During the WWDC developer fair, Apple announced a new framework for self-learning object recognition, Core ML . This technology must make things like face recognition and text recognition much faster than currently, and that on existing hardware. If developers build in that technology now, it will only get faster when Apple finally inserts a special AI chip into an iPhone.
Qualcomm, the largest maker of chips for Android devices, takes a different approach. The company has created a software platform that allows developers to optimize their AI apps for existing chips in the Snapdragon 600 and 800 series. This Neural Processing Engine ensures that, for example, image recognition apps run more smoothly on phones with a Snapdragon chip. The same applies here: if developers already use the framework now, the introduction of a chip with a separate AI part will only become easier. Facebook is already using the platform to make face filters work better.
Facebook also works on such a software platform, Caffe2Go . That framework is first trained on powerful computers with large data sets, and can then get started on smartphones without the need for huge computing power. According to Facebook, this approach takes longer than methods that require a constant connection to a powerful server. A big advantage of course is that the AI functions of Caffe2Go apps work well without a good connection to the server being required.
AI in the Cloud
At the beginning of this year, Google announced the Cloud TPU (Cloud Tensor Processing Units) during development fair I / O. A self-developed, very powerful AI chip that has been installed in large numbers in Google’s data centers. The same applies here as on smartphones: special AI chips are better at running and training artificial intelligence than normal hardware.
Because Google’s data centers got too busy with things like image recognition and translation, the company developed a chip that could do better. Google speaks of an AI supercomputer, which companies can use as an AI cloud service to perform tasks that normal computers are not powerful enough for.
Microsoft and Amazon also have such AI data centers where companies can use their artificial intelligence. However, those companies do not make their own chips, but use those from Intel and Nvidia. The second became big with video cards for gamers, but has been developing chips for artificial intelligence for a number of years now. Intel missed the boat with chips for smartphones, but is now developing both powerful AI chips for data centers and small AI chips to, for example, provide drones with self-learning computer vision.
Without you noticing it as a consumer, artificial intelligence has been working behind the scenes for years. The combination of extremely powerful AI data centers to train artificial intelligence and special chips in our smartphones is waiting.
Years of theory, development and preparation come together to make our smartphones really smart with artificial intelligence.