Noisy downtown: The inflation era of mobile AI

Although only a quarter past, but if for 2018 years of the mobile phone industry to find a key word, that AI will have a large probability of election.

Led by  and Apple, the two manufacturers in the mobile phone after the concept of artificial intelligence, as if all the other manufacturers clairvoyant overnight, to achieve a technological breakthrough, but also began in the first quarter of the launch of new aircraft to promote AI. In this year's MWC, Ai's thunder is also almost better than the communications industry closer to the 5G, the manufacturers and the media are more favored.

Today's mobile AI is like a downtown, everyone is hawking. But who is bluffing, who really has an AI hole? After studying several models of AI phones, we can simply divide them into three categories, explaining the essence of the product under the sound of Hawking.

Hardware Intelligence: Evolution under perception and decision-making

The first type of mobile phone, we can call it hardware intelligence. Through the improvement of the chip calculation, the original computing environment and computing ability of the mobile phone can be completely changed, so as to undertake more machine learning, neural network computing, improve the ability of the mobile phone all the way.

At this level basically with the Apple A11 chip and the HiSilicon Kylin 970 chip manufacturers mainly-Apple,  and glory. The neural network engine in the NPU and A11 in Kirin 970 distinguishes AI computing tasks from common computing tasks so that AI computing tasks do not have to be uploaded to the cloud and can be performed more frequently and efficiently on the end side.

Another advantage is the terminal computing real-time response to the active service capability, which in the Mate10, Glory V10 and other products embodied in the more obvious, such as through the perception of the user on the use of mobile phone habits, active cleaning memory, closed background programs and other services. All of this is done locally on the phone, and users do not have to worry about their personal privacy being uploaded to the public space in the cloud, with the ability to enjoy better battery life and running speed.

This may be somewhat different from the user's traditional perception of AI, after all, the majority of manufacturers now focus on promoting the ability to take pictures of mobile AI. But battery life, speed and so on in the past when users buy mobile phones, the traditional considerations, and with hardware-supported AI capabilities combined, can also be a new evolution.

Application Intelligence: Counting forces limits AI imagination

The second type of mobile phone we can call "application Intelligence". Under the premise of limited lifting calculation, the AI arithmetic is applied to the places where the user is most likely to be perceived, such as taking pictures and so on. This kind of cell phone basically can't run out the dragon 845 of millet, oppo and so on flagship model.

Unlike AI Solutions and HiSilicon, Qualcomm does not set up a separate hardware unit for AI operations, but instead builds AI engine--to better adapt to AI operations by debugging multiple components such as GPUs, CPUs, and DSPs. The result of this solution is that while mobile computing has improved, the general calculation and AI calculations are confusing and there is a good chance of queuing up. So AI computing is hard to do like hardware intelligence all the time, and computing power is lesser.

A more typical embodiment is  and two new machines, both sides of the main smart camera, but the ability of the mix 2S stay in the smart scene recognition, but P20 can be used to spend more power of the handheld night shooting. and equipped with 845 of the mobile phone basically do not have the active service to the user, the reason is in the simple task of row and row sitting, can not realize the user behavior habit of learning all the time.

Although this kind of mobile phone is also in the propaganda speech operation Big dozen AI card, but the AI and the smart photograph always equate. This is not because the AI can only do smart photos, but because of the limitations of the application of intelligence itself, limiting the AI imagination of these phones.

Enigma Intelligence: An ancient AI that is hard to perceive

To the last category of mobile phones, we can only call it "puzzle intelligence" ... This kind of cell phone usually does not change on the chip, in the software also did not see where intelligence, but people just want to say that they are smart.

Here we have to understand a concept, in the absence of AI chips, and even the mobile phone computing ability is very low age, mobile phones can still be equipped with AI applications. For example, the automatic beauty function of self-timer software, including the basic human face fixed-point recognition, neural style migration and so on AI operations. But when you enable these features, you usually upload your photos to the cloud and use the computing power of the cloud to complete the calculations.

The use of cloud computing power will have many shortcomings, such as the upstream of the data consumption time, so that many functions can not be carried out in real-time, such as connecting to the cloud need to rely on the network, once there is no signal can not be related to the calculation.

The smart phones of these mysteries are usually just cloud computing, and then advertised as the ability of their own phones. This year's MWC, LG made such an embarrassing show, launched the new machine LG V30s ThinQ, and claimed to be the main AI.

Results v30s Thinq's AI features only two aspects, on the one hand, can take pictures of the AI Vision, on the other hand, in the field of speech AI voice. And the phone is equipped with the Dragon 835, which is more than 845 of the force is worse. In the video of the AI is basically copying the image of the software's filter function, voice more embarrassing, LG did not develop a voice assistant, chose and Google assistant cooperation set up a 50-function voice instruction set, through Google Assistant calls some common functions of mobile phones, turning Google Assistant into a remote control with 50 buttons.

To say that these phones do not have AI, it seems a bit unfair, after all, they can also give users a number of related functions, but still rely on the "ancient times" the old-fashioned way to achieve. Whether it's a technical kernel or a sense of use, there's almost no elevation. What is the point of "AI" that cannot be perceived?

AI abuse, will the cell phone homogeneity change?

As the AI industry media, we naturally do not want to see the mobile phone AI is confusing, or even become a manufacturer hype tools. But looking in the right direction, the emergence of these three types of mobile phones proves that the homogeneity of the handset is being improved.

Before the advent of mobile AI, mobile phones have been caught up in "micro-innovation" for a long time. Not only a curved screen of water, a full screen, a dual camera, to the end the manufacturers even began to look at the shell color, as if the collection of seven colors can summon the dragon. Not only users themselves are bored, even the mobile phone industry itself is often self-examination-mobile phone This product, is not already touched the ceiling?

But the emergence of mobile AI to break the looming ceiling, the ability to improve the photography of mobile phones can be compared with the card machine or even micro-single, the promotion of active service capabilities, so that the phone can remember the user's habits, travel, and eventually become the user's life assistant ...

These are just the beginning, and we can see that the leaders of mobile AI in the two areas of iOS and Android are building a different mobile AI ecosystem. Apple uses AI to achieve greater computational power and spatial awareness, helping arkit achieve better visual effects and creating an AR-centric mobile image entertainment revolution. 

As the two development ecosystems mature, the experience of mobile AI will also lead to other changes in the day difference, while the first echelon of AI mobile phones will be more and more distance from the later. The time has long been no need for the media for users to read the authenticity of those phone AI, users can feel the huge difference.

Now manufacturers are abusing the name of AI, which is the inevitable inflation in the advent of technological innovation. After a few tumultuous and calm, the market law will slowly wash the participants, and finally let everything go on track.

Noisy downtown: The inflation era of mobile AI