Artificial intelligence on smartphones makes all the sense in the world and is not smoke, but, today, it is not going to take advantage of it. Don’t buy a terminal just for it.
Artificial intelligence on smartphones has already been talked about at length. All manufacturers are betting on integrating neural processing units into their chips, with which to execute machine learning tasks with up to 25 times higher performance than a traditional CPU and 50% more energy efficient. The public and the specialized press have come to think that this is “smoke”, which are empty figures. With certain tests, you can prove that they are true, and that launching new chips makes all the sense in the world.
Artificial intelligence on smartphones makes all the sense in the world for manufacturers. After reaching extremely high performance figures with traditional CPUs, and thus achieving a high consumer satisfaction rate, the market can not stagnate, differentiation must be achieved. This, added to the new needs in terms of facial recognition, object recognition or detection of behavioral patterns in the daily use of smartphones, makes dedicated units on chips a great incentive to promote new functions and invest in marketing talking about “intelligence”.
The reality is that although all these functions can benefit from a lot of speed and efficiency in chips such as the Kirin 970 or still unpublished Snapdragon 845 of Qualcomm, there are no applications with which to take advantage of its extra power. Even at the system level, using a Huawei Mate 10 or Honor View 10, maximum exponents of the artificial intelligence in smartphones, the differences are not noticeable by the user in the day-to-day, and where they should be, as in the camera app, the user help is quite disappointing.
Companies like Huawei promote their new terminals from the perspective of AI, and the truth is that, at the moment, you can hardly take advantage of it.
«We have to wait for the applications, which will arrive soon“, is the message that manufacturers and enthusiasts are sending. And the truth is that yes, that those who make use of these dedicated units may be on their way. But the number of terminals that offer something like this is so small, compared to the cost of developing a new function in an application, that the incentive is close to zero in most cases.
There are universal APIs like TensorFlow Lite or Caffe2 of Facebook, but they do not offer the same performance as private ones offered by manufacturers like Qualcomm and Huawei, and as long as there is no good integrated solution at the system level, the ball will not move. Recommending acquiring a terminal for its artificial intelligence today is like having done it for the support of LTE networks in 2011, although in Spain we could not enjoy the new speed of telephony networks until 2013. Only the iPhone, because it uses only one system, iOS, can speak of a standardized solution for all developers, and it is not that the expectation is overwhelming. In a few years, it will be an aspect to prioritize, but for now, no.