With smartphones becoming indispensable in our lives and with all the new and groundbreaking features they are coming up with, artificial intelligence (AI) is a technology which seems to be the most highlighted. So, to measure the AI capabilities of smartphones, Chinese benchmark testing company AnTuTu has released a public beta version of its AI testing app called ‘AI Review’. We should also mention that the website has made the apk file public to so that it can be downloaded from its official website.
However, AnTuTu has also stated in its blog post that it might not actually be possible to measure the capabilities of AI on a single score as there are many SoC makers, smartphone companies and software developers and they are all working to make their own version of the technology. However, there are some phone companies which have not released their SDK to allow other apps to read into their AIs and that puts a negative score to the smartphone’s AI. The version of Android on the smartphone will also affect the AI scores since Android 9.0 has new AI optimizations of its own.
Smartphone companies are putting AI in their devices as a means to get ahead of the competition. The usage of AI in a smartphone are many such as – obeying voice commands to open or close an app, clicking better pictures with AI enhancements in camera, AI Gaming mode to block the calls and message notifications and give a lag free gaming experience, AI System OS which watches the user’s usage of the device and apps and then optimizes the apps so that they consume less battery. But there are still limitations about what AI can do and how its performance can be measured.
That’s why the company has put a unified standard for testing of AI’s capabilities and major companies like Samsung, Qualcomm, Mediatek and Nvidia are cooperating with this idea. These companies together are developing a way to quantify AI performance across various implementations.
About how the benchmark works, it is divided into two sub-categories – Image classification and Object recognition. The Image classification test reviews test data which comprises of 200 images and is based on Inception V3 neural network whereas the second part of the benchmark, the Object recognition test reviews a 600-frame video which is based on the MobileNet SSD neural network. Both these neural networks are then translated a common term or language that is supported by the manufacturer of the AI through SDK. And if the chip does not support AI-related algorithms, and fails in these tests then AnTuTu marks it as unsatisfactory and unreliable.
Antutu has released a Chinese version of Antutu AI benchmark public beta today, which is available for download from Antutu official website https://t.co/ZLKczZyXac and China app stores now.
How does your phone score on #AntutuAIBenchmark? Download the new app and have a try. pic.twitter.com/c1WASkBwcf
— AnTuTu Labs (@AnTuTuLabs) January 25, 2019
However, it is still not that simple, so AnTuTu has made a guard to keep the phones from cheating in this test. The score is determined by speed and accuracy, meaning accuracy cannot be satisfied to give speedy by wrong results. In that case they fail the test. AnTuTu wishes to set industry standard for AI. But even so it is clear that measuring AI’s performance on a number is complex and seems impossible. And high or low scores on the benchmark doesn’t necessarily mean that the phone or AI is excellent or obsolete. AI is still in its early stages of development and the AI makers still have a long way to go.