Released this week update of Android Wear to become the largest since the advent of the operating system. It brought many changes, such as the ability to launch apps directly on the watch, and Google Play, setting the dials with widgets, new text entry methods that support payment system Android Pay and the voice assistant Google Assistant. Also, Android Wear 2.0 has a “Smart answers” that allows you to quickly reply to an incoming message if you are busy. It turns out that it is based on machine intelligence.
In his research blog , Google has described the principle of operation of the function “Smart answers”. It analyzes the text and offers a different quick answers depending on the context. The interesting thing is that machine intelligence which analyzes messages and generates responses, is directly on device and not on the company’s servers, and therefore able to work even without Internet connection. While initially the developers believed that to realize this is impossible.
“We went back to the drawing Board to create a completely new and lightweight architecture is machine learning not only to activate the “Smart answers” on Android Wear, but also to open new opportunities in many other mobile apps,” reads the blog.
Initially, the developers wanted to use the existing neural network technology and so-called “student schedule”, but they are simply “not fit” on a smart watch, but an attempt to limit the number of responses did not meet their requirements. As a result, experts have created a simple system that groups messages where you can use such responses as, “Hey, how’s it going?” or “How you doing, buddy?”, don’t occupy a lot of memory. Next, use “paleobotany schedule” that test user responses to communications, similarities in words and phrases and also other factors to predict the best possible answers. And all this is done directly on the device.
“The model can also adapt to user writing style and individual preferences to provide a more personal experience,” explain the developers.
The developers were so surprised by the quality of this system that decided to apply it in other areas. Therefore, in the coming months this machine intelligence will appear in other applications of the company.