This month Google has just introduced Gmail’s Smart Reply used for iOS and Android. Depending on the content of the email, the cyber intelligence as well as neural networks of Smart Reply will propose up to 3 possible replies to the email received.
The offers from the system will be polished after studying from responses of the users to suggestions the networks gave. Smart Reply can be downloaded from Google Play and Apple’s App Store, and its available language is English.
According to Emma Ogiemwanye, the spokesperson of Google, now the source of Smart Reply has reached the number of 20,000 and it will increase day by day. Susan Schreiner, a senior editor and analyst of C4 Trends agreed that the launch of this feature would give users a hand to when they were too busy to do all “communications-related” activities.
The system of Smart Reply works depending on two recurring neural networks. One of them is the encoding network, which is used to scan every word of the email arrived and then gives a vector – a list of numbers that summaries the main point of the email.
The vector has no dependence on the sentence structure. For instance, the two questions “Are you free on Monday?” and “Does Monday work for you?” can produce the same vector.
After that, the other network will use the vector – or can be called a meme – to generate a response with correct grammar “one word at a time.”
In fact, long short-term memory (LSTM) network architecture is used for the neural networks as it will be able to function even in case of long delays. Moreover, it can deal with signals whose components have both low and high frequency. This LSTM architecture will aim at the most useful part of the email that enables response prediction.
Also, a machine learning system was developed to create meaningful responses with nature language. Schreiner of C4 Trends believed that though the system could not always interpret memes correctly users would be tolerant towards some errors of it at first because “for anything, there’s a learning curve”. Google states that all the privacy of information included in the emails will not be violated as there are no humans read them. The machine learning system which is used to analyze the data was set that they could not read.
However, Mike Jude, program director of Frost and Sullivan seemed not to be in favor of the idea. He thought that email’s auto-reply was not really necessary. According to him, an ideal “digital assistant” would be able to “answer your phone, take a message, and then email the message to you”. It could even sort the emails out and put a suitable label on them such as “urgent, routine and garbage”. Normally people do not need email’s auto-reply as “most people are becoming fairly thoughtful as to what they put into [emails]”, Jude noted. He also said that it would take more time to open emails and see the suggestion of the machine, as it is like playing “an unending game of 20 questions.”