Gmail Smart Reply generates answers to emails for you, and this futuristic tool employing artificial intelligence will actually be widely available this week.
Android and iOS users will be able to access this feature which has been unveiled on Tuesday, and no longer have to struggle to find the perfect response to the emails they receive from family, friends or co-workers.
Instead, they will get to choose from 3 alternative answers generated in real-time by Google’s deep neural network, integrated in the Gmail app.
The tool initially reads the e-mail from the Inbox, then it parses the meaning of words and phrases, by using “long short-term memory” (LSTM).
After this processing stage, it creates relevant short sentences, based on the original content of the message, and on prior responses given in similar exchanges. The chosen reply can be further edited by the user or sent immediately.
Google officials believe that this will greatly improve productivity and speed, when it comes to keeping track of emails on mobile devices.
Normally, it isn’t too convenient or comfortable to type replies or dictate them using such small gadgets. Therefore, automated, yet sensible replies composed thanks to machine learning sound like the best solution to deal with this problem.
Artificial intelligence has become the center of attention for several tech giants, such as Google, Microsoft, Apple and Facebook, and there is fierce competition between these companies to achieve supremacy in this emerging field.
Efforts are underway to simplify the way that consumers access media, to combat information overload and to generate deeper engagement with products and services.
The more users interact with software that employs artificial intelligence, the better neural networks become at understanding requests and catering for every need. Such knowledge is also employed by Gmail Smart Reply, and it is similar to the way natural language works.
The tool doesn’t just process the meaning of each individual word. It actually gets the gist of that communication, by “remembering” the beginning as it reaches the end of the email.
While this might raise fears that privacy would be violated when emails are read, it must be noted that the entire algorithm used by the email assistant is operated by artificial intelligence.
In the development stage of this system, Google did employ researchers in order to test the relevance and accuracy of replies generated by the program. That is how it was discovered that many responses were actually different ways of conveying the same meaning.
As a result, programmers had to create another neural network to identify these similar messages and remove them, so that eventually there could be enough disparity between the three options provided to the user.
Another issue which had to be corrected was that replies saying “I love you” were generated too frequently by the program, even in the most unlikely or inappropriate scenarios.
That is because the algorithm had noticed a higher frequency of messages such as “Sounds good”, “Thanks” and “I love you”, by studying email exchanges between humans. As a result, it came to rely on these stock answers when dealing with unfamiliar e-mails, which were more difficult to process or understand.
Experts fixed this error by making the prototype select responses which bore more relevance to the original message, instead of opting just for the most popular answers given in ordinary cases.
As a result, Gmail’s Smart Replies might be less lovey-dovey now, but their accuracy is so high that this upcoming tool is definitely worth testing.
Image Source: Google Blog