Google’s Inbox e-mail app has a new feature called Smart Reply, which is able to read incoming e-mails and send back an appropriate response.
Machine learning, which understands the essence of an e-mail and writes various reply options, is used by Smart Reply.
Computers use machine learning to recognise patterns and make judgements on their own with the help of a large mass of data. Machine learning can be defined as a type of artificial intelligence that develops algorithms to evolve behaviours based on empirical data. Currently, Google is investing a lot of time and money in machine learning.
The Google Now service and Gmail’s spam filters use machine learning to predict what type of information the user would prefer to see.
On Tuesday it was announced that machine learning will be part of Google’s Inbox app, and people can use it to automatically respond to e-mails. Smart Reply analyzes the incoming e-mail and gives three different reply options that the user can insert with a tap or click.
Bálint Mikló, a software engineer said that the new Smart Reply feature can save a lot of time – that would otherwise be spent typing – especially in the case of e-mails that need a quick response.
The Smart Reply feature has a pair of deep neutral networks. These are essentially algorithms that process data to figure out sentence structure, tone, and style.
According to Greg Corrado, senior research scientist at Google who works in artificial intelligence, scalable machine learning, and computational neuroscience, the system can handle new inputs a lot better than previous rule-based systems.
The first neutral network analyzes the text of an incoming e-mail to get the gist of what is being asked or said (which is the list of ‘thought vectors’). For instance, for an e-mail that reads “I sent you the documents”, Smart Reply would probably suggest the response “I’m working on them.”
Once the networks can learn from greater bodies of text, the Smart Reply feature will become a lot more accurate, scientists say. The system also makes sure not to suggest overlapping options that say the same thing, but use different phrasing.
Image Source: wired