Home > News content

Let Google senior researchers tell you: Allo is how to help you automatically reply to the message

via:博客园     time:2016/10/17 20:31:09     readed:1582

In the fall conference held earlier this month, Google released the most unexpected product is a mobile chat Google Allo. Google Wave, Google Buzz and Google+ and other social products have failed, after the transition to office IM Hangouts, Google decided to try again in social products.

Google Allo direct competitors will be WhatsApp, Facebook Messenger, iMessage, micro letter and Line, the IM application users add up to billions, then the social relations chain can not be basically broken circumstances how Google Allo and they Competition?

The answer is artificial intelligence. This mainly includes two aspects, one is the smart reply (Smart Reply), one is Google Assistant intelligent voice assistant.

The benefits of smart replies for social networking products have been highlighted in Google's Inbox, a mailbox-like application. In March this year, after the version update, Inbox can automatically analyze the mail you receive, and gives a series of quick reply options, such as "trying", "immediately sent to you" and "tomorrow". & Rdquo; and so on. This not only greatly improves the efficiency of people to deal with e-mail, you can also reply to content in the user when the way to train Inbox smart reply neural network system, for Google and the user is simply the best of both worlds.

151103-scenario151103-scenario

Inbox auto-reply demo

Greg Corrado, a veteran researcher behind Google's smart-reply technology, said: "AI is not a single thing, nor is it a technology that is alive in a system," said Greg Corrado, an expert on the subject of PingWest and other media interviews in Beijing on Friday. But a new tool for its application. & Rdquo;

In other words, intelligent reply to help Inbox changed over the past decade people use e-mail the way artificial intelligence applications, Google is being widely used in its other social products, such as Google Allo.

151110_p11_google151110_p11_google

Greg Corrado is a Senior Research Scientist at the Google Institute. His research interests include biological neuroscience, artificial intelligence and machine learning

There was a script that you use the QQ auto-reply "Oh" and you can chat on the day TB. In the case of the other party is indeed very stupid you can indeed do so, but normal people will not succeed. But joined the artificial intelligence technology Google Allo smart reply can help you do, maybe even 忽悠 a lot of people.

This is Google Allo in the face of other competitors when resorted to killer - mdash; both efficiency and fun, so IM application for different groups of people. After Snapchat relies on reading and burning features, and Line rely on the rich expression of the capsule was hundreds of millions of young users, the world's social networks are clearly aware of the importance of young users, even Apple's iMessage also recently joined the Textures and other entertainment features. The Allo's intelligent response not only to meet the workplace for a large number of work automatically reply to the needs of information, but also to meet the young users for the fun of the request, and more people close to Google Assistant also made a lot of contributions.

Corrado is a Senior Research Scientist at the Google Institute, where he focuses on biological neuroscience, artificial intelligence, and machine learning. In June, Corrado published an article at Google Researchpaper, Describes the intelligent reply behind the principles and techniques.

Corrado notes that Google has studied millions of emails and found that about 25% of those replies contain only 20 or fewer words, and that email is still an important way of communicating with billions of people around the world, Learning to help mailbox users to deal with increasingly complex e-mail is a viable thing.

In an intelligent response system, Google uses a technique called Long Term Memory Networking (LSTM) to analyze messages that can be placed on the local system, meaning that users' mail messages will not be sent to Google. The LSTM's work is to understand and analyze the contents of incoming e-mail messages in a semi-supervised manner and then use a feedforward neural network system to provide possible replies. For example, for the same e-mail, the system will give the following three "I suggest we meet tomorrow." Although these three sentences express the same general meaning, but there are still semantics Of the small differences, so this step needs to be manually selected according to personal habits.

google_allo_lstm2google_allo_lstm2

Google Smart Response process

Corrado said, "Privacy is important to consumers and to everyone, so it's important for businesses," he said, adding that privacy concerns for smart responses built into important social products such as Allo and Inbox are likely to be a problem for Corrado. The point is that users must be very clear and clear communication with the company's privacy policies and standards is what the user can use this information to decide whether to recognize and continue to use these products and services. & Rdquo;

Prior to this, Google and Facebook, Amazon, IBM, Microsoft and other five technology companies have announced the establishment of an artificial intelligence alliance (Partnership in AI), its main purpose is to answer questions about artificial intelligence and other artificial intelligence, the proposed direction of development , And collaborate with experts in the academic and ethical fields to propose artificial intelligence standards to be followed by future researchers.

Corrado said that one of the key responsibilities of the AI ​​alliance is to help users protect privacy and regulate the individual company's operations. "But the final decision is returned to the user and the consumer's own hands." At the same time a company if you want users to continue to use their own platform, they must find ways to win and maintain the user's trust. "He said.

We have a previous articlearticleIn the use of machine learning to improve the efficiency of iMessage through the end-to-end information transmission and differential privacy technology (Differential Privacy) to protect user privacy, and Google because the user information sent to the server analysis and trigger criticism . But in Corrado view,This is because Google and Apple in the privacy of this matter on the handling of different ways.

Corrado believes that this artificial intelligence alliance (Apple is not involved) can call on more companies in how to use technology to protect user privacy research, "the promotion of technology research, once the results, coupled with the user privacy protection for existing technology Of the support will be fruitful. For example, an HTTP page and an HTTPS page, we recommend that you use an HTTPS page if the user wants to have some more private conversations. All of the technology research and development on privacy protection is actually to provide users with more choices to deal with their relatively private information and data. "Says Corrado.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments