The company most likely to develop "Her", Facebook's Artificial Intelligence Road

Zuckerberg often warns employees with "code is better than eloquence." In 2013, the â€œAlways Onâ€ slogan was hung on Facebookâ€™s New York office. Yann LeCun, who believes this sentence to be a creed, is leading the Facebook artificial intelligence team to overcome the brunt of artificial intelligence.

Today, artificial intelligence has become an important part of Facebook, and related technologies have been used in chat robots, face recognition, and personalized recommendations. At the same time, AI-related technologies are also open to the outside world, providing third-party developers with technical resources.

Â 2012 - 2014: Build a team and get rid of dependence

From 2012 to 2014, since Facebook's voice and graphics technologies were provided by third-party companies, in order to get rid of the dependence on technology providers, it took a series of acquisition measures, intended to swallow the core technology. On the other hand, as Google and Microsoft stepped up in the direction of artificial intelligence, Facebook also began to form its own AI team in full swing.

June 2012, Acquisition of Face Recognition Company Face.com

This is Facebook's second acquisition after the listing, this acquisition opened the curtain for Facebook to enter the field of artificial intelligence. Face.com is an Israeli facial recognition company that previously provided face detection and recognition technologies to Facebook and other companies. Its own camera application Kilk can automatically help users tag Facebook friends on photos. In addition, it can determine the age of the characters in the photo, but Microsoft's How Old with the same function did not start to fire until 2015.

Aug. 2013 Acquisition of Speech Recognition and Machine Translation Corporation MT

Facebook's information flow translation has been completed with Microsoft's Bing, in order to get rid of Bing's dependence, Facebook began to accumulate experience in content translation. Mobile Technologies (MT for short) focuses on researching speech recognition and machine translation technology. MT's Jibbigo application supports voice and text translation in more than 25 languages.

In September 2013, Yann Lecun joined heavily

In March 2013, Geoffrey Hinton joined Google and shocked the entire academic circle and industry. After Geoffrey Hinton joined Google, Yann Lecun, an expert in artificial intelligence, announced that he was heavily involved in Facebook. Zuckerberg himself took the lead of the Facebook artificial intelligence laboratory. Yann Lecun promoted the development of the convolutional neural network. Geoffrey Hinton's comment on him was that LeCun held the torch high and passed the darkest era.

In December 2013, the artificial intelligence laboratory FAIR was established

In the third month after joining Yann Lecun, Facebook formally established the artificial intelligence laboratory FAIR, Lecun at the helm. The laboratory is mainly devoted to basic science and long-term project research. 70% of the FAIR team's work is long-term research, 30% of the work is short-term product development, and the project is usually considered in a 5-10 year cycle. In addition, an application machine learning part AML was established, led by Candela. AML applies the research results in the field of artificial intelligence and machine learning to Facebook's existing products, usually only planning for six months. The former focuses on scientific research, while the latter emphasizes application.

March 2014, acquisition of Oculus VR and drone company Ascenta

Revenue Oculus is Facebook's very important and adventurous acquisition in recent years. With regard to the attitude of virtual reality technology, Facebook is very direct: â€œattackâ€ can be used as an â€œappleâ€ in the field of virtual reality. â€œShouâ€ can prepare for the next social age. A year later, Facebook acquired two VR companies, Surreal Vision and ARPebbles, to apply its technology to the Oculus Rift.

In addition, Facebook acquired British drone manufacturer Ascenta in the same period. This transaction is intended to assist Facebook in the implementation of a project called â€œConnectivity Labâ€ to bring Internet services through the manufacture of drones, satellites and lasers. In areas where the development of the Internet has been hindered.

Â 2015-2016: Focus on firepower, main attack chat robot

The FAIR team once claimed that they have an ambitious goal: to teach the machine the common sense and let the machine learn like a baby or an animal. LeCun mentioned that the biggest project of the current FAIR is for the natural language understanding of the dialogue system, and LeCun's phrase was vividly demonstrated between 2015 and 2016. Facebook's vast majority of artificial intelligence work is based on the dialogue system.

In January 2015, acquisition of speech recognition company Wit.ai; open source deep learning Torch module

Wit.ai 1 graduated from Y Combinator and developed an API interface that can create a voice-activated interface. After being integrated into Facebook, Wit.ai will work with Parse, a mobile development back-end service provider, to provide voice control development tools for Facebook, strengthen the voice input function of Facebook Messenger, and improve the level of semantic understanding. After that, Facebook established the language technology department.

At the same time, Facebook has open-sourced some modules for training neural networks more quickly on Torch, and Facebook hopes to promote the development of the entire deep learning field.

In March 2015, announced a new artificial intelligence system

After the open source neural network module, Facebook chief technology officer Mike Schroepfer announced a variety of new artificial intelligence systems that can identify the content and related background in videos and sentences. One type of artificial intelligence system can identify 487 different types of exercise, and can find small differences between different types of exercise, such as distinguishing between skating and ice hockey.

In June 2015, the Paris AI laboratory was established

After setting up an artificial intelligence research center in the United States in 13 years, Facebook set up a laboratory in Paris to tap into European artificial intelligence talents and build artificial intelligence machines that can understand massive amounts of data.

August 2015, the public killer feature Facebook M

Facebook has opened an artificial intelligence project called "Moneypenny" (abbreviated as M) , which can help users complete some information consulting work. For example, it is possible to complete reservations for hotels, shopping, etc. These are Microsoft, Apple's and Google's voice assistants are unable to complete now. The interactive mode is to enter the text of the desired service in the dialog box. Facebook M is a hybrid product and uses the artificial + artificial intelligence mode to serve the user. It is reported that Facebook M is a killer feature. Although it is smart enough, it is still in the test period.

In April 2016, launched a chatbot platform

At the F8 Developer Conference, Facebook developed Messenger as a platform that allows businesses and users to communicate via chat bots, and launched the Messenger chatbot platform at the appropriate time. Facebook plans to use "chat bots" to help users and businesses conduct automated communications and online purchases, and expand the use of Messenger applications in customer service and corporate transactions. There are now more than 10,000 chat robots on the Messenger platform .

In July 2016, the 404 laboratory was officially completed

Facebook has created a new 22,000-square-foot hardware lab called Area 404 at its headquarters in Monroe, California. Facebook's infrastructure team, the Oculus team, and the connectivity group all develop data center equipment, virtual reality devices, drones, and satellites. It is reported that Facebook's innovative products in the next 10 years will be completed in the 404 area.

Â The prototype of "Her"

Facebook's measures in artificial intelligence are not as extensive as the Internet, and they are just as powerful as Google. Facebook is a project that focuses on advancing dialogue systems and integrates them into existing applications. Every step is implemented.

What is worth looking forward to is that the number of Facebook users continues to increase, and the large and irregular life data generated has caused Yann Lecun more headaches, but it has to be said that these data closely match the data required for the training of human-computer dialogue systems. . Boldly think about it, when Facebook's data analysis level and computing power have increased significantly, it is likely to be the first company to develop "Her".