The development of global artificial intelligence is changing with each passing day, AI+ industry, AI+ medical care, AI+ manufacturing, AI+ education … … It can be said that a hundred flowers blossom. With the continuous evolution of the Internet and the high maturity of mobile intelligent terminals, the era of information explosion has quietly come to the public. Behind this massive information, a new kind of "media people" has emerged. They are not human beings, but "robots".
In October this year, Guangzhou International Artificial Intelligence Industry Research Institute officially settled in Nansha District, Guangzhou, and the institute took the lead in Nansha with 16 high-end talent teams in the field of artificial intelligence. Among the key laboratories, there is an AI R&D team focusing on the application of artificial intelligence and natural language processing (NLP) in the field of information content — — Yuntianyi, that is, the "writing robot" that everyone has heard about for a long time.
Besides playing Go, intelligent robots can also write manuscripts. As early as 2009, Major League Baseball used the intelligent robot StatsMonkey to complete the writing of the game manuscript for the first time. In 2015, Tencent, a domestic internet company, launched the writing robot DreamWriter, and the system has been in operation for nearly three years. Artificial intelligence has gradually integrated into the current content creation. How high is the "IQ" of the "writing robot"? What changes can artificial intelligence bring to the media industry?
In the future, with the in-depth development of the Internet of Things technology, experts in this field will paint a more "sci-fi" scene for all-media reporters under the background of the Internet of Everything.
Wen Jing, all-media reporter of Guangzhou Daily (except signature)
The number of intelligent writing robots is growing.
Speaking of intelligent writing robots, it can be traced back to 2009. For the first time, Major League Baseball used the intelligent robot StatsMonkey to identify major events during the game and summarize the overall game dynamics through statistical analysis, and automatically compiled a sports report.
In 2015, the "writing robot" really broke out in the industry, and major media at home and abroad began to "test the water" writing robot. It is understood that in foreign countries, The New York Times uses the Blossomblot system to screen articles and push them to social networking sites and other platforms; The intelligent system of Los Angeles Times focuses on dealing with sudden news such as earthquakes … … In China, in 2015, Tencent’s writing robot named Dreamwriter was officially launched, and a report entitled "CPI rose by 2% in August to a 12-month high" was released. Subsequently, the domestic media also "early adopters" one after another. Xinhua News Agency launched the robot writing project, Alibaba and CBN launched the "DT Draft King", and today’s headline launched "Xiaoming BOT" & HELLIP; … In the field of content production, traditional media and mobile internet media have followed this technical upsurge in succession, trying to introduce the manuscript-writing robot into the increasingly large number of news reports, content production and distribution; In the technical field, more and more research teams in the artificial intelligence industry are involved in the research and development of "writing robots".
Secret writing robot
So, how does the "writing robot" "write" a manuscript? Dr. Jin Lijie, CEO of Yuntianyi, a domestic AI news laboratory, said in an interview that "writing robot" is just a name that makes it easier for the outside world to understand. In the industry, they prefer to call this kind of artificial intelligence system "intelligent assisted creative system", which is also a more appropriate statement.
Taking the "chess writing" platform as an example, it quickly extracts content clues from massive information, generates an article framework with one click according to the keywords input by users, and automatically judges the content value from multiple dimensions. Users can search the hottest news topics through the list of "Hot Topics in the System" and generate articles on related topics with one click; You can also let "Xiao Yi" write a propositional composition by inputting keywords.
All-media reporters learned that the "writing robot" is different from the traditional news gathering and writing mode. The "writing robot" automatically "crawls" all kinds of massive information every day. When reporters and editors need a certain news content, they input keywords, and the system will "crawl" tens of thousands of relevant information from various channels according to the keywords and generate the first draft. On the "chess writing" platform, the active data of international information spans two years, and the whole information is stored for about 5~6 years, which also reflects the massive data behind the writing robot platform. In fact, this is also a prerequisite for being an artificial intelligence platform — — Only with a large enough data reserve, the robot can classify and learn, thus constructing the knowledge map of events.
Of course, different artificial intelligence platforms have different algorithms and ideas. For example, the AI system of Yuntianyi does not rely entirely on robots to write a complete manuscript. "Most manuscripts are post-processed and polished by editors."
"Robots" are mainly concentrated.
In sports and financial news reports
The research content released by Tencent Media Research Institute points out that "writing robots" can be divided into three categories, all of which rely on natural language processing technology. The first type is template-based writing, that is, using an article template to fill in the structured information (including specific figures, percentages, etc.) in the database. The second category is post-extraction, that is, analyzing long natural language texts containing useless information, extracting key information from the article, reorganizing the language and outputting it. The third category is that it is automatically generated by machines. Generally, the language is modeled mathematically by training the language model, and then the text that the algorithm thinks "best reflects the original meaning and looks like a sentence" is generated sentence by sentence without using the original text.
It is understood that at present, many writing robots, or most "intelligent assisted creative systems", mainly focus on the first type of template writing and post-extraction. His more dedicated and good writing fields are mainly sports, finance and disaster news. What these types of news have in common is that most of their contents rely on data to "speak". Take the news of NBA events as an example, most of them are composed of score+players’ scoring performance, that is, the news type based on data analysis. Therefore, there will be templates for this kind of news in the database of "robot". At the same time, through the deep learning ability of artificial intelligence, the system analyzes how journalists/editors handle this kind of news while writing it, so as to get more data and establish more diverse models.
All-media reporters have noticed that most of the financial news written by Tencent Dreamwriter is real-time news about the rise and fall of a stock price. According to Liu Kang, the head of Dreamwriter, the core problem that Dreamwriter had to solve when it was first born was that "the production capacity of financial news could not meet the demand of content".
Yuntianyi’s "chess writing" is an intelligent information writing system based on a wide range of topics. It is a combination of the second and third categories, and its writing scope is not limited to the news field that "speaks" by data. According to estimates, a small editorial team, with the help of the "Little Game Writing" intelligent creation platform, can improve the work efficiency by five times. "Traditional journalists/editors can probably write two long-length news stories a day, while in ‘ Small game writing ’ With the cooperation, about ten articles can be completed in one day. " In the whole process of intelligent assisted creation, the system can not only help to select the topic, but also "generate" an article frame with high correlation with what the reporter/editor wants to write, which can greatly simplify the writing process and shorten the time of writing.
In January of this year, the robot "Xiaoyi" was "employed" in the GET information App. Generally speaking, in the writing of traditional news information, the workload of editors accounts for 80%, and that of machines accounts for 20%. At present, the GET information team can achieve 65% machine completion and 35% manual completion. In the future, we hope to achieve 80% machine completion, and the editor is only responsible for 20%. "In our opinion, only a small part of the content can be produced by machines and automatically released, because any media needs supervision and content review. In the future, the manpower of the news information platform will also be mainly concentrated in this part. However, it is also inevitable to reduce labor costs and improve timeliness. For example, the work done by a team of 30 people in the past can now be completed by a team of 10 or less. " Jin Lijie said.
According to the reading data statistics of domestic large-scale content information platforms, with the help of "small game writing", the reading volume of many news written by GET information can reach tens of thousands to hundreds of thousands, and some even millions.
After training, it can be effectively improved
Content generation ability of "machine"
Many Internet companies and R&D teams have focused on writing robots, which is all related to the vigorous development of mobile news information platforms in the past three years — — The emergence of many news content clients means that a huge amount of news content information needs to be produced every day, and the update speed is maintained by minutes. On the other hand, it is the Internet users’ thirst for news and new information. Nowadays, on the way to and from work by subway or bus, everyone will "brush" their mobile phones, and the demand for content information will increase geometrically. It is worth noting that last year, Penguin Think Tank released the Map of the Future: Report on the Trends of New Media in China in 2017. The report shows that Chinese netizens spend an average of 67 minutes reading news information every day, among which the proportion of "heavy users" who spend more than two hours a day is 14.6%.
Nowadays, domestic large-scale content information distribution platforms pay attention to "thousands of people and thousands of faces" — — Through big data, algorithms, labels, etc., what everyone sees is different. Similarly, the "writing robot" can also push different content for everyone through the big data and machine learning ability in the background. Yuntianyi is like this: "Imagine that a manuscript-writing robot is like a fisherman, and every day ’ Fishing ’ It will classify and combine the returned fish, shrimp, crab, etc. through artificial intelligence algorithms. " Because everyone’s "entrance" to the intelligent platform is different, and his needs and content are different, the platform can "know" what content he needs and what content he needs more. In addition, the content provided by the machine for readers is updated 24 hours a day.
"The intelligent assisted writing system and automatic writing system driven by artificial intelligence can produce hundreds of information articles a day after training of massive high-quality content, which will greatly improve the production efficiency and timeliness of information content. Artificial intelligence technology has achieved innovation in some aspects of the life cycle operation of traditional information content. " Yun Tian Yi said so.
Discussion:
Can writing robots replace human work?
The industry thinks it is more auxiliary function.
At the beginning of the writing robot’s launch, "the robot came to grab the reporter’s rice bowl" and "the reporter was going to cry" … … For a time, it has attracted great attention from the technology and media industries. In recent years, whether the writing robot will replace human beings as "news reporters" has been increasingly fierce with the development of technology.
Ma Zhijie, dean of the School of Humanities and Communication, Guangdong University of Finance and Economics, and Ph.D. in journalism, said that advanced technologies, including intelligent robots, are only writing AIDS and tools are extremely valuable. "If you want to do a good job, you must sharpen your tools first." The news and communication industry will not disappear, but will only take the advanced technology express train and be stable and far-reaching. However, the news industry must rely on media workers who are good at using advanced technology, not the technology itself. Intelligent robots are highly integrated with cutting-edge technologies such as big data, cloud computing and AR/VR, so they have strong capabilities of data collection, extraction, calculation and information self-organization, which helps media workers to improve content production efficiency, but also strengthens media competition.
When talking about the advantages and disadvantages of "writing robot", he believes that compared with people, the advantages of intelligent robot content processing and production are obvious, such as large amount of data, many information points, fast production speed, accurate computerized production service and breakthrough of time and space constraints. However, the shortcomings are also very obvious, such as starting from data, purely objective perspective, "arrogant", and lack of freely controllable ability in information appreciation such as agenda setting, news commentary and public opinion guidance.
Writing robots can’t be a branch of the future media. Human beings have been on the road of using new technologies to improve the efficiency of content production. At present, the cutting-edge technology is intelligent robots, so it is understandable that intelligent robots are preset by media, but intelligent robots can only be referred to as "it" forever, because they are machines and intelligent technologies, and they can never become media, and the core of media is people with strong subjective initiative.
Dr. Jin Lijie said frankly that the original intention of R&D was to use the intelligent assisted creative system to free journalists and editors from some boring, hard and repetitive work of news reporting and focus on the core of content creation.
Dr. Ma Zhijie also stressed that the unemployment of journalists is not the result of technological progress, but the self-elimination of journalists caused by the industry’s advancing with the times. Some journalists and editors who can’t adapt to the requirements of the changed times will inevitably not adapt to the changed news and communication industry. Therefore, the unemployed will inevitably exist.
Related links:
Can’t catch up with China’s first AI anchor "on the job"
At the world internet conference in Wuzhen just held, sogou, a Chinese Internet company, and Xinhua News Agency jointly launched the world’s first "AI synthetic anchor". It is understood that the virtual anchor is based on a host of Xinhua News Agency, which can imitate not only human voice, but also human expression.
It is reported that AI synthetic anchor can achieve efficient news broadcast. Users only need to input the news text, and it can broadcast with the same voice as a real person, and even the lip shape and facial expression are consistent with the real host.
The specific operation is that the live anchor records a video for broadcasting news to the camera. With this video, "sogou avatar" can extract the voice, lip movement, facial expression and other features of the live anchor, and then "clone" an AI virtual anchor with news broadcasting ability that is very similar to the live anchor through technologies such as speech synthesis, lip synthesis, facial expression synthesis and deep learning.
According to sogou, "sogou doppelganger" uses its AI ability to personify AI from the aspects of image expression, sound language habit and logical thinking, and then "clones" human AI doppelganger, thus helping human beings to improve the efficiency of information expression and transmission.
关于作者