AI+ Writing Let Editors "Laid off"? Can only be human assistant at the present stage

Science and technology dynamic writing is the essence of language and the soul of culture. It is incompatible with the data composed of 1 and 0, words and symbols. How can the product of inspiration, intuition, and skillful language training be composed of data and code? ? In the wave of AI sweeping across all major industries, writing, as a form of art, seems to have become an "island" in people's minds.

However, this "island" was also "invaded" by the AI ​​wave. Recently, an e-commerce platform has launched an AI writing system called “Shakespeare” that can “write” thousands of copies in one second. People in the industry joked that it was finally the editors who were “off the job”.

Is AI+ writing really powerful?

Don't panic! "Shakespeare" is a "copywriter processor"

According to the press release issued by the e-commerce platform, the system is based on the traditional NLG and language model methods, and based on the big data accumulated by the platform itself at the level of product labels and search databases, the system analyzes and trains the structure at the sentence level. The model and language are generated so that one thousand words can be spit out in one second, and various styles can be selected according to the different needs of users. According to reports, the system can also optimize machine algorithms based on the user's choice of text.

However, when the system was first launched, it was met with netizens' tantrums: “Since we can intelligently write a copy, can we intelligently praise it?” “Is the artificial intelligence + copywriting a thousand people or a thousand people?”

In response, the person in charge of the system stated: “Traditional manual creation is time-consuming and laborious, especially during the “promotion” of the e-commerce platform, the time is tight and the task is heavy. Many businesses are often hard-pressed when they modify pages. This system is designed to greatly empower Businesses, improve efficiency, reduce labor costs, and gradually realize machine copywriting, system layout pages."

Obviously, the main application scenario of this system is the writing of commodity copywriting. There is still a long way from people writing about poetry and writing articles. The reporter found an article written by artificial intelligence. Although it seems to be logically fluent, its main content comes from an exclusive interview with the news media. The work done by the machine is only the deletion and interception of the manuscript.

“Actually, AI technology still has considerable limitations in its application to writing.” Lin Tingyan, an analyst at Tuoba Industry Research Institute, told a reporter from Science and Technology Daily, “AI writing requires the input of a large number of articles of a specific type for machine learning training. In order to let AI learn the cost of training that includes time and money, we can learn how to write certain types of articles." To a certain extent, the system is more like a rich copy database, and it can be recommended to users based on different styles. The text that is suitable for this product.

Limitations: Can only "expand" in specific areas

It is not the first time that AI has "stepped on the doorsteps" in the field of art. Google used AI to create a painting that eventually sold for $8,500. In one test, the paintings created by AI successfully “deceived” the eyes of more than half of the audience, causing them to mistakenly believe that the work was done by people.

There is no shortage of explorers on the road to AI writing. In 2016, an AI robot once participated in the composition of the college entrance examination essay that year, but the results can only be described with horrible. The two writing styles were flawed, the logic between the sentences was also fluent, and the words were monotonously repeated. Only a few sentences were considered to be somewhat “artistic literature”. Unfortunately, in the process of contextual integration, they lost their flavor. In fact, Japan has also used AI to participate in examinations for higher education. After accumulating a number of failure experiences, it has been able to achieve the level of university admission. However, it is also good at mathematics, history and other disciplines, but it is very general in language processing ability.

However, in the news news writing, report data collection, etc., AI writing has achieved a lot of success. In 2007, a U.S. company launched a news writing software for writing financial and sports news releases. As long as you import the latest data, you can generate up to 2000 reports per minute. In recent years, Chinese media have also begun to use artificial intelligence technology for news writing.

Lin Zhenyun said that for the AI ​​Institute to write, it must lock in a specific application target, conduct a large amount of data collection and model training; the larger the training range, the less effective and accurate the training will be. “As a result, AI technology can currently only generate texts in areas with similar styles and small vocabulary usage, such as weather forecasts, financial news, etc. AI can output relevant reports and news well, but it has innovations for those styles. Requirements, emotionally rich articles, AI are powerless."

To be improved: At this stage only "do not understand to understand"

"At this stage, we use more NLG (natural language generation) technology. Take this platform artificial intelligence system as an example. It uses a large amount of data to accumulate templates, and ultimately requires manual intervention to correct it." Tsinghua Lin Hui, director of the University of Socrates Institute of Large Data Center, told a reporter from Science and Technology Daily.

More importantly, artificial intelligence at this stage is just "do not understand and understand." “This is just like our voice processing method. The machine learned through big data analysis that certain words and phrases can be used together. The content that artificial intelligence writes out is just what people seem to think the machine has learned about. But in fact the machine does not know." Lin Hui said. For example, like the artificial intelligence sci-fi movie “Her” released in 2013, when the heroine said “I love you”, she only uttered a syllable but could not fully understand the emotions behind her.

Lin Hui introduced that due to technical limitations, the current AI + writing is still relatively low-level, on the one hand, it requires the participation of manpower; on the other hand, the machine still has obstacles in natural language processing and understanding. From the current point of view, the main purpose of AI writing is not to replace manpower but to serve as a person's assistant to help people deal with some tedious and repetitive labor. Some professors of journalism also wrote that the artificial intelligence that will be replaced by artificial intelligence will have clear rules, but in the news field, in-depth reporting, news review, and other parts that need creative thinking will be the areas that humans are further pursuing. This is The machine cannot replace it.