“Lonely cities lie buried in the night, / they remember their fairest queen, / Falling under the setting sun, / to dream first…” This is a sentence from “Sunshine Lost the Window,” a collection of poems published by Microsoft’s “Xiaoice” in May 2017. Xiao Ice is said to have been trained more than 10,000 times in modern poetry by 519 poets since the 1920s to develop her poetic skills. It would take a human being about 100 years to read these poems 10,000 times.
The chief planner of the poetry collection said that xiao Ice, as a new poet, may not be enough. The original version of the poem presents all of Xiao Ice’s writing without any modifications. Misspellings in the poem are deliberately preserved, with parentheses marking the edges. In addition, some of the poems may feel repetitive: “We want the AI to be 100% original.”
Sooner or later, AI will enter the realm of content production, not in literature, such as poetry, but in media that produces information 24 hours a day. Starting with shallow, fast, short, formatted information that needs to be compared with a lot of data, AI is taking over editing.
Husniff searched several types of AI products serving different media as follows.
Reutres News Tracer removes fake from Social media
Recently, NetEase News “News Society Salon”, with the theme of “How ARTIFICIAL Intelligence will Reshape Journalism”, discussed with experts and scholars the impact of AI on the media industry. During the session, Haiming Wang, market development manager of Reuters, introduced the Reutres News Tracer.
For the past two years, Reuters has been using a tool called Reutres News Tracer on Twitter to monitor emerging News events and verify social media messages.
Its working process is: collect all possible news information, find the original source; Find the first sender; The reliability test is carried out to mark the authenticity through cross-comparison and “score” from 0% to 100% according to the verification results.
Features of Tracer include: real-time monitoring of social media news; Use algorithms to ask questions that journalists might normally ask, and backtrack on the authenticity of the news; Algorithms and machine learning are used to pick out what is news and what is advertising, rumor or general conversation, and pick out the main points, while stacking the same types of news into data. Provide reporters and editors with reliable material and resources.
Wang Haiming once used “Tracer” to witness the rumor “come true” : At that time, the news that a company was going to go public gradually increased from more than 800 clicks, and then was reprinted by BIG V or other media. When reprinted by the website, the rumor turned into news, but was clarified 24 hours later. “I witnessed the whole process because I received the first news source very early,” he said. “The whole chain of fake news spread was confined to a small area at the beginning and broke out with retweets and big V retweets from news websites.”
Chat news bot
At present, the mainstream approach in the media industry is to use AI robots as news feed tools, enabling them to carry the interactive nature of social media. For example, daily automatic push news; Users choose the news they are interested in according to the options given. You can also send keywords to get relevant news.
CNN has launched a “personality distribution”. The bot sends headlines to users with three options below: Read Story, Get Summary and Ask CNN. Users can Read the Story, Get a synopsis of the Story and Ask the bot questions.
On the basis of news push, some media have developed AI products with more journalist attributes: assisting reporting, collecting clues and materials, asking questions to users and collecting attitudes towards news events.
Buzzbot, developed by BuzzFeed, aims to put “a reporter in everyone’s pocket, and everyone can tell Buzzbot what’s going on.” When a user opens a dialog box, “Buzzbot” will ask a question, for example, about the Democratic Party convention and the election. Based on the user’s answers, it will record the user’s election attitudes to help reporters and editors analyze voters’ political preferences and opinions, and complete interviews and recycling tasks.
There are also more vertical-focused products, such as sous-Chef, the Guardian’s food chatbot on Facebook, which introduces and recommends recipes as users chat with it.
A writing robot that produces manuscripts in seconds
Heliograf, a news writing robot developed by The Washington Post, went live in 2016. In November, the bot reported republican Steve King’s defeat of Democrat Kim Vive: “Republicans remain in control of the House of Representatives, losing just a few seats and still holding a commanding majority. It’s a stunning reversal of fortunes after many Republican leaders feared double-digit losses.” The robot not only did “basic work” but also predicted election trends, which is considered to be in the style of the newspaper’s reporters.
Washington Post reporters and editors created narrative templates that included key phrases for various potential outcomes, such as “Republican control of the House” and “Democratic retaking of the House.” After that, “Heliograf” is associated with a structured data source. When the election comes, it connects to VoteSmart.org, a data-exchange site, and Heliograf identifies the data, matches and integrates the words in the template, and ensures that different versions of the story are published on each platform.
In November 2016, “Heliograf” created more than 500 articles and received more than 500,000 hits. The Washington Post received 1.1 billion hits that month.
Following the AI wave, Chinese media are focusing on “fast and template” news writers
After an earthquake struck Jiuzhaigou county in early August 2017, a China Earthquake Network robot wrote a press release in just 25 seconds. The manuscript consists of 540 words, including quick report parameters, introduction of epicenter topography and other contents. The robot combines digital technology and intelligent writing programming system: real-time monitoring of information sources; Information extraction; Machine learning algorithm is used to compose news by template and extracting information from knowledge base.
Earlier in September 2015, Tencent Finance published a story about the August CPI using its robot Dreamwriter, citing statistics bureau data and adding analysis and forecasts from experts and analysts. Later, “Dreamwriter” was widely used by Tencent Finance. According to the team, “Dreamwriter will automatically generate manuscripts according to the algorithm in the first time, instantaneous output analysis and analysis, and deliver important information and interpretation to users within one minute.”
In April this year, DreamWriter posted more than 2,000 articles per day on finance + tech apps.
Statistics about the 2016 Rio Olympics show that Xiaomingbot posts more than 30 articles a day. “The bronze medal match of women’s table tennis singles was held at the Rio Convention Center – Hall 3. Kim Song Yi, ranked 50th in the world, faced Ai Fukuhara of Japan, ranked eighth in the world. In the end, Kim Won the game 4-1 in 49 minutes. The scores were: 11:7, 11:7, 11:5, 12:14, 11:5. This message is written by “Xiaomingbot”. In 16 days, “Xiaomingbot” has completed more than 450 press releases and received more than a million views.
Right now, AI robots don’t have the ability to create great content, but how close are we? The Associated Press is already using deep learning to automatically generate data-backed articles, with the benefit of diverting resources to higher-value stories.
“Heliograf isn’t about making journalists obsolete, it’s about making the newsroom more efficient,” noted Selehi Prakash, vice president of digital product development and chief information officer of the Washington Post. At the moment, the hope for AI in the media industry is that it will increase efficiency, freeing journalists from basic tasks, allowing them to complete more complex topics, and allowing humans to tell stories that only humans can tell.
PS: this article is written by artificial, but it may be like the poem xiaoice wrote at the beginning, with wrong words, if there are mistakes, please point out, thank you.
mrpuppybunny
The tiger sniff network
The tiger sniff network
The tiger sniff the App