Robot - writer March 25th, 2016
The set of input data, determined by scientists in advance, included the gender of the main characters, a description of the plot, as well as a set of phrases and sentences that were to be used when writing the work. After this, the program independently generated the text of the novel, called “The Day the Computer Writes a Novel.” No other details about the operation of the algorithm are provided.
The text of the novel was nominated for participation in a literary competition named after the Japanese science fiction writer Hoshi Shinichi, and was accepted for consideration by the jury. Of the 1,450 entries submitted to the competition, 11 were written using various text-generating programs, but after four stages of screening, only one novel written using artificial intelligence made it to the final selection. At the same time, the jury members did not know that the author of the novel was a “robot writer.”
According to Japanese writer Hase Satoshi, The Day the Computer Writes a Novel is quite well written. However, the writer noted some shortcomings in the text written by the machine - in particular, poor character development.
This is not the first time that a robot has been used to write quality text. Previously, Automated Insights opened free access to the beta version of the “robot journalist” service Wordsmith, which allows you to create short notes written in natural language based on data sets.
Here's another interesting example.
Scott Horsley competes with a robot. Source: An NPR Reporter Races A Machine To Write A News Story. Who Won? NPR, May 29, 2015
In May 2015, Scott Horsley, NPR's White House correspondent and former business journalist, issued a bold challenge to Automated Insights' Wordsmith writing algorithm. "We wanted to see," writes NPR, "how our best pen would fare against the machine." Since NPR is a radio station, the bio-journalist was trained enough to write quickly. According to the terms of the competition, both opponents had to wait for the publication of the financial report of the Denny’s cafe chain and go ahead. Moreover, Scott had an advantage - after all, he was a regular customer of Denny’s. He even had a favorite waitress there, Genevieve, who knew his order by heart: a bacon and egg sandwich. Did not help. Although... how to judge.
The robot did it in two minutes, Scott Horsley - in a little over seven. NPR publishes both notes and invites the reader to sort of determine which text is robot-generated and which is human-generated.
Journalistic "Turing test". Source: An NPR Reporter Races A Machine To Write A News Story. Who Won? NPR, May 29, 2015
The robot wrote, of course, the wrong note. It obviously has a higher density of numbers and is drier in style. Whereas Scott, remembering either the menu or Genevieve, let in some lyrics that were not necessary for the financial report. For example, he inserted the phrase: “Rising sales show that consumers are willing to open their wallets for pancakes, eggs and hash browns.”
Formally, the robot’s dictionary is larger, since it must include the entire vocabulary of the language (this is over 1 million words for the English language). The vocabulary of an educated English-speaking person can reach 100 thousand words. But the robot must use the most frequent words, which makes its language drier. In addition, this robot’s vocabulary is also limited to financial specialization. It never even occurred to him that culinary or sports names (grand slam) could be used in a financial report. A person, on the contrary, is not limited by frequency and is free to use as rare and pretentious words as he likes, expanding the context and imagery. Moreover, a human writer is a writer because he uses original images. A robot simply doesn’t need this for a financial report.
“But that could change,” writes NPR. If the owner decides to feed Wordsmith arrays of more relaxed NPR texts and slightly tweaks the algorithm, then the “Wordsmith” will quickly rebuild its vocabulary, or rather, expand the boundaries and principles of its use. It's all customizable.
Who won the competition? The robot wrote faster, more efficiently. Scott Horsley, no matter how trite it may be, is slower, but more humane. The audience for this note is financiers. Is the lyrical insert about wallets and pancakes valuable to them? As long as readers are still people and not robots, then, probably, yes, it is valuable.
In general, it's a draw. Although two minutes versus seven... For radio, for financial market news, this may turn out to be critical.
The Yesenin competition between a foal and a locomotive was also held in the academic environment. In 2012, Christer Clerwall, a professor of media and communications at Karlstad University, Sweden, asked 46 students to read two sports reports - one written by a robot and one written by a human. The human note was reduced to the size of a robotic one, but the robotic one was also slightly corrected by the editor: headline, lead, first paragraphs - as an editor usually does in the media. Students were instructed to evaluate the materials on a number of criteria: objectivity, credibility, accuracy, tediousness, interestingness, clarity, pleasure to read, usefulness, integrity, etc.
The results showed that one note won in some parameters, the other in others. The human text received more points according to the criteria “well written” and “pleasant to read.” The robotic text, also predictably, received more points according to the criteria of “objectivity”, “clear description”, “accuracy”, etc. That means it's a draw again.
How do you like this story?
This earthquake went down in the history of journalism. On March 17, 2014, at 6:25 a.m., Los Angeles Times journalist and programmer Ken Schwencke was awakened by tremors. He ran to the computer, where a note written by his Quakebot algorithm was already waiting for him in the publishing system. Ken scanned the note and pressed the “Publish” button. So LAT became the first media outlet to write about the earthquake - 3 minutes after the shock. The robot journalist is ahead of his bio-colleagues.
The first earthquake note written by a robot.
Scientists have been trying to make robot journalists for a long time. There are also real advances in the use of mathematical algorithms for reporting. The robot processes traffic situations, weather forecasts, reports and other news topics. In fact, any data can be converted into coherent text, presenting the material for reading and publication.
This technology has been developed by scientific companies for a long time. In connection with progress in development, discussions are ongoing on the topic of possible competition between robots and real journalists.
Cyber journalist vs bio journalist
Computer science still makes up a huge part of what is included in the concept of journalism. The process is quite robotic in collecting information, processing and its subsequent publication. But a journalist, as a person, is more necessary for dialogue with the public. The presence of personality makes journalism come alive and this aspect cannot be replaced by a robot.
Accuracy is important when working with information, especially when the data is selected algorithmically. Robots make significantly fewer errors than people collecting data manually. The program carefully processes channels and sources and can compare data with historical information.
The main trends are more correctly identified, side information is taken into account, after which phrases and sentences are formed. Less adjustment required. But if the algorithm was the culprit of the error, a person must interfere with the code or redo the program.
Additional ethics, requiring human intervention, are necessary when creating a style. After all, it is possible to include various jargons in the robot’s vocabulary. Only human discretion can add prohibited or permitted techniques to the algorithm. The same can be said about the objectivity of coverage of an event, regulated by people. You cannot force a robot to be more objective than a journalist. Although an automated system can be trained to present a topic.
The robot is capable of generating up to two thousand articles per second, which cannot be compared with human work. Moreover, such publications do not have an animated author. Different opinions stop at the fact that every journalist should know how robots operate. And also to what extent they comply with professional ethics, helping in the work of the press.
Robo-journalism, the third threat
Considering the possibility of complete robotization of journalism in the coming decades, the elimination of the journalistic profession is not expected. The robot can only answer questions, and only real people have a real interest in the situation. The Internet has deprived journalists of authorship; modern marketing does not need the help of the media at all. And now comes threat #3 from the automated press.
Modern readers may not realize that some of the news was compiled by robots. Humans do not interfere with this process. This way the programs can process many thousands of quarterly profit reports. The video generation platform prepares hundreds of news clips and uploads them to websites and television.
Robots are highly efficient and can analyze and prepare much more reports than journalists.
Professionals are somewhat nervous and anxious about the prospect of losing their jobs. But with thousands of stories produced by robots, there have been no reported job losses for professional reporters.
A note on earthquakes and tectonic shifts
Robots can be useful by providing warnings of impending earthquakes. There is a well-known case of a robot publishing an article about a similar natural disaster in California. 3 minutes after the event, a message appeared in the newspaper with the date, indicating the radius of action and the strength of the shock. The note was signed by an algorithm, which was the only unusual fact.
There is a journalist behind this who used his program to write articles. Waking up from tremors, he discovered that the computer had already created a note. All he had to do was press a button, sending the article for publication. It has been proven that the programs being developed are quite capable of analyzing data on the movements of the earth’s crust, transmitting them to the press.
Generative journalism
Modern software can be used by anyone by downloading it to their computer. For example, games are being produced with a robot journalist. There are platforms that provide such programs and systems to various organizations. These are automatic generators of articles on the topics of profit reports, sports or for educational institutions. In America, the technology has been used by large news agencies and search engines for quite some time.
One method of automation is to create "branching paths." Intended text elements from single words to entire sections are added. Some of them are used in the required article or report, while others are deleted. The user enters results such as sports reports or financial reports into the program.
Based on the data, branches are formed that are subject to further construction and development in the form of a verbal structure. This reporting structure is the basis for many articles, each of which will be unique depending on the data entered.
Two arguments about the robot's unsuitable abilities
Whether there is a "disadvantages" of robots when creating news? Criticism of technology is undoubtedly present. For example, Internet users were asked to vote for articles from robots and journalists. According to the results, the human article turned out to be richer and more attractive.
There are debates that a robot cannot imitate a person's style and also show understanding. The debate is not without foundation, although robot manufacturers argue that a robotic article can be stylized to resemble the handwriting of a real journalist.
Roadmap for robojournalism
Some enterprises that develop and produce robots for journalism claim that the company is focused on personalized information. The platform does not strive to create one article that will be read by the masses. However, it can produce many reports that concern everyone and suit individual preferences. The article will be thanks to the data invested in it and the user needs, which he waits for and consumes.
In addition to the versatility of one article, there are programs for creating multiple reports. Manufacturers intend to bring into considerations and
Robotic devices are occupying more and more niches, threatening to displace human specialists in most professions. There is an opinion that artificial intelligence systems will never learn only one thing - creativity.
On the other hand, when transferring craft tasks to robots, people will have the opportunity to realize themselves in more interesting and enjoyable areas.
Continuing the topic of robot journalists, which we are talking about, we will look at specific examples of the introduction of AI in the media sphere.
Advantages and disadvantages of robot journalists
In the previous article, we worked on the Wordsmith platform, which is used by the Associated Press news agency. This product from Automated Insights, in research, showed results that largely exceeded the capabilities of traditional journalism. In addition to the fact that the agency increased the number of companies regularly reviewed in terms of income by 900%, it was also possible to improve the quality.
Out of 12 evaluation criteria, the software solution outperformed journalists on seven points, and only one of them was negative:
- imagery;
- utility;
- information content;
- accuracy;
- objectivity;
- disposition to trust;
- boring.
If a person is prone to subjectivity, then he is deprived of this; he is not able to be guided by emotions. If the program is properly maintained, the AI will never repeat a mistake once made. Plus, with better analysis, AI can better match facts and coincidences, finding stories that a human journalist would miss.
But texts written by people also have their advantages. According to the study, these are:
- connectivity;
- quality of presentation;
- understandability;
- enjoyable to read;
- interestingness.
According to experts, robots will never learn how to take and process goods efficiently. They are not able to identify questions that develop a topic or recognize sarcasm. AI shows itself best in writing those materials that only require a comparison of facts and their subsequent analysis.
Eg:
- financial reports;
- crime reports;
- weather forecast;
- coverage of the traffic situation;
- sports reviews.
The writing of materials by robot journalists, and the quality of their analytical publications is above average. It is not surprising that many large publications are already choosing to involve AI in such materials.
The most popular robot journalists
The main competitor of the Wordsmith platform is the development of the Narrative Science company called, which became famous for its publication in the Los Angeles Times back in 2014. Since then, the program has created volumes of text that are unrealistic for humans, and its algorithm has been improved.
Another star robot journalist is Heliograf, who “works” for The Washington Post. During the 2016 elections, he prepared more than 500 publications. What’s noteworthy is that the company’s management does not plan to monopolize its robot, but is considering options for leasing the device to competing publications, such as the Chicago Tribune and Los Angeles Time.
The popularity of another robot journalist called Syllabs is associated with the election races. In 2015, this company, together with the newspaper Le Monde, prepared 150 thousand texts about the results of regional elections in France in 4 hours. The first prototype of this program was created back in 2011, and in 2015 Syllabs was already completely ready for work. It can produce text in French, English and Spanish, and adding a new language is quite easy.
The leader in the speed of text production was a Chinese robot developed by specialists from Peking University. He wrote a 300-word piece for the Southern Metropolis Daily in just 1 second.
But not everything is so perfect in the work of robot journalists; they also sometimes make mistakes. For example, in 2015, the algorithm prepared an overview of the financial condition of Netflix. Using the official Q2 report, AI wrote that the company's shares fell 71%, missing investors' expectations. In fact, Netflix shares almost doubled, they just split. The robot journalist was not able to take this fact into account, whereas a human would have understood the situation.
Robot journalists: expert opinion
"One day the day will come when a machine will win the Pulitzer Prize"- Chris Hammond (Narrative Science).
“Robot journalists have already grown up to be considered a full-fledged part of the news media community. Whether you like it or not, our industry is at the forefront of this phenomenon,”- Laurence Dierickx (Universite Libre de Bruxelles).
“Thanks to automation, we prepare up-to-date income reports for 4 thousand companies quarterly, whereas previously their number was about 400,”- Justin Myers (Associated Press)
To summarize, we note that the direction of robotic journalism is extremely promising and is actively developing. By 2016, there were 11 companies developing AI systems for media, and their number is growing rapidly. Research agency Gartner predicts that by 2018, a fifth of all business content will be generated by algorithms.
However, there is no need to fear a full-fledged takeover of journalism by robots. Areas that require creativity and out-of-the-box thinking will still require human input.
Artificial intelligence is penetrating all areas of our lives, including journalism. Despite the fact that an autonomous robot reporter writing analytical texts is still something fantastic, the world's leading media have long been using AI and technology to create some materials, reports The New York Times .
Almost a third of the content published is created using automation technologies. Thanks to the system, which is called Cyborg, the publication creates thousands of news items.
For example, AI can analyze a financial report as it appears and instantly publish news that will include the most relevant facts and figures.
Uses robots to create news about US minor league baseball - AI is great at announcing game results and identifying the best players. AP began using technology to write materials back in 2014, signing a contract with Automated Insights.
In addition to baseball, robo-journalists at The Associated Press also cover the financial statements of global companies. Thanks to artificial intelligence, the number of such news increased from 300 to 3,700 in one quarter.
Last week, the Australian version released its first story written by artificial intelligence - it was devoted to the volume of donations to a particular political party. is currently testing a tool called Bertie, which will help correspondents by generating material templates for them.
It employs the Heliograf robot, which proved its usefulness during the 2016 Summer Olympics, as well as the US presidential election.
In addition, Heliograf was used for publications covering local elections in individual regions - this practice is called geotargeting.
“When it comes to international media, there is a risk of losing audiences who are interested in stories about their small communities. So we thought - how can we expand our reach?” - recalls the publication's director of strategic initiatives, Jeremy Gilbert.
The above publications also have an internal warning system if the AI encounters information that is not standard for itself. In this case, a human journalist analyzes the news feed and decides whether the publication will need an expanded version of the material that will complement the work of the robot.
What awaits journalists
AI journalism is not as simple as it seems - developers put a lot of effort into creating a program, which is then filled with a large number of templates with different versions and endings.
Despite the fact that there are mistakes in the work of artificial intelligence, experts welcome the use of robots in the media. First of all, AI does not require spell checking - it is “naturally literate” and follows all the language rules that are built into it.
Moreover, unlike regular correspondents who find monotonous work boring, the system does its job quietly and “never complains.”
However, robot journalists will not yet become a full-fledged replacement for people, says Lisa Gibbs, director of partnerships for The Associated Press. According to her, freeing a person from routine will allow him to spend more time on creative work.
“The work of a journalist implies creativity and curiosity. It's about storytelling, finding the facts, holding government accountable, critical thinking and evaluation - that's what we want our journalists to spend their energy on," Gibbs said.
And Dow Jones is experimenting with ways to get technology to do other tasks that make the newsroom's lives easier, such as translating audio into text or recognizing images that have been photo-edited.
“If a few years ago AI was a new technology used in IT companies, but now it has become a necessity for everyone. I believe that soon a huge number of journalism tools will be controlled by artificial intelligence,” said WSJ head of development and research Francesco Marconi.
Marconi compared the introduction of AI to the invention of the telephone and its ubiquity: “This is a new field, but technology is moving forward. Today it’s AI, tomorrow it’s blockchain, and in ten years there will be something else. The only thing that doesn’t change is the standards of journalism.”
Andrey Miroshnichenko
Cyber-journalist versus bio-journalist: draw so far
In May 2015, Scott Horsley, NPR's White House correspondent and former business journalist, issued a bold challenge to Automated Insights' Wordsmith writing algorithm. “We wanted to see,” writes NPR, “how our best pen would fare against the machine.” Since NPR is a radio station, the bio-journalist was well trained to write quickly. According to the terms of the competition, both opponents had to wait for the publication of the financial report of the Denny’s cafe chain and go ahead. Moreover, Scott had an advantage - after all, he was a regular customer of Denny’s. He even had a favorite waitress there, Genevieve, who knew his order by heart: a bacon and egg sandwich. Did not help. Although... how to judge.
The robot did it in two minutes, Scott Horsley in a little over seven. NPR publishes both notes and offers the reader a kind of Turing test to determine which text was generated by a robot and which was generated by a person.
The robot wrote, of course, the wrong note. It obviously has a higher density of numbers and is drier in style. Whereas Scott, remembering either the menu or Genevieve, let in some lyrics that were not necessary for the financial report. For example, he inserted the phrase: “Rising sales show that consumers are willing to open their wallets for pancakes, eggs and hash browns.”
Formally, the robot’s dictionary is larger, since it must include the entire vocabulary of the language (this is over 1 million words for the English language). The vocabulary of an educated English-speaking person can reach 100 thousand words. But the robot must use the most frequent words, which makes its language drier. In addition, this robot’s vocabulary is also limited to financial specialization. It never even occurred to him that culinary or sports names (grand slam) could be used in a financial report. A person, on the contrary, is not limited by frequency and is free to use as rare and pretentious words as he likes, expanding the context and imagery. Moreover, a human writer is a writer because he uses original images. A robot simply doesn’t need this for a financial report.
“But that could change,” writes NPR. If the owner decides to feed Wordsmith arrays of more relaxed NPR texts and slightly tweaks the algorithm, then the “Wordsmith” will quickly rebuild its vocabulary, or rather, expand the boundaries and principles of its use. It's all customizable.
Who won the competition? The robot wrote faster, more efficiently. Scott Horsley, as banal as it may be, is slower, but more humane. The audience for this note is financiers. Is the lyrical insert about wallets and pancakes valuable to them? As long as readers are still people and not robots, then, probably, yes, it is valuable.
In general, it's a draw. Although two minutes versus seven... For radio, for financial market news, this may turn out to be critical.
The Yesenin competition between a foal and a locomotive was also held in the academic environment. In 2012, Christer Clerwall, a professor of media and communications at Karlstad University, Sweden, asked 46 students to read two sports reports—one written by a robot and one written by a human. The human note was reduced to the size of a robotic one, but the robotic one was also slightly corrected by the editor: headline, lead, first paragraphs - as an editor usually does in the media. Students were instructed to evaluate the materials on a number of criteria: objectivity, credibility, accuracy, tediousness, interestingness, clarity, pleasure to read, usefulness, integrity, etc.
The results showed that one note won in some parameters, the other in others. The human text received more points according to the criteria “well written” and “pleasant to read.” The robotic text, also predictably, received more points according to the criteria of “objectivity”, “clear description”, “accuracy”, etc. That means it's a draw again.
But the most important thing that the study of the Swedish professor revealed , – that the differences between the average bio-journalist's text and the average cyber-journalist's text are negligible. This is a critical factor for assessing the future and present of robo-journalism. Cyber-skeptics always say that a robot cannot write better than a human. But this is the wrong approach. “Maybe the robot’s note doesn’t have to be “better”? What if she just enough good?" Professor Clearval shares his concerns with Wired.
Robo-journalism, the third threat
The Internet has liberated private authorship. Millions of people themselves inform each other about everything in the world. What’s worst is that it’s free, but with great desire. Yes, the Internet is full of junk, but we consume information carefully selected to suit our interests. Content on the network is filtered not before publication, but after - during distribution, thanks to the virus editor. As a result, the old media are deprived of their monopoly on agenda setting. So the matter will not be limited to the death of newspapers. The Internet threatens old media not so much with the transition from paper to digital, but with the involvement of the audience in authorship.
Another threat to traditional journalism is corporate media and other content marketing. Corporations also got the opportunity to become authors. This means that they need traditional media less and less as an intermediary. Corporations can now do it themselves.
And if among amateur authors the content is “improved” through collaboration (viral editor), then corporations improve their media presence due to competition for the public’s attention. In the media arms race, they lure media professionals away from them, apply innovation and, most importantly, move from direct advertising to socially significant topics. After all, brands need an audience: advertising only wastes the audience, but content is able to collect it. And although these processes are not very noticeable to the general public, the media of corporations, as they develop, causes no less damage to traditional media than the blogosphere.
However, the blogosphere and corporate journalism at least consist of people. A third threat is approaching unfortunate journalism, a soulless, inhuman one. If the blogosphere deprives the media of subscriptions, corporations deprive them of advertising, then algorithms threaten to take away the profession.
Until recently, the topic of robot journalism was very foreign. And in fact, for those who are familiar with the life of Russian editorial offices, it is quite difficult to imagine news algorithms in their everyday life. But at the end of October, Yandex announced that it was creating a news agency where robots would write news. Among editors, exhausted by the struggle with the Internet and losses, this topic caused a new attack of vague anxiety and, of course, rejection. “Okay,” say those who are a little familiar with the problem, “robots will someday write sports notes, financial analytics or weather reports. But they are incapable of more.”
Incorrect assessment. It’s not just that robots will “someday” write about the weather, finance or sports - they are already doing it with all their might now, and in mind-boggling volumes. Accordingly, the question is whether they will be “capable of more.” The immediate answer is: yes.
A note on earthquakes and tectonic shifts
This earthquake went down in the history of journalism. On March 17, 2014, at 6:25 a.m., Los Angeles Times journalist and programmer Ken Schwencke was awakened by tremors. He ran to the computer, where a note written by his Quakebot algorithm was already waiting for him in the publishing system. Ken scanned the note and pressed the “Publish” button. So LAT became the first media outlet to write about the earthquake - 3 minutes after the shock. The robot journalist is ahead of his bio-colleagues.
Since the exact dates have been named, it is not interesting to name others, I can only predict that we will face quantitative and qualitative competition with our cyber colleagues. In the quantitative competition, bio-journalists are losing right now. In terms of quality, we will yield within 5-7 years.
It is interesting that in the early stages of journalism's transition from people to robots, native editors will become the killers of the profession. Editors are forced to produce as much content as possible to increase traffic. A journalist has no time to deal with serious topics, he must send materials to the site: “motion for motion"s sake" - movement for the sake of movement. Journalism theorist Dean Starkman called this effect the hamsterization of journalism - from hamster wheel, squirrel wheel. Hamsterization of journalism in pursuit of traffic, he reduces the journalist’s time working on material for the sake of the quantity of material: “do more with less.”
One hundred thousand readers of one article is quality journalism. But is it necessary to fight for one hundred thousand readers for one article, if you can deliver one hundred thousand articles, each of which will bring ten readings? Who will the editor choose: a capricious journalist with a growing salary and three texts per week, or a trouble-free algorithm with a decreasing subscription fee and three texts per minute?
The Associated Press is not buying the services of Wordsmith because the “blacksmith” writes better than a person. But because he writes more and faster. So the debate about the quality of the text is completely irrelevant. Robots will take over newsrooms due to their economic, not literary, merits.
Robot journalists in Russia
As for the forecast for Russia, we first need to understand where we are now. Bot creation is actively developing in the political sphere and in commerce, but not in journalism. There are known interesting experiments of Sports.ru - perhaps the most innovative editorial office in this field. For example, a robot writes a sports chronicle there. Another robot writes short phrases and selects gifs for them, sometimes it even turns out funny.