Artificial intelligence for writing texts in the media and literature

Main article: Artificial Intelligence

2023

Popular tech publication Gizmodo closes Spanish branch, replacing journalists with AI

On August 29, 2023, G/O Media, the parent structure of the popular Internet resource Gizmodo dedicated to design, technology and science, announced the reduction of all employees of the Spanish branch of the named site. Instead, artificial intelligence will be used to translate materials. Read more here.

German media holding Bild fires 200 editorial staff, replacing them with artificial intelligence

In mid-June 2023, the German media holding Bild announced a cost-cutting program of €100 million, which will lead to the dismissal of about 200 employees. The company warned workers that it expected further editorial cuts due to AI capabilities. Read more here.

What neural networks can already be used in the media today

Today, more and more people understand that the future lies with neural networks, and that things can be done on them that were previously impossible. Like any innovative product - to a wide audience, neural networks seem to be something of little use, but curious. They know how to write music, process and generate images, highlight the main thing, voice text, and maintain a simple dialogue. But after the first delight, everyone will play enough, and the novelty will become a working commonplace in all areas. For example, several ways were selected specifically for the media to potentially use neural networks to solve real problems.

The article "Media of the Future: What Neural Networks Can Be Used in the Media Today" presents the results of a study by experts who, based on their many years of experience in online media, analyzed: what could simplify journalists' work, improve the quality of materials and increase business efficiency. Read more here.

British newspaper used ChatGPT to write news and became frustrated with results

ChatGPT chat bot based on artificial intelligence, developed by OpenAI, is not yet suitable for the role of a reporter. This, as it became known in early March 2023, is evidenced by the results of an experiment conducted by a British newspaper. Read more here.

2022

MISiS taught artificial intelligence to write plays

The experimental model of the neural network - the generator of plays "NeuroStanislavsky" was presented in Moscow. The project opens the next era in theatrical art and in the field of neural networks: artificial intelligence has already created music and paintings, but for the first time in Russia he wrote a play. NUST MISIS (National Research Technological University) announced this on November 11, 2022. Read more here.

Neural network from "Sberbank" wrote a collection of short stories with writer Pavel Pepperstein

Neural network from "Sberbank" ruGPT-3 wrote a collection of stories along with writer Pavel Pepperstein, who published the publishing house Individuum. About this "Sberbank" reported TAdviser on May 24, 2022.

Neural network from Sberbank wrote a collection of short stories with writer Pavel Pepperstein

Trying To Wake Up is the world's first storybook to be born in as a result of cooperation between the writer and his "double" neural network. Of the 24 texts in only half of it belongs to Pepperstein, she composed another dozen a generative ruGPT-3 neural network further trained on Paul's stories and fragments of his interview.

The algorithm wrote texts in several calls, with intermediate tuning and additional training on classical literature: team first SberDevices set a vector for the text of the neural network, and then the most successful machine creativity options were selected by the editor-in-chief of Individuum Felix Sandalov and co-author Pavel Pepperstein.

Шаблон:Quote 'author = said Felix Sandalov, editor-in-chief of Individuum.

Шаблон:Quote 'author = noted Pavel Pepperstein, writer and artist.

In addition to creating the architecture of the neural network, its training and selection of texts, human intervention in the work of Pepperstein's digital "double" limited to the work of the corrector. Neural network compositions are offered to the reader in the form in which the neural network wrote them. Before writing your stories for collection, Pavel Pepperstein got acquainted with the texts written by the machine, so that not only the algorithm mimics the style of the writer, but vice versa.

In the collection, each story has one author, and readers are left with the opportunity guess who owns this or that story - a person or a car. In section with the content of the book readers are invited to put notes who wrote this or that story, and share your guesses with other readers.

The book "Trying to Wake Up" came out with a holographic cover with an image two paintings. One was written by Pavel, and the second by the ruDALL-E neural network, created by SberDevices and Sber AI with the assistance of SberCloud, which previously "studied" the work of Pepperstein the artist.

2020: Court first copyrighted text written by artificial intelligence

In mid-January 2020, a court in Shenzhen (China) ruled that an article created using artificial intelligence is protected by copyright. This is the first time a court has copyrighted a text written by AI.

For five years, Chinese tech titan Tencent published articles created by Dreamwriter's automated software, focusing on business and finance-related material. Tencent experts developed algorithms for AI that creates news content in 2015. On August 20, 2018, the Dreamwriter program wrote a financial report that was published on the Tencent Securities website indicating that the report was automatically generated by the Dreamwriter robot.

A court in China has ruled that texts written by artificial intelligence must be copyrighted

A few weeks later, an online platform run by Shanghai Yingxun Technology Company posted the same financial statement on its website. Since the text was generated by AI, the Shanghai Yingxun online platform considered it possible to copy it without permission. The company decided that the material compiled by AI does not have an author in the traditional sense of the word, that is, formally they did not violate copyright law.

However, the court found that the wording of the article had a "certain originality" and met legal requirements that should be classified as a written work. The form of presentation met the requirements of written work, and when creating it, the selection, analysis and processing of relevant information was required. Thus, the text created by AI was subject to copyright law.

Although the defendant has already removed the article from his website, he will still have to pay a fine of 1,500 yuan ($217).^[1]

2018: Machine translation system successfully passed newstest2017

On March 30, 2018, a group of Microsoft Research scientists announced that the machine translation system they created successfully passed the newstest2017 test, which includes 2 thousand proposals taken from professionally translated news articles. Such standardized tests are widely used in the development of artificial intelligence systems and help to bring objectivity to the assessment of their work.

External experts who speak both languages ​ ​ assessed the results of artificial intelligence when translating sentences from a sample, comparing them with two professional translations.

Recent discoveries in the field of neural networks have helped to achieve such success. At the disposal of scientists every year is an increasing set of various techniques, the combination of which allows you to improve the results. For example, one of the techniques of working with artificial intelligence is "double learning." The system translates the phrase from Chinese to English, and the person translates the result back into Chinese - approximately as if the teacher translates the phrase back for the student so that he understands his mistake.

Scientists still have a lot of work ahead of them: in particular, they plan to add new language pairs and achieve similar results in translating texts into broader topics.

2017: Neural network taught to write poems of a certain genre and on any topic

Researchers at Facebook AI Research (a division Facebook developing software for artificial intelligence) introduced a new approach to automatic poetry writing in early July 2017. The created system neuronets is able to write poems of a certain genre and on any topic, practically indistinguishable from those that a person could write. However, the main goal of poetry - the transmission of thoughts and feelings through the figurative means of language - still remains inaccessible to artificial intelligence.^[2]

Any poetic work has two important components: form and content. The form is expressed in the metric characteristics of the verse: rhythm and rhyme; the content is responsible for the fullness of the poem - lexical and figurative means used to convey a certain thought.

Can a computer compose real poems? Until recently, it was believed that not, but the neural network created on Facebook AI Research has practically learned to do this.

The authors of the new study proposed two models that are based on existing language neural networks. The first model extracts both the form and the content of the poem from the training sample, and gives the output a work of a certain poetic genre. Such a model was trained on a small (300 thousand words) sample of sonnets and was able to create a work written by five-leaf iambic (the size traditional for a sonnet in English).

However, such a model has one important limitation: it can create works only of the poetic form that was presented to it in the sample for training. Taking this into account, the researchers proposed another approach, which is to use a generating adversarial network, one part of which (the generator) is responsible for content and the other (the discriminator) for form.

The model of automatic versification was taught on a large sample of poetic works (7.56 million words) of various genres. Then the scientists set up an experiment: they took poems created by a neural network, mixed them with amusingly absurd poems by real poets and invited seventy readers to determine who wrote this or that poem - a person or a computer. Respondents correctly established the authorship of people in 51.4% of cases; in the case of computer poetry, this share was 53.8%.

Respondents also needed to note on a five-point scale the "readability" of poems ("How easy is the presented fragment to read?"), Their emotionality ("How much does a given fragment evoke emotions in you?") and beauty ("How much is a given fragment pleasant to read?"). It turned out that the respondents considered the most "human," emotional and beautiful poems to be precisely those that were created by a computer.

Here, for example, is one of the quatrains used in the experiment:

How dreary to be somebody,

How public like a frog

To tell one’s name the livelong day

To an admiring bog.

The meaning of the poem seems dark. The phrase admiral bog ("admiring mire") looks absurd, and what exactly expresses the publicity of the frog? However, it is not written by a computer. The author of this quatrain is Emily Dickinson ("I'm nobody! Who are you?»).

And here is another example used in the survey of respondents:

The frozen waters that are dead are now

black as the rain to freeze a boundless sky,

and frozen ode of our terrors with

the grisly lady shall be free to cry.

In this passage, the metaphor of winter as death and fear, understandable to the reader, is visible. The poem is dark, it is filled with various comparative means and seems quite emotional, but it was written by his computer.

Despite the fact that the work of the new neural network brought artificial intelligence closer to real poetic creativity, he is still far from real poets.

As the poet Rishi Dastidar noted in an interview with the Internet publication New Scientist, there is no subtext for computer poems: a neural network can put one word after another so that the result corresponds to a certain poetic model, and select them so that they relate to a certain topic, but such poems have no idea and aesthetic depth. In order for the works of a computer to express thoughts and feelings (one of the most important tasks of poetry in its everyday understanding), artificial intelligence must learn to think and feel itself.

You can get acquainted with the work of the neural network and evaluate it on the website neuralpoetry.getforge.io.

2016

Narrative Science: By 2025, 90% of all texts will be written by robots

By 2016, it uses robotic journalists and Forbes magazine, for which Narrative Science has created an appropriate specialized platform. Articles written by artificial intelligence are still very simple, but the co-founder of Narrative Science, Christian Hammond, believes in great prospects for this direction, believing that by 2025 the 90% of all texts in the world will be written precisely by artificial intelligence. The assumption is certainly bold, but some experts agree with this in part.

"In the future, robots journalists will well be able to be competitive in the labor market, especially in terms of creating simple news notes. At the same time, they can have a number of advantages: a clearly structured algorithm can be laid in the machine, according to which the text will be written, for example, what information to put in the beginning, which in the middle, and which in the end. This will increase the readability of news notes. Often journalists forget about this rule, providing all the relevant information already in the first lines of the text, "said Vladimir Sungorkin, general director of Komsomolskaya Pravda

Nevertheless, most experts are absolutely confident in the inability of artificial intelligence to completely replace a person in such a difficult task as writing interesting articles.

"The car is unlikely to be able to replace a person completely. Robots are not able to feel an interesting story - they are only able to answer the questions posed, "Yury Pogorely, executive director of the Interfax Financial and Economic Information Service

The book written by the robot passed the 4 stages of the selection of the literary prize

However, despite such significant skepticism, practice shows that the prospects for artificial intelligence are still not so vague. In the spring of 2016, the book "The Day the Computer Writes a Novel," created by artificial intelligence, was able to reach the final of the Hoshi Shinichi Literary Prize. Yes, the work could not win the main prize at the end, but the very fact that it passed four stages of selection speaks volumes.

The Hosi Shinichi Prize is primarily known for the fact that works written by the machine can be put up for competition. In 2016, out of 1,450 books, about 11 were written by artificial intelligence, but only one was able to qualify for the final. At the same time, the jury was not informed that it was considering a book written by AI.

was surprised by the work of artificial intelligence, as we have a really well-structured novel in front of us. However, he still has a number of problems that prevented him from winning the top prize. For example, the characters are not fully revealed. A number of studies have yet to be carried out on the solution of such a problem by the developers, "science fiction writer Hase Satoshi said at a press conference of the competition

Developers from the University of the Future Hakodate, where the novel was created, have not yet revealed the algorithm for the work of artificial intelligence. It is only known that the plot details, the characteristics of the heroes and the words and phrases used were originally set to write the work.

Auto Poetry Google

In May 2016, artificial intelligence Google after analyzing 11 thousand unreleased books, he began to write his first literary works. However, so far they are rather gloomy:

"He went silent for a long time.

He's tar for a moment.

It was quiet for a second.

It was dark and cold.

There was a pause.

Now it's my turn. "

2015: Yandex robots begin to write notes on the weather and the situation on the roads

In November 2015, the direction of preparation of automatic texts was opened by the Russian company Yandex. At this time, Yandex's artificial intelligence produces only short notes on the weather and the situation on the roads, but in the future, company representatives promise to expand the list of topics for publications.

2014: Associated Press launches automatic text preparation on companies' financial accounts

In 2014, the Associated Press announced that from now on, most of the news related to the companies' revenues would be created using robots.

"It has taken us a long time over the years to sort out the company earnings figures in preparing the items. However, the Wordsmith platform from Automated Insights allowed us not only to automate this process, but also to significantly increase performance, "said Associated Press editor-in-chief Lou Ferrara

The application of automated technology allowed The Associated Press to increase the number of quarterly company earnings news from 300 to 4,400. Later, the Yahoo News news portal used a similar service. In 2016, Associated Press robot reporters expanded their subject matter somewhat. They began to be trusted with small news notes related to the US Minor League Baseball.

2013: Yandex.Avtopoet service launch

Yandex and Google specialists conducted active research aimed at teaching the poetry machine. The Yandex.Avtoet service, which was created in December 2013, has been compiling poetic lines based on user requests and news headlines for several years. For these purposes, the robot was specially taught to determine poetic dimensions and rhyme lines among themselves. You can get acquainted with the full list of Autopoet's works here. One example of his poems (spelling and punctuation preserved):

"Echo of the Samara police,

the euro is losing ground,

Kursk faces flooding,

the dollar continued to fall "

Notes