A whirlwind of generative artificial intelligence (AIGC) is sweeping many fields.
At the end of 2022, OpenAI released ChatGPT, a natural language dialogue application, and iteratively launched GPT-4 in March this year, which quickly attracted the attention of various industries and the public. Large-scale models are booming all over the world, capital, technology, and talents continue to pour in, and technology companies continue to deploy their own large-scale models. It is estimated that by 2030, the AIGC market size is expected to exceed one trillion yuan.
A few years ago, the news that AlphaGo defeated the Go player Li Shishi triggered a wave of artificial intelligence. However, the wave of AIGC is more violent, because in addition to technological breakthroughs, its low threshold and practicality make the perception of the application side more significant. Therefore, the public does not just stay in the discussion stage, but can directly experience the power of generative artificial intelligence.
Every round of technological innovation will outline a new era. "In the era of AIGC, all industries are worth reshaping with AI." Affected areas include education, finance, e-commerce, film and television, design, etc. Among them, journalism is one of the most severely affected areas, and the response to AIGC is also the most positive.
Internationally, many media have already carried out related attempts. The news aggregation website BuzzFeed released quizzes, a test column answered by AI, and said that it will use AIGC to write test content to replace part of the manpower. On May 24, The Washington Post announced the establishment of a cross-departmental AI collaboration mechanism, including the strategic decision-making team AITaskforce and the executive team AIHub, to better adapt to AI innovation practices. The British "Financial Times" also appointed an AI line editor for the first time to pay close attention to the latest developments in this field. More than 100 domestic media organizations, such as The Paper, Cover News, and Upstream News, announced their access to AIGC products in February this year. Tencent Research Institute conducted a survey on the topic of "ChatGPT's Impact on Journalism" in June this year. The results showed that 80% of domestic journalists had used ChatGPT or similar products, and more than half (56%) of them were affected by it. Interviewers believe that ChatGPT (or similar tools) has brought practical help to their work.
While the ability to generate text is outstanding, ChatGPT also has the ability to generate multimodal content. For journalism, where content creation is at its core, this will bring about many intersections with significant implications. This technological breakthrough that is "comparable to the industrial revolution" (Microsoft CEO Satya Nadella) and a technological form that is no less than "the birth of the personal computer or the Internet" (Bill Gates) what will it lead to? What kind of change? Will it bring new opportunities for journalism?
Through this report, we attempt to discuss the impact and challenges that AIGC technology represented by ChatGPT has brought to the journalism industry, as well as the new possibilities it creates. AIGC is starting a "technological revolution", and journalism is one of them. Through this incision, we may be able to capture the true meaning of this revolution for human society.
The Great Reshuffle: A Triple Transformation of Journalism
Since the turn of the millennium, journalism has experienced bright moments full of hope, and has also been mired in pessimism.
The brand-new digital ecology endows traditional media with vigorous development vitality, and at the same time, it also gave birth to a group of digital media upstarts. However, in recent years, various factors such as the change of flow logic, the impact of new media forms such as short videos, and the shrinking of online advertising revenue are causing the news industry to fall into a new predicament.
(1) The "traffic era" is over, the news media has been reshuffled, and connecting readers has become extremely important
The logic of content distribution has evolved several times.
After the financial crisis in 2008, advertisers' advertising investment shifted from traditional media to online media. Search engines represented by Google and social media represented by Facebook have completely reshaped the content traffic pattern. Traffic originating from these two ports becomes the most important source of exposure for online media.
Of particular concern is Facebook, whose founder, Mark Zuckerberg, was a firm believer in the positive value of news content: enhancing the reputation of the platform and improving user retention and engagement. Therefore, Facebook once vigorously strengthened the recommendation proportion of news content, allowing relevant content to gain more exposure. The decade between 2006 and 2016 saw a honeymoon period between social media and journalism.
According to a 2015 survey by the Pew Research Center, as many as 64% of Internet users get news from social media. This is the "Journalism Traffic Era" created by social media, which has created a large number of digital media upstarts. The business models of BuzzFeed and VICE, the most well-known digital media in the 21st century, are based on the viral spread of social media. Huge traffic and user attention poured into these media, followed by a large amount of venture capital. At their peak, BuzzFeed and VICE were valued at $1.7 billion and $5.7 billion, respectively.
But for media outlets, the foundations of this model are fragile, and their survival depends entirely on the platform. Once the algorithm and rules of the platform are changed, the business model will be severely damaged, and it is completely out of control. The turning point of the story happened in 2016. During the US presidential election, Facebook was questioned by the outside world for using algorithms to manipulate the election results. The "Cambridge Analytica" incident directly sent Mark Zuckerberg to the hearing. Faced with criticism from all sides, Facebook announced a reduction in the proportion of news content. In 2020, Facebook will further strengthen relevant measures to significantly reduce the push of news content and political content.
This is not just a turn of the Facebook platform, but the overall trend of social media. The adjustment of the algorithm has resulted in less and less exposure of news content, which has severely hit the media that rely on social media traffic, resulting in a collective dilemma for the industry. In 2023, BuzzFeed founder Jonah Peretti announced the closure of its news business BuzzFeedNews, VICE announced the closure of the news brand VICEWorldNews, and its main website is also planning to file for bankruptcy. VoxMedia, Insider, ABCNews and other media have all laid off employees to varying degrees.
The common problem of these media is that they have not accumulated a user base by strengthening paywalls and subscription services. In the face of huge traffic and advertising revenue when the big wave comes, no one will think this is a problem. But when the tide goes out, you can find out who is swimming naked.
In contrast, when upstarts such as BuzzFeed were in the limelight, the old media "New York Times" was looked down upon by the industry because of its slow adaptation to the Internet, and was even regarded as a representative of the revolution. Spurred on by BuzzFeed, The New York Times was forced to transform itself into the larger context of digital journalism. But this pace is not in a hurry. It puts its own content behind the paywall, and attracts subscribers by deepening the content, but it becomes the capital that can survive the cycle.
In the era of frothy and fleeting traffic, it has never been more important to build a closer connection with readers based on core readers. More and more news outlets are recognizing this, and it has spawned three new trends:
**First, use new media forms to establish a direct connection with readers. ** In recent years, podcasts and RSS reading have begun to rise, and many media have launched their own RSS subscription services and podcast brands, aiming to establish a close relationship with readers more directly and effectively, strengthen media brands and enhance commercial value;
**Second, focus more on local content rather than global hot topics. **For example, the media group MvskokeMedia adjusted its editorial strategy to focus on local community reports, reflecting its focus on core readers;
**Third, strengthen the transparency and openness of reports. On the one hand, it helps readers understand the concept of reporting, and on the other hand, it also reversely understands what kind of news reports readers need more. ** For example, the Honolulu Civil Beat held "pop-up newsrooms" similar to pop-up events in its area to strengthen communication with readers.
(2) With the rise of "short video journalism", the audience's attention has shifted, and traditional news concepts have been impacted
In 2023, the data statistics agency "PressGazette" (PressGazette) released a ranking list of 25 media information companies established since the millennium, of which Facebook topped the list, followed by TikTok. The influence of social media is self-evident. The rise of short video platforms represented by TikTok has had a profound impact on the development trend of the journalism industry.
TikTok is quickly becoming one of the largest content platforms and traffic bases in the world. Not only are a large number of young audiences gathered on the platform, but audiences of different age groups are gradually turning their attention to short videos instead of graphic content or serious news reports that news media are good at. Simultaneously with the diversion of audience attention are advertising revenue and venture capital, which also flow to TikTok and Instagram platforms that young users pay more attention to.
Not only that, but a new news format is emerging: "TikTok journalism". When major news events such as the new crown epidemic and the conflict between Russia and Ukraine occurred, people discovered that the main source of information was no longer the news media, but TikTok. A large amount of real-time and first-hand video content is spreading rapidly on TikTok. TikTok has gradually transformed from an entertaining short video platform to a comprehensive content platform including audio and video content, and has become an important way for Internet users to obtain information. For many young audiences, it's not that they don't watch the news anymore, they just don't read the news in the news media. Similar phenomena are also evident in China.
When the main carrier of news changes from text to video, this is a challenge for most news media. Some traditional media are actively seeking changes and trying to integrate into the short video news ecology. According to statistics from the "2022 Digital News Report" released by the Reuters Institute of Journalism at Oxford University, about half (49%) of mainstream media organizations regularly publish content on TikTok. As a representative of established media, the "Washington Post" specially hired a third-party team to produce content for its own TikTok account; the "Los Angeles Times" formed a content team called "404" to conduct experimental content production based on the preferences of young audiences.
Some native news outlets have also sprung up from TikTok journalism. For example, NowThis, which started with short video news, already has 8.5 million fans; Spanish content company Ac2ality takes "telling news in one minute" as its core concept, and has accumulated 3.9 million fans on TikTok since its launch in 2019.
The rise of TikTok journalism means a shift in the focus of audience attention. On the one hand, the impact on the news industry is reflected in the loss of advertising revenue and the deterioration of the media's living environment. Although efforts have been made to integrate into the short video news ecology, the inherent incompatibility between traditional news reports and video media has made this transformation ineffective. restricted. On the other hand, the widespread impact of short video news has impacted traditional news concepts. Values such as "objectivity" and "authenticity" that are regarded as the standard in the news industry are no longer emphasized. Fast, sensational, and visual impact have become new production standards. Data such as likes and retweets have become new indicators of news quality. "Yellow news" has gained more and more traffic and audiences, and the living space of traditional news has been further squeezed.
From the perspective of the audience, people are gradually accustomed to obtaining news and information through channels such as short videos, which is also affected to a certain extent by the increasingly prominent phenomena of "news fatigue" and "news avoidance". News avoidance is driven by both cognitive and emotional factors: the cognitive aspect is manifested in the perception that certain topics or events are reported too much, and reading these news will lead to a sense of fatigue, and it is difficult to obtain information increments, resulting in "news overload"; The emotional aspect refers to people actively avoiding news that triggers negative emotions, such as reports on epidemics, violence, and natural disasters.
In 2017, 29 percent of respondents said they "often or sometimes avoid the news," according to a report by the Reuters Institute of Journalism and Oxford University, a figure that rose to 32 percent in 2019. After the outbreak of the new crown pneumonia epidemic in 2020, people's demand for news surged briefly, but the phenomenon of news avoidance rebounded quickly, with 59% of people saying that they "sometimes or always actively avoid news". The audience's information receiving habits and changes in mentality have become factors that the news industry has to consider, and have also become obstacles to the transformation of news media.
(3) Shutdown and layoffs have become the norm, and journalists actively seek changes
Three years of the new crown epidemic have had a huge impact on the global economy, and the journalism industry cannot stay out of it.
The closure of news organizations became the norm. Foreign media, including BuzzFeed, VICE and other digital media, have closed their news businesses, and print publications such as The Livonia Observer have substantially ceased publication. The domestic situation is also not optimistic. The "Media Blue Book: China's Media Industry Development Report (2022)" jointly released by the School of Journalism and Communication of Tsinghua University and other institutions shows that the impact of the new crown pneumonia on some areas of the media industry is still continuing, and the advertising revenue of traditional journalism continues to decline. Budget for periodicals, newspapers and other media. Domestic newspaper advertising and distribution revenue fell sharply, and the TV advertising market was weak and declining. From 2020 to 2023, dozens of newspapers, including "City Pictorial" and "Southeast Express", announced their suspension or suspension of publication.
The income of journalists has dropped significantly. Economic uncertainty has affected the jobs of about two-thirds of journalists, according to the Press Gazette. More than 80% of the respondents are full-time journalists, most of them (71%) have an annual income of less than $100,000, and the average freelance writer's fee is less than $300.
Layoffs set the tone for media organizations. According to incomplete statistics, since 2020, dozens of media have announced layoff plans. BuzzFeed cut off a certain percentage of employees due to its plan to use AI to generate quiz content. On April 20, 2023, the founder of BuzzFeed once again announced the closure of its news business, laying off about 180 employees, involving content, technology, administration and other departments. 15% of the total. According to Forbes statistics, since January 2023, more than 30 newspapers and media organizations have carried out layoffs of varying degrees. The latest occurred on June 7. unique challenges” and will lay off 74 newsroom employees.
The global economic downturn and technological shock are causing the news industry to face a double crisis, and the living conditions of journalists are worrying, which is also closely related to the application of new technologies by news organizations. The introduction of automatic reporting and automatic editing systems has liberated part of the manpower, but also caused some practitioners to become redundant. The iteration of media technology has not brought about progress in labor relations, which is especially evident in the content industry. In the first half of 2023, a strike action continues to be staged in Hollywood, affecting many on-air dramas. Behind the strike is the transformation of the production method and broadcasting form of drama series by the streaming media platform represented by Netflix, which squeezes the living space of screenwriters. The substitution effect of technological mechanisms also impacts journalism.
Faced with limited living space, many news organizations and journalists began to shift their focus to social media and short video platforms. For example, opening channels and publishing video content on TikTok and YouTube to attract younger audiences while increasing revenue through ad-sharing models. On the other hand, journalists share knowledge and insights through social platforms such as Twitter and LinkedIn, and build personal brands and influence.
Emerging: AIGC is setting off a
New technology and industrial revolution
AIGC, which uses AI to automatically generate content (AIGeneratedContent). It is not a new thing. It can be traced back to 1957, when Lejaren Hiller and Leonard Isaacson completed the first computer-generated musical composition in human history. AI-generated models and AI-generated works continue to appear, but 2022 is truly the first year of AIGC's outbreak. The significance of AIGC is not just that technology generates content, but that AI has the ability to generate and create like humans. Benefiting from the unlimited creative potential and future application space, AIGC is setting off a new technological and industrial revolution, pushing artificial intelligence to usher in the next era.
(1) Large model is the cornerstone of AIGC outbreak
With the advent of deep learning in 2010, the development of artificial intelligence has advanced to the third climax, and the large model has brought this climax to a new stage. In 2017, Google released the landmark Transformer algorithm in the article "AttentionisAllYouNeed". Although it is still a continuation of deep learning, it made the deep learning model parameters exceed 100 million. Transformer replaced RNN and CNN and entered the era of large models. This is undoubtedly is a major milestone.
Transformer is a neural network model based on the self-attention mechanism. It was originally used to complete text translation tasks between different languages. The main body includes the Encoder and Decoder parts, which are responsible for encoding the source language text and converting the encoded information into the target language. text. Then based on the Encoder and Decoder, the development of the large model has roughly embarked on three roads: the first is to abandon the Decoder part, and only use the Encoder as the pre-training model of the encoder. The most famous representative is the Bert family; the second is to abandon the Decoder part. The Encoder part is based on the GPT family of the Decoder part; the third is the Google T5 large model route used by both the Encoder and the Decoder.
AI large model, also known as pre-training model or foundation model, is a model trained based on a large amount of data and has a large number of parameters, which can be adapted to a wide range of downstream tasks. These models, based on the ideas of transfer learning and recent advances in deep learning, as well as large-scale applied computer systems, exhibit surprising emergent capabilities and significantly improve the performance of various downstream tasks. In view of this potential, the large model has become a paradigm change in the development of AI technology, and many cross-domain AI systems or product services will be directly built on the large model. Specifically in the field of AIGC, AI large models can achieve multi-task, multi-language, and multi-mode, and will play a key role in the generation of various content. According to the basic types, pre-training models include natural language processing (NLP) pre-training models, computer vision (CV) pre-training models, and multimodal pre-training models. These three types of models have broad application prospects in journalism and other fields.
Why is it said that the large model is the cornerstone of the outbreak of AIGC? It is because the large model has triggered a qualitative change in AIGC's technical capabilities. Although various generative models have emerged in an endless stream in the past, the high threshold for use, high training costs, simple content generation and low quality are far from meeting the flexible, high-precision, and high-quality needs of real content consumption scenarios. The large model solves many of the above landing problems. For example, ChatGPT can provide high-quality text content generation services for people from different countries, different cultural backgrounds, different professional fields and age groups at the same time, which was unimaginable before. ChatGPT also demonstrates the magical ability brought by large models beyond text generation itself. ChatGPT, GPT-4, Bard, PaLM, LLaMA, etc. have brought the current prosperity of large models, and also brought the dawn of AGI.
In general, AIGC's outbreak in 2022 will benefit from large-scale model technology. The AIGC large model, which has the characteristics of versatility, basicity, multi-modality, multiple parameters, large amount of training data, and high-quality and stable generated content, has become a "factory" and "assembly line" for automated content production.
(2) Industrial ecology is the guarantee for the development of AIGC
The previous slow development of the AI industry is closely related to the lack of a more mature industrial system. Any mature industry has a relatively complete upstream and downstream industrial ecosystem. For example, in the automotive industry, there are only a few manufacturers of core components such as engines and gearboxes in the world, but there can be many consumer-oriented automobile manufacturers. In the previous AI industry, from basic model research and development to online sales of products and services, every company seems to have to cover the entire industry chain, and it is difficult to balance cost input and profit return.
The development of the industry is in a difficult situation. Previously, the lack of versatility of the AI model was the core problem, but now based on the large model, the AIGC industrial ecosystem has been initially formed, presenting a three-tier structure of upper, middle and lower.
**The first layer is the upstream base layer, which is the AIGC technical infrastructure layer built on the basis of the large model. **Due to the high cost and technical investment of large models, they have high barriers to entry. Taking the GPT-3 model launched in 2020 as an example, AlchemyAPI founder Elliot Turner speculated that the cost of training GPT-3 may be close to 12 million US dollars. Therefore, the main institutions currently entering the pre-training model are leading technology companies and scientific research institutions.
In the field of AIGC, American infrastructure companies (in the upstream ecological niche) include OpenAI, Stability.ai, etc. Because of the technical support of the basic layer, the downstream industry can develop like mushrooms after rain, forming the current AIGC business flow.
**The second layer is the middle layer, that is, vertical, scene-oriented, personalized models and application tools. **The pre-trained large model is the infrastructure. On this basis, it can quickly extract and generate scene-oriented, customized, and personalized small models to realize industrial pipeline deployment in different industries, vertical fields, and functional scenarios. Advantages of on-demand use, high efficiency and economy. Based on the large model, Model-as-a-Service (MaaS) becomes a reality, which realizes the transformation of AI from "manual workshop" to "factory mode". AI large models have stronger versatility and intelligence. MaaS provides safe, efficient, and low-cost model use and development support for downstream applications. It can be applied in industries on a large scale and empowers applications in various industries. Bring about the improvement of the production efficiency of the whole society. OpenAI CEO Sam Altman once clearly pointed out that the middle layer is the core position of future AI entrepreneurship.
For example, based on the open API interface of ChatGPT, many large models or application tools used in the financial and medical fields have been produced. JasperAI relied on GPT-3 to automatically generate creative marketing content, and turned from scratch into a unicorn in 18 months. Also, after StableDiffusion was open-sourced, there have been many secondary developments based on open-source models, training specific styles of vertical domain models have become popular, such as the famous Novel-AI generated by the two-dimensional painting style, and various styles of character generators, etc. .
**The third layer is the application layer, that is, content generation services such as text, pictures, audio and video for C-end users. **In the application layer, it focuses on meeting the needs of users, and seamlessly connects the AIGC model with the needs of users to achieve industrial landing. NotionAI based on the GPT-3 large model is such a product, which can meet users' professional text content generation needs. Take StableDiffusion open source as an example. It not only opens up programs, but also has trained models. Successor entrepreneurs can better use this open source tool to dig out more abundant resources with the computing power threshold of C-end consumer-grade graphics cards. The content ecology plays a vital role in the popularization of AIGC among a wider range of C-end users. Now there are more and more tools for C-end users, including web pages, locally installed programs, mobile applets, group chat robots, etc., and even content consumption services that use AIGC tools to customize and generate maps.
At present, from the infrastructure layer company that provides large-scale models to the application layer company that focuses on building AIGC products and application tools, AIGC has grown a prosperous ecology, technology innovation has triggered waves of application innovation, and technology empowers thousands of industries. As the integration of the digital economy and the real economy continues to deepen, and the digital scenarios of the Internet platform become more and more abundant, the overall human demand for the total amount and richness of digital content continues to increase. As a new type of content production method, AIGC has taken the lead in achieving major innovation and development in news media, e-commerce, film and television, entertainment and other industries with high digitalization and rich content demand, and its market potential is gradually emerging. At the same time, in the process of promoting digital-real integration and accelerating industrial upgrading, AIGC applications in various industries such as finance, medical care, and industry are also developing rapidly.
(3) Scenario application innovation, embodied intelligence, and ability equal rights are the future orientation of AIGC
Scenario application innovation is the future development path of AIGC. Any emerging technology can only be widely used if it is applied in a specific scenario and generates economic and social value. At the same time, in a wide range of applications, technology can continue to iteratively innovate and develop. This forms the flywheel effect of "scene application and technology iteration". In 2022, the Ministry of Science and Technology successively issued the "Notice on Supporting the Construction of New-Generation Artificial Intelligence Demonstration Application Scenarios" and "Guiding Opinions on Accelerating Scenario Innovation and High-level Application of Artificial Intelligence to Promote High-quality Economic Development". Ten demonstration application scenarios such as ports and smart mines. It has become an industry consensus to promote the implementation of artificial intelligence technology through application traction. At present, OpenAI's strategy is also to try to establish an application ecology and apply large models to various industries.
Scenario application innovation also means that AIGC will be more vertical and lightweight in the future. First, although the big model is generalist, it lacks industry depth. Its future development trend may be "verticalization" in six aspects, including industry depth, enterprise personalization, capability specialization, scale miniaturization, deployment distribution, and ownership privatization. Second, in the future, AI will be embedded in all areas of social production and life, especially mobile devices and embedded devices, that is, localized deployment is required. At present, large models have high requirements for hardware computing power and memory, while mobile devices or embedded devices often have limited computing power. Therefore, model lightweight will be an important direction for the future development of AIGC.
Embodied intelligence is an inevitable form of AI development. Embodied intelligence means that AI is not only digital or virtual, but also has a physical form in the physical environment, such as robots or other devices that can interact with the real world. True intelligence and learning require interaction with the physical world, as most biological intelligences have evolved in direct contact with their environment. Embodied intelligence can better learn perception and behavior through interaction with the environment. Similar to this point of view, some scholars believe that ChatGPT will not be able to achieve super artificial intelligence in the future because they lack the ability to interact with the real world. Therefore, embodied intelligence is considered to be the key to general artificial intelligence, and "embodied intelligence" robots are the ultimate form of artificial intelligence
In July 2023, the team led by AI scientist Li Feifei released the latest achievements in embodied intelligence. They connected large models to robots and transformed complex instructions into specific action plans. Humans can use natural language to give instructions to robots. More importantly, by combining LLM (Large Language Model) + VLM (Visual Language Model), the ability of the robot to interact with the environment is further improved, and tasks can be completed without additional data and training.
Ability equality is the inevitable result of the development of AIGC. At present, the development of AIGC has given users more creative power and freedom. For example, ordinary people can use AIGC to create novels, music works, 3D content, etc., all of which can be generated on demand based on input words. Not only that, but in the future everyone may have their own "Jarvis" - a personal intelligent assistant like Iron Man. In 2021, Microsoft first introduced the concept of Copilot (copilot) on GitHub. GitHub Copilot is an AI service that assists developers in writing code. In May 2023, with the blessing of the large model, Microsoft will usher in a comprehensive upgrade of Copilot, launching Dynamics365Copilot, Microsoft365Copilot and PowerPlatformCopilot, etc., and put forward the concept of "Copilot is a brand new way of working". Work is like this, and life also needs "Copilot". Li Zhifei, the founder of Go Ask, believes that the best job for large models is to be a "Copilot" for humans. The AIGC large model may become everyone's intelligent assistant, so that everyone can enjoy the AIGC technology dividend.
In addition to "supply-side reform":
What does the AIGC bring to journalism?
The overall recession of the global economy, the substitution effect of new technologies, the impact of short videos, and the reduction in traffic from social media have caused the news industry to face difficulties. In this context, the emergence of the AIGC may be a ray of hope for news production and journalism as a whole. So, what new possibilities will AIGC bring to journalism? Could it be a way out of a difficult situation?
(1) AI-assisted news production is not new
Before discussing the changes brought about by AIGC, looking back at the development history of journalism, we can see that AI's involvement in journalism, especially news production, is not without precedent. Over the past decade or so, the wave of journalism innovations triggered by artificial intelligence can be divided into three stages: the stage of automated reporting, the stage of enhanced reporting, and the stage of generating reports.
**The first stage, the stage of artificial intelligence automated reporting. **At this stage, it is mainly to use AI's natural language generation (NLG) ability to automatically report news. Media organizations such as the Associated Press, Reuters, Bloomberg, and Agence France-Presse all have representative practices. Automated reporting uses programs to automatically generate text content, which has advantages in reporting efficiency and accuracy, but due to lack of thinking and empathy, it is difficult to write reports comparable to human reporters, so it is only applicable to specific fields, such as finance, Sports and other news types that can be templated.
In terms of applications, for example, the automatic news generation system called "AI News Production Line" developed by Reuters can generate news such as stocks, sports and weather; the Washington Post uses an automated writing robot called Heliograf, which can Generating simple news reports in the fields of science, politics and sports; the automatic writing system launched by NHK TV station in Japan was outstanding in the report of the Tokyo earthquake in March 2011. Products such as DreamWriter launched by Tencent in 2015 and Kuaibi Xiaoxin by Xinhua News Agency are representative practices of automated reporting in China. During the two sessions of the country in 2018, the "media brain" launched by Xinhua News Agency sorted out the hot words of the two sessions of the country from 500 million web pages, and generated and released the world's first machine-produced video news about the two sessions, which took only 15 seconds.
**The second stage, the stage of AI-enhanced news reporting. **This phase focuses on using machine learning and natural language processing (NLP) techniques to analyze data and reveal relevant trends. For example, the Argentine newspaper La Nación has been using artificial intelligence to support its data team since 2019, and then cooperated with data analysts and developers to establish an AI laboratory to further strengthen AI applications.
The application of AI in public opinion analysis is also an example of AI-enhanced news reporting. In the public opinion analysis process, AI can assist in tasks such as sentiment analysis, topic detection, forecasting and trend analysis, helping organizations better understand public opinions and attitudes in order to cope with complex public opinion and market environments. For example, the application developed by the Associated Press and NewsWhip can help professionals track the dissemination of content, analyze how the content will drive the social participation of members and customers, and adjust the content strategy to better meet the needs of users. There are also media that use the data capabilities of AI to optimize content. For example, Forbes launched the AI content publishing platform Bertie in 2019, which can generate more attractive headlines and automatically match pictures for the content of the report to optimize the communication effect; Washington Post "Also continue to explore the practice of incorporating AI into the business, such as launching the ForYou recommendation system, and using AI models to detect subscription tendencies and user loss.
**The third stage is the stage where generative artificial intelligence (AIGeneratedContent) participates in news production with multi-modal generation capabilities. **ChatGPT, Google Bard, Microsoft NewBing and other products are based on a large-scale language model (LLM) that can generate narrative text. Compared with the automated reporting stage that is only applicable to financial reports, sports reports, etc., AIGC can conduct longer , higher-quality report writing, and can imitate specific work styles according to instructions. AIGC's multi-modal generation capability also brings many new possibilities for the visualization of news reports. At present, the journalism industry is still at this stage, and relevant practices still need to be deepened. However, it is foreseeable that AIGC will affect news gathering, production and presentation, and then change the entire pattern of the journalism industry.
(2) AIGC will realize the "supply-side reform" of journalism
Multi-modal content such as Wensheng text, Wensheng picture, Wensheng audio and video, Wensheng code, etc., all belong to AIGC, that is, the category of artificial intelligence generated content. Traditional content production models, such as UGC, PGC, etc., mainly differ in the professionalism and composition attributes of the authors, but in essence, people are the main body to produce content, while AIGC uses AI to produce different forms of content.
AIGC's influence on journalism is mainly concentrated in the stage of news production. With the improvement of AIGC technical capabilities such as ChatGPT and the deepening of its application, its impact on the journalism industry will also deepen. The current application practice shows that the impact of AIGC on journalism mainly includes the following aspects:
**Firstly, the collection and processing of news information to optimize the production process. **
With the help of plug-ins such as plugins, ChatGPT can quickly capture and collect massive amounts of data, and perform automatic processing, such as quickly browsing texts and generating summaries for further analysis by reporters. This ability provides a possibility to improve the efficiency of information acquisition. In the data retrieval stage, reporters and editors do not need to read a large amount of full-text materials, but can use ChatGPT’s data analysis and semantic analysis capabilities to generate abstracts and quickly obtain core information to improve work. s efficiency. ChatGPT's language generation capability can also be used to translate cross-language texts, making it easier for reporters and editors to obtain materials and information in different languages. At the same time, AIGC tools can assist journalists to identify and organize interview audio and video content, improve productivity and optimize the creative process. According to our research, "document retrieval" and "translation content" are currently the two most commonly used AIGC by media practitioners, accounting for 54.8% and 44% respectively.
Using AIGC to enhance the ability to collect and process information will play an increasingly critical role in news reporting. Roula Khalaf, editor-in-chief of the Financial Times, pointed out that the newsroom should establish an AI technology team to assist reporters in data mining, content analysis and translation. AI's ability to mine stories.
**Secondly, the generation of news content improves the reporting efficiency. **
ChatGPT has strong learning ability and text generation ability. After networking, it can quickly collect Internet data to generate news content. Through the setting of prompt words (), ChatGPT can also generate news reports of a specific style. In addition, ChatGPT can be applied to generate interview outlines, article frames and titles, etc. It can also translate news reports into multiple languages, break language boundaries, and spread news to diverse audiences.
Some media have incorporated AIGC into the production process of news content. For example, BuzzFeed uses ChatGPT for quiz content generation; before Valentine’s Day in 2023, The New York Times created a Valentine’s Day message generator using ChatGPT. Users only need to enter a few prompt instructions, and the program can automatically generate a love letter; Germany Publishing group AxelSpringer and UK publisher Reach have also recently published articles written by AI on local news sites.
NewsGPT.com, the world's first platform for news reports generated entirely by artificial intelligence, has also been launched. According to the statement, the website has no human reporters, and NewsGPT scans and analyzes news sources from around the world in real time, including social media, news websites, etc., and creates news reports and reports. Its founder claims that NewsGPT is "not influenced by advertisers or personal opinions" and provides "reliable" news 24/7.
**Finally, the multi-modal presentation of news reports has given birth to news types such as "interactive news". **
With the improvement of technical capabilities, GPT-4 already has the ability to generate multi-modality. In addition to Wenshengwen and Wenshengtu, it may generate more media forms in the future. At the same time, with the help of AIGC tools such as Midjourney, it has also achieved multi-modal content such as text generation, pictures, audio, code, and 3D content, which has created new possibilities for the generation of news content. The "media convergence" and "all-media reporter" that the journalism industry once pursued are now seeing the light of day due to the emergence and application of AIGC. The multimedia report "Avalanche" produced by "New York Times" in 2012, including pictures, videos, data, 3D content, etc., took 6 months and a team of 11 people to spend 250,000 US dollars to complete. Modal generation capability will greatly reduce the production cost and threshold of similar content.
At the same time, thanks to ChatGPT’s real-time interaction capabilities, it can be used to develop dialogue robots for journalism, integrate them into news reports, answer readers’ questions in real time and provide supplementary information based on data. This may expand a content form of "AIGC Interactive News", emphasizing the interaction with readers, and presenting a complete news picture through continuous questions and answers. AIGC can also enhance technical forms such as "virtual anchor" and optimize the effect of news presentation.
In terms of advertising and marketing content, AIGC has also demonstrated strong generation capabilities, such as using ChatGPT to write advertising copy or using products such as Midjourney to directly generate advertising content to improve creation efficiency. In addition, ChatGPT can also be used to analyze data sets to help advertisers understand consumer behavior patterns and market trends in order to optimize the effectiveness of advertising. AIGC is poised to bring about a revolution in the world of digital marketing.
(3) Objectively understand the role of AIGC in journalism
Overall, the AIGC technology represented by ChatGPT has the potential to improve efficiency and even realize changes in news information collection, content generation, and multi-modal presentation. In the future, with the further improvement of technical capabilities and the deepening of its application in the journalism industry, AIGC will replace some conventional content production links, liberating reporters and editors from tedious work that consumes time and energy, and focusing on more creative work. However, in this process, the problem of manpower reduction caused by "technological substitution" is inevitable, so the survival status of journalists in the new technology environment deserves attention.
With its powerful content generation capabilities, AIGC is expected to realize a "supply-side reform" of the journalism industry. But in terms of actual application, it is still too early to "reform". Currently, tools such as ChatGPT are mainly used to improve the efficiency of content production, which is an "upgraded version" of automated reporting. Because it still does not have empathy, thinking, common sense judgment, etc. Basic ability, AIGC cannot really be used for writing in-depth reports, but is used in specific fields such as sports and stocks, as well as "leftovers" such as the generation of test content. Deputy director Cao Feng commented that ChatGPT is still unable to replace writing needs in high-demand and high-limit scenarios. It can also be seen from industry practice that after the ChatGPT fire, although many media organizations have made relevant attempts, they have not No authoritative media has really applied ChatGPT to the production process of news reports. Including our survey results, only 38.1% of news media organizations are actively using AIGC tools like ChatGPT.
There are several reasons for this, including:
** Content is poorly readable. **Although ChatGPT can quickly generate content based on prompts, its readability is poor. The generated content is more like an expository text, which is not thoughtful and interesting to read. News is a report of recent facts. Although readers want to quickly understand the dynamics of the environment around them, they prefer to read more readable news reports than boring "explanatory texts". Part of the reason for the poor readability is that ChatGPT lacks analytical and investigative capabilities, and cannot perform the same original expressions as humans, so it cannot provide an in-depth view of events, and can only stack in-depth "pictures of soup". On April 18, 2023, the official account "Daily People" published an article titled "This is our first manuscript written entirely by ChatGPT". The reporter entered prompt words, and the content was all generated by ChatGPT. However, no matter from the actual text or the feedback from readers, this article cannot compare with the level of human authors. Key words such as "dull", "elementary school students' composition", "routine sense", "stiff" and "translation accent" appear frequently in the comment area . The human author who cooperated with ChatGPT also expressed his feelings about this cooperation: "It is definitely not pleasant, and it can even be described as painful."
**Information sources are confusing. **The technical principle of AIGC is a large model, and the data set composed of massive data constitutes the model training samples of AIGC. However, these data often include books, media reports, academic journals, as well as self-media articles, advertising and marketing copywriting, and social media content. For professional media, the news reports they release should be responsible not only for the readers but also for the reputation of the institution. AIGC with confusing information sources is obviously not an ideal choice. As Julia Beizer, chief digital officer of Bloomberg Media, commented, the position of the media is to provide readers with fact-based information, but AI is not enough to be an accurate source of information.
** Made up information indiscriminately. **The concept of "machine hallucination" is used to describe AIGC's ability to "talk nonsense seriously". The word "hallucination" comes from the mental illness "Confabulation" in psychology, which means that individuals will answer questions by fabricating content out of fear of disappointing the other party or avoiding appearing stupid. Due to the setting of the program, tools such as ChatGPT must give answers to users' questions. If the training data set does not contain this question or the data set is wrong, ChatGPT will fabricate a wrong answer. At the same time, it lacks basic common sense and judgment, so it cannot realize that the answer given is wrong. If it is applied to news reports, it needs to be matched with manual proofreading and verification, which in turn increases the workload of humans. In 2023, the American technology news website CNET.com once launched dozens of articles generated by AI. Although the website editor claimed that the articles had been "checked and edited" before publishing, readers soon discovered that there were a large number of these articles. Fundamental mistakes, and half of them have problems of plagiarism and plagiarism.
Therefore, we need to objectively understand the role of ChatGPT in journalism. It is still too early to say that AIGC will revolutionize or even replace journalism. As a content industry, the news industry's demand for excellent talents will never change, and in-depth content based on first-hand interviews will become more and more important. As Madhumita Murgia, artificial intelligence editor of the Financial Times, puts it, although generative AI tools can synthesize information and edit it, they cannot output original content or have analytical capabilities. Can replace someone with original ability".
Sword of Dachmoth:
Will the AIGC be the death knell for journalism?
For the journalism industry, AIGC will set off a supply-side reform in the content production link. However, given the current level of AIGC technology, "reform" is far from coming. The AIGC has been incorporated into journalistic production practices rather limitedly and has not really begun to be of value. Therefore, it is actually too early to discuss the AIGC's challenge to the journalism industry. However, technology has been iterating. From the perspective of technology development history, we cannot underestimate the transformative effect caused by any technology. When the more advanced AIGC is incorporated into the journalism industry and widely used in the future, what challenges will it bring to the journalism industry? This is something we need to think about.
(1) Destroying the field effect of news production and impacting news concepts such as "objectivity"
AIGC's involvement in the content production link of the journalism industry will inevitably bring about destructive effects while improving efficiency.
ChatGPT is applied in the news production process. After a news event occurs, the program captures, analyzes and summarizes the relevant information, and quickly produces a collage of content, which maximizes the efficiency. However, as far as the journalism industry is concerned, multiple forces originally in the news field will have an impact on the content of the news report. Therefore, the birth of a report is not just a reporter's personal inspiration, but the product of the game balance of multiple forces. The result of the institutionalized operation of the news media. During this process, journalists also accept the discipline of journalistic professionalism to ensure the balance and authenticity of the reports as much as possible. But when the generating subject becomes ChatGPT, this "field effect" of news production gradually disappears.
Correspondingly, as Professor Wu Xiaoning of South China University of Technology mentioned in the paper "The Impact and Challenge of the ChatGPT Information "Revolution" on the Journalism Industry", in this process, the importance of news facts in historical texts has increased. Since the principle of ChatGPT is to use existing content as a training data set, the longer the influence of a phenomenon or event, the more relevant content, and the easier it is to be captured and integrated into the news content produced by the machine. In the same way, if certain news figures and news events have higher popularity, they are more likely to be captured and re-presented by artificial intelligence, which may form an "information polarization" effect and form an artificial intelligence-made "Information Cocoon".
At the same time, the information capture process itself involves legal and ethical issues, such as whether AIGC captures network content and uses it as a training data set in compliance with legal requirements? Should the subjects of the captured content (especially content creators such as journalists) be compensated financially? In February 2023, the image provider Getty sued StabilityAI on the grounds of "copyright infringement". These issues, at least for now, are still in the fog stage.
In addition, the ChatGPT-style news generation model will impact the existing news concept. Journalism professionalism emphasizes the dimensions of authenticity, objectivity, and publicity. These concepts are a set of operational norms gradually formed in journalism practice to ensure that news reports do not deviate from the truth. In the traditional journalism industry where people are the main body of production, journalists are disciplined by professionalism and professionalism, and pursue these concepts in their personal production practices. However, ChatGPT has no subjective consciousness and cannot understand the meaning behind these news concepts, and these concepts cannot be converted into a "language" that ChatGPT can understand as a string (prompt words).
There is a view that ChatGPT gets rid of the subjectivity of the individual subject and seems to be able to report more objectively and fairly. As NewsGPT advertises, this website will present news objectively and truthfully. But the problem is that the algorithm itself still has values, and the algorithm will also extend the discrimination in the real world. This is an unavoidable problem that is more difficult to solve than people as the subject. Professor Hu Yong from the School of Journalism and Communication of Peking University pointed out that the "objectivity" of journalism is endorsed by the reputation and word-of-mouth of people and institutions, but the "objectivity" of algorithms excludes any institution. The logic behind it is that technology is neutral Yes, there is no human bias, so objectivity can be guaranteed. But the problem is that technology is never neutral and lacks human judgment, so it is not the savior of "objectivity".
It is worth noting that the impact of ChatGPT on news production is also reflected in the irregular use of ChatGPT by practitioners, which can easily lead to problems such as plagiarism and unclear sources. According to our research, most (81.9%) media organizations have not issued specifications and guidelines for the use of tools such as ChatGPT. This is a practical issue that needs attention.
The impact of ChatGPT on news production will also be reflected in the employment replacement issues brought about by new technologies. This phenomenon is happening intensively due to ChatGPT's higher content production efficiency, which can replace human reporters on certain types of reports. For example, after BuzzFeed announced that it would use ChatGPT to assist in the generation of quiz content, it immediately announced its layoff plan. At the same time, in the "Hollywood Strike" movement that will occur in May 2023, how to prevent AI from replacing the work of human screenwriters has also become the core appeal of those participating in the movement. While these two examples do not point directly to journalism, this phenomenon will soon occur as ChatGPT is more deeply used in news production.
(2) "Hijacking" traffic, AIGC changes the content distribution pattern
At present, the proportion of information generated by AIGC is still low, but with the widespread promotion of AI-generated content and the in-depth application of AIGC technology, the field of content distribution will face a major impact.
In the digital age, a large portion of the traffic of online news media comes from search engines, and generative artificial intelligence is gradually becoming the main source of information for search engines. Microsoft's Bing browser integrates ChatGPT and is upgraded to NewBing; Google also announced that it will give priority to displaying content generated by artificial intelligence (such as its Bard) in search results. According to Google's test in March 2023, Bard only provided basic answers and summaries, but did not include links to news sources.
For search engines, this is a natural "market behavior" because it can directly present sorted search results, greatly improving the efficiency of users' information retrieval and optimizing user experience. However, once a pattern develops in which search engines allocate more traffic to the results generated by generative AI, more in-depth, long-form news content will be ignored.
Not only does this impact traffic to news outlets, it can significantly dent news outlet revenues. As more and more users get the desired content directly from the search page instead of clicking into the homepage of the news media, the living space of the news media that relies on advertising revenue sharing will be compressed. The revenue model centered on advertising will face a huge impact, and at the same time, the subscription revenue of the media will also be directly damaged.
Social media has also been affected. In the first half of 2023, the collapse of digital media such as BuzzFeedNews and VICE has confirmed the importance of social media. Once such traffic sources are cut off, the media that rely on it will be hit hard. News media such as "New York Times" and "Wall Street Journal" also set up accounts on social media platforms such as Twitter and Facebook to distribute content. When AIGC content floods into social media, similar "news bot accounts" will also appear. Taking away users' attention, users tend to choose to obtain quick and easy-to-obtain news summaries, thereby affecting the exposure of news media content.
(3) The Birth of Audience 4.0: From "News Consumer" to "News Producer"
For the journalism industry, AIGC will not only change the content production method, but also reconstruct the production relationship.
The reason is that, as an underlying technical capability, AIGC has a relatively low threshold. As long as network problems and account problems are solved, not only journalists can use it, but ordinary users can also use it. For the former, due to its high level of specialization, considering factors such as readability, production time and cost, the degree of acceptance of AIGC technology may not be deep. As for the latter, that is, ordinary audiences, they are more willing to use related technologies because they do not have similar "professional baggage".
In this case, ordinary people can also generate news information by using AIGC's generating ability. For example, for a certain news event, let ChatGPT quickly generate a news report explaining the cause and effect, or let ChatGPT generate a summary of a series of recent news, so that you can quickly understand the news. In addition, content such as news comments can be directly generated.
In this process, audiences are no longer just consumers of news information, but creators and producers of news information, turning from passive to active, thereby realizing the transformation of identity subjects. Looking back at the history of technological development, the emergence of the Internet has achieved a round of transformation. In the era of Web 2.0, the application of personal blogs (Blog), social media and other media forms has enabled ordinary people to obtain the "right to publish", that is, they can express their various opinions on the Internet. This has reversed the monopoly of traditional media on publishing rights in the pre-Internet era. Due to the extremely high cost of establishing a media organization, a newspaper or a TV station, it has formed a high threshold for information release, and it is difficult for ordinary people to have the opportunity and sufficient capital to establish their own channels. With the help of the Internet and mobile devices, everyone They have all become "news reporters", recording and publishing anytime and anywhere.
If the Internet has changed the pattern of content distribution, then the AIGC technology represented by ChatGPT has realized the "civilianization" of content production. With the help of AI, ordinary people can cross the professional threshold and become content producers comparable to professionals. Generate customized news content according to your own needs. With the help of social media, the cost of distribution is also negligible.
The research field classifies the "audience". The audience as the main body of daily dialogue is "Audience 1.0", and the audience as media content readers and attention commodities is "Audience 2.0". In social media where "everyone is a journalist" In this era, audiences who can record and publish at any time become "Audience 3.0". Then, entering the era of AIGC, with the help of AI, we can obtain an audience comparable to professional production capacity, and directly enter the era of "Audience 4.0".
The implications for journalism are profound. After the audience has the ability to collect and produce content, they can consume content more independently, reduce their dependence on news media output, and further reduce the latter's influence and "gatekeeper" status. The boundaries of the journalism industry will become increasingly blurred. How to differentiate from ordinary creators, strengthen professional boundaries, and how practitioners can deal with the crisis of professional identity will be challenges that the journalism industry must face.
(4) Crisis of trust in journalism triggered by the prevalence of fake news
AIGC has democratized content production, but it may also lead to the proliferation of rumors and fake news.
As the subject of content production, journalists are restricted by their media organizations and production mechanisms on the one hand, and constrained by news professionalism on the other hand. In the process of news production, they will pay attention to following various principles to ensure that news reports can Balanced, objective and authentic. Authenticity is the most basic requirement for publicly released news reports, including authenticity of facts, authenticity of details and authenticity of sources.
However, after the production subject is generalized, these limitations will no longer exist, and AIGC has the potential to become a tool for generating fake news and rumors. In February 2023, a "press release" about "Hangzhou Municipal Government will cancel traffic restrictions" was circulated on the Internet, and it was later discovered that the owner of a community used ChatGPT to generate it, and was forwarded by other owners with screenshots, resulting in the spread of wrong information. Similar incidents include the "Notice of Hangzhou Municipal Government on Adjusting Property Market Policies" circulated on April 18, 2023. The news stated that Hangzhou will implement a new property market policy in May, which was later confirmed to be fake news generated by ChatGPT. These fake news may bring extremely high political and economic risks, and damage the interests of relevant subjects. For example, in May 2023, a fake news written by generative AI "Warning of Major Risks of HKUST Xunfei" attracted widespread attention , leading to a sharp drop in HKUST Xunfei’s stock price.
In these incidents, AIGC has become the right-hand man of rumormongers. Its generation ability reduces the cost of dissemination and production of false information. If it is not controlled, the unverified false information generated by it will seriously pollute the information ecosystem. cause serious social impact.
AIGC's ability to create websites could also be used to spread fake news. With ChatGPT, anyone with basic coding skills can create a fake news website. This will also pollute the information ecology and cause great risks. At the same time, due to the characteristics of AIGC, after false news flows into the content market, if it is not screened, it may continue to form the corpus for large-scale model training, leading to further spread and strengthening of rumors, resulting in more serious and continuous consequences. The dissemination of fake news will affect the audience's recognition and trust in the news, which may overwhelm the facts, create confusion, and even bring about a new round of crisis of trust in journalism.
AIGC Era
Six Possibilities for the Development of Journalism
The application of new technologies often brings about disruptive changes. As media scholar Joshua Merowitz said: The intervention of any kind of media will create a new environment. Although AIGC has not yet been used on a large scale in news reports, in the face of the menacing AIGC wave, the news industry cannot stay out of it, and is bound to be involved in it, and even be completely reshaped.
From the perspective of historical development, as an observer and recorder of social development trends, the journalism industry does not resist new technologies, but instead integrates its capabilities into its own development to achieve self-innovation. This report believes that with the improvement of AIGC's technical capabilities and the continuous deepening of its application, the journalism industry will have the following six possible directions:
(1) Large media-specific models will be developed and applied
At present, the application of AIGC in journalism is still shallow. The key reason is that its information sources are unknown and its content is uneven. There are articles from authoritative journals, articles from self-media and marketing accounts, and a lot of fake news and fake news. It is because the current large models mostly use general-purpose training databases, so the quality of the presented content varies. These are the difficulties that hinder the application of news reports that focus on rigorous details, accurate information, and clear sources of information.
On the other hand, news reports have certain expression norms and discourse habits. In this case, it may become a trend to develop a dedicated large-scale model for the news industry. Its training data sets are all from news media reports, and the source can be traced to ensure that the information is true and accurate, the source is clear, the bias is reduced, and the content presentation is more in line with the professional expression norms of journalism.
At present, the cost of large-scale model training is gradually decreasing, and large media organizations may have their own exclusive large-scale models. This trend may not be limited to the journalism industry. For industries with clear industry boundaries and requirements for information sources and content presentation (such as the legal industry), it will be a development to develop dedicated large models instead of using off-the-shelf general large models. direction. There have been many practical examples in this regard, such as the "CCTV Media Large Model" jointly released by Shanghai AI Lab and China Central Radio and Television on July 20, which combines massive audiovisual data from the media and advanced algorithms and technologies from the lab. Improve the quality and efficiency of audio-visual media production.
(2) Fact checking and content proofreading will play a key role
Fact checking and content proofreading play a pivotal role in the traditional news industry, and almost all traditional newsrooms have a dedicated proofreading department (copydesk). However, with the accelerated process of media digitization, the importance of verification and proofreading has gradually decreased. A very clear example is that when the media has undergone large-scale layoffs in recent years, verification and proofreading departments are often the hardest hit areas, which is enough to show the neglect of verification and proofreading functions in the digital media era.
However, with the application of AIGC, the role of fact checking and content proofreading will become more and more critical. Similar positions will continue to play the role of "gatekeeper" to proofread and verify the content and details generated by AIGC, so as to avoid AIGC's random fabrications and prevent uncontrollable phenomena such as "machine hallucinations". In the face of increasingly advanced technology, the media should also strengthen cooperation with academic institutions and technology companies to improve the ability to identify wrong content.
At the same time, since the operating principle of AIGC is to reassemble and collage the content in the training data set, for the journalism industry, the originality of reporting is the bottom line that must be defended. Therefore, the accusation of verification and proofreading also includes "duplicate checking" of AI-generated content, deleting or marking the source of non-standard referenced content, avoiding the risk of public opinion caused by "plagiarism", damaging the reputation of the institution, and preventing media Ethical anomie and legal and moral issues.
(3) AIGC usage ethics and norms in journalism will be established
As a professional field, journalism has its own professionalism, ethics and normative requirements. For AIGC, a new technology form, relevant usage ethics and norms should also be established to form a unified principle within the profession, which is easy for practitioners to follow. These ethical norms include not only basic principles, such as "content generated using ChatGPT must be marked to ensure readers' knowledge", "content generated using ChatGPT must be manually checked and proofread before release", but also some specific ones. Operational regulations, such as in a report co-created by humans and AI, the content created by AI must not exceed a certain proportion, etc., in order to minimize the chaos caused by the application of AIGC. The "Ten Basic Principles of Journalism" applicable to the AIGC era is about to come out.
At present, the media has begun to promote such practices. For example, the technology media "Connection" has formulated relevant regulations, clearly defining the purpose and workflow of using AI to ensure content quality. Norms are not constraints, and reasonable norms will help technologies better integrate and exert their value. The main body for establishing norms may be industry associations, and each news organization will also form its own relevant norms and requirements based on its actual operating conditions. In addition to the code of ethics, it is equally important to help practitioners better understand and use AIGC's instruction manuals and courses. How to use AIGC to assist one's own news reporting practice will become one of the key capabilities of future journalists.
(4) News stratification, authoritative professional news reports will be more important
In the era of AIGC, the importance of authoritative and professional news reports will become increasingly prominent, and reshaping professionalism will become an important mission and a way out for news media organizations. AIGC has greatly improved the efficiency of content generation. However, there is a difference between machine-generated text and human-written content. Although the former is fast and has a complete framework, it cannot replace "good" news reports, and the latter will always have an audience market. The "good" mentioned here includes excellent writing, high readability, and strong empathy... These factors together constitute the conditions for touching readers.
AIGC intervenes in news production, and can quickly generate a report with complete elements when a news event occurs, which will meet the basic information needs of the audience. However, for the in-depth excavation of events and the supplement of background information, human reporters still need to go deep into the scene and conduct first-hand interviews and investigations. Therefore, the types of news will further differentiate in the future. On the one hand, real-time event reports and information reports will be completed by AIGC. In this field, the space for human reporters will become increasingly narrow. On the other hand, authoritative professional news reports and In-depth reporting will become more important and get more attention.
Correspondingly, the connection between media organizations, journalists and readers will become increasingly critical. One of the problems with AI as the main body of production is that it cannot establish an emotional connection with readers. In most cases, readers can clearly realize that AI is AI, a system without emotion and consciousness, which will weaken readers’ trust in the content degrees, and that’s where the opportunity for human journalists lies. Strengthening the connection with readers and building the brand of the organization and the personal brand of journalists will become key issues.
(5) There will be a "localized news" shift in the journalism industry
Due to the training principle of the AI large model, general-purpose text constitutes the main body of the training data, and the amount of text based on local content is small. Even if it is included in the training data set, it is easily overwhelmed by other types of information, so AIGC is not good at generating localized content. good. At the same time, the audience's attention to localized reports has not weakened. Therefore, the journalism industry in the era of AIGC may have a trend of localization.
The neglect of local news has become increasingly evident since the advent of digital media. Due to the flatness and low threshold of the Internet, the potential audience of a website is theoretically Internet users all over the world. For online media, in order to increase the traffic and exposure of website content, they often adopt a global strategy in content production and presentation, expand the scope of attention as much as possible, and report important events happening around the world. This tendency has also affected traditional media in turn. More and more local newspapers gradually expand the proportion of national reports in news gathering and editing.
At the same time, the reporting of localized news was gradually neglected. This is also an important reason for the audience to have a "news avoidance" emotion. The audience's demand for localized news is not met. Many times, the audience only wants to know what is happening around them, and does not want to pay too much attention to distant news events. Many media outlets have noticed this trend and are returning their focus to localized reporting. This shift will continue in the AIGC era, with more and more news outlets focusing on localized news reporting.
(6) AIGC Application Deepening Promotes News Type Innovation
The journalism industry has been relatively positive about embracing new technologies. The journalism industry is good at applying various new media forms to news reports to achieve richer presentation effects. For example, with the help of big data and algorithm technology, data journalism has emerged, featuring the visual presentation of objective data; as another example, with the help of multimedia technology, the "New York Times" conducted a comprehensive report on the avalanche that occurred in Tunnel Creek in the Cascade Mountains in Washington State. Reporting, the digital special report "SnowFall" (SnowFall) was launched, including text, pictures, video, data content and other media forms, which is considered to "redefine news reporting".
Similarly, by absorbing the characteristics and advantages of AIGC technology, new types of news will also emerge. One of the more likely innovations is "intelligent interactive news", that is, the main body of the report focuses on the core of the news event, and readers can interact at any time through the dialog box attached to the report page to understand the background information of the news, the cause and effect of the event and Historical context, and even the progress of the latest events, etc., the interaction between the audience and news reports will be enhanced like never before. Of course, this is only one of the possibilities. With the continuous deepening of the application of AIGC in the news industry, more imaginative news types and formats may appear in the future.
Conclusion:
Will the AIGC replace journalism?
German scholar Staubel summed up three stages of technological evolution: first, "invention", second "innovation", and finally "institutionalization", that is, the formation of culture. In a nutshell, "invention" is creation from scratch, and "innovation" is the utilization and improvement based on invention. As far as the current situation is concerned, AIGC is still in the stage of invention, and is moving towards the stage of innovation integrating with various fields. From the perspective of the history of technological development, it takes a long process for any technology to be accepted, adopted by society and really play a role. We should neither underestimate the change that the AIGC may trigger, nor overestimate the speed with which it will be achieved.
AIGC is promoting innovation in news collection, production, and presentation, but it is still too early to "disrupt" and "change". In our survey, most practitioners (50.5%) also believe that for journalism, tools such as ChatGPT are more of an auxiliary role, and only 10.5% believe that these tools are quality improvement tools. The most fundamental impact of AIGC on the journalism industry is that it has triggered a change in the way news is produced, thereby realizing the reconstruction of production relations. Specifically, AIGC has improved the efficiency of news production and lowered the threshold for news production. Using AIGC technologies such as ChatGPT, audiences can generate customized news information and comments based on their own information needs. As a result, traditional audiences have completed their identity transformation, from passive information consumers to active news producers, which will change the pattern and existing cognition of the journalism industry. This is the trend that journalism should be most wary of and needs to deal with.
Of course, advanced technology may change the way of production, but it cannot change the place of responsibility. For journalism in particular, humans will always be the moral actors and ultimate gatekeepers behind AI, even if all articles are generated by the AIGC. From this perspective, the responsibility of human beings will be more important. It will also become more and more important to strengthen the responsibility of the main body, strengthen the verification, and form the application ethics and norms of AIGC.
The term "news" not only refers to the "news reports" we can read, but also refers to the journalism industry and the news traditions it carries, including values, operating norms, ethical principles, and so on. As an unconscious subject, AI has never been able to inherit and follow these traditions, which are the basis for the existence and continuation of journalism.
ChatGPT will not replace journalists, just some of their jobs. Experienced journalists have high sensitivity, insight and empathy for news events, and can extract news value and put it into writing with fluent words. These subjective characteristics are the abilities that ChatGPT cannot replace. With the waves washing away, excellent journalists and authoritative news organizations will become more and more important. The origin of instrumental rationality is bound to stick to value rationality. For the journalism industry, strengthening professionalism and authority, emphasizing investigative reporting and explanatory reporting will be a way out in the AIGC era.
Many people think that ChatGPT has already appeared, so let GPT write articles and even replace journalism. But this view obviously ignores the complexity of journalism and the significance of its existence. The real journalism industry is "the lookout at the bow", safeguarding the public interest and expressing the demands of the people. This is the responsibility of the journalism industry and the starting point for the struggle of generations of journalists. Technical tools cannot understand this passion, nor can we attempt to transfer responsibility and professionalism to ChatGPT line by line. AIGC can never replace journalism at this point.
View Original
The content is for reference only, not a solicitation or offer. No investment, tax, or legal advice provided. See Disclaimer for more risks disclosure.
Inflection point? Journalism in the Age of AIGC
A whirlwind of generative artificial intelligence (AIGC) is sweeping many fields.
At the end of 2022, OpenAI released ChatGPT, a natural language dialogue application, and iteratively launched GPT-4 in March this year, which quickly attracted the attention of various industries and the public. Large-scale models are booming all over the world, capital, technology, and talents continue to pour in, and technology companies continue to deploy their own large-scale models. It is estimated that by 2030, the AIGC market size is expected to exceed one trillion yuan.
A few years ago, the news that AlphaGo defeated the Go player Li Shishi triggered a wave of artificial intelligence. However, the wave of AIGC is more violent, because in addition to technological breakthroughs, its low threshold and practicality make the perception of the application side more significant. Therefore, the public does not just stay in the discussion stage, but can directly experience the power of generative artificial intelligence.
Every round of technological innovation will outline a new era. "In the era of AIGC, all industries are worth reshaping with AI." Affected areas include education, finance, e-commerce, film and television, design, etc. Among them, journalism is one of the most severely affected areas, and the response to AIGC is also the most positive.
Internationally, many media have already carried out related attempts. The news aggregation website BuzzFeed released quizzes, a test column answered by AI, and said that it will use AIGC to write test content to replace part of the manpower. On May 24, The Washington Post announced the establishment of a cross-departmental AI collaboration mechanism, including the strategic decision-making team AITaskforce and the executive team AIHub, to better adapt to AI innovation practices. The British "Financial Times" also appointed an AI line editor for the first time to pay close attention to the latest developments in this field. More than 100 domestic media organizations, such as The Paper, Cover News, and Upstream News, announced their access to AIGC products in February this year. Tencent Research Institute conducted a survey on the topic of "ChatGPT's Impact on Journalism" in June this year. The results showed that 80% of domestic journalists had used ChatGPT or similar products, and more than half (56%) of them were affected by it. Interviewers believe that ChatGPT (or similar tools) has brought practical help to their work.
While the ability to generate text is outstanding, ChatGPT also has the ability to generate multimodal content. For journalism, where content creation is at its core, this will bring about many intersections with significant implications. This technological breakthrough that is "comparable to the industrial revolution" (Microsoft CEO Satya Nadella) and a technological form that is no less than "the birth of the personal computer or the Internet" (Bill Gates) what will it lead to? What kind of change? Will it bring new opportunities for journalism?
Through this report, we attempt to discuss the impact and challenges that AIGC technology represented by ChatGPT has brought to the journalism industry, as well as the new possibilities it creates. AIGC is starting a "technological revolution", and journalism is one of them. Through this incision, we may be able to capture the true meaning of this revolution for human society.
The Great Reshuffle: A Triple Transformation of Journalism
Since the turn of the millennium, journalism has experienced bright moments full of hope, and has also been mired in pessimism.
The brand-new digital ecology endows traditional media with vigorous development vitality, and at the same time, it also gave birth to a group of digital media upstarts. However, in recent years, various factors such as the change of flow logic, the impact of new media forms such as short videos, and the shrinking of online advertising revenue are causing the news industry to fall into a new predicament.
(1) The "traffic era" is over, the news media has been reshuffled, and connecting readers has become extremely important
The logic of content distribution has evolved several times.
After the financial crisis in 2008, advertisers' advertising investment shifted from traditional media to online media. Search engines represented by Google and social media represented by Facebook have completely reshaped the content traffic pattern. Traffic originating from these two ports becomes the most important source of exposure for online media.
Of particular concern is Facebook, whose founder, Mark Zuckerberg, was a firm believer in the positive value of news content: enhancing the reputation of the platform and improving user retention and engagement. Therefore, Facebook once vigorously strengthened the recommendation proportion of news content, allowing relevant content to gain more exposure. The decade between 2006 and 2016 saw a honeymoon period between social media and journalism.
According to a 2015 survey by the Pew Research Center, as many as 64% of Internet users get news from social media. This is the "Journalism Traffic Era" created by social media, which has created a large number of digital media upstarts. The business models of BuzzFeed and VICE, the most well-known digital media in the 21st century, are based on the viral spread of social media. Huge traffic and user attention poured into these media, followed by a large amount of venture capital. At their peak, BuzzFeed and VICE were valued at $1.7 billion and $5.7 billion, respectively.
But for media outlets, the foundations of this model are fragile, and their survival depends entirely on the platform. Once the algorithm and rules of the platform are changed, the business model will be severely damaged, and it is completely out of control. The turning point of the story happened in 2016. During the US presidential election, Facebook was questioned by the outside world for using algorithms to manipulate the election results. The "Cambridge Analytica" incident directly sent Mark Zuckerberg to the hearing. Faced with criticism from all sides, Facebook announced a reduction in the proportion of news content. In 2020, Facebook will further strengthen relevant measures to significantly reduce the push of news content and political content.
This is not just a turn of the Facebook platform, but the overall trend of social media. The adjustment of the algorithm has resulted in less and less exposure of news content, which has severely hit the media that rely on social media traffic, resulting in a collective dilemma for the industry. In 2023, BuzzFeed founder Jonah Peretti announced the closure of its news business BuzzFeedNews, VICE announced the closure of the news brand VICEWorldNews, and its main website is also planning to file for bankruptcy. VoxMedia, Insider, ABCNews and other media have all laid off employees to varying degrees.
The common problem of these media is that they have not accumulated a user base by strengthening paywalls and subscription services. In the face of huge traffic and advertising revenue when the big wave comes, no one will think this is a problem. But when the tide goes out, you can find out who is swimming naked.
In contrast, when upstarts such as BuzzFeed were in the limelight, the old media "New York Times" was looked down upon by the industry because of its slow adaptation to the Internet, and was even regarded as a representative of the revolution. Spurred on by BuzzFeed, The New York Times was forced to transform itself into the larger context of digital journalism. But this pace is not in a hurry. It puts its own content behind the paywall, and attracts subscribers by deepening the content, but it becomes the capital that can survive the cycle.
In the era of frothy and fleeting traffic, it has never been more important to build a closer connection with readers based on core readers. More and more news outlets are recognizing this, and it has spawned three new trends:
**First, use new media forms to establish a direct connection with readers. ** In recent years, podcasts and RSS reading have begun to rise, and many media have launched their own RSS subscription services and podcast brands, aiming to establish a close relationship with readers more directly and effectively, strengthen media brands and enhance commercial value;
**Second, focus more on local content rather than global hot topics. **For example, the media group MvskokeMedia adjusted its editorial strategy to focus on local community reports, reflecting its focus on core readers;
**Third, strengthen the transparency and openness of reports. On the one hand, it helps readers understand the concept of reporting, and on the other hand, it also reversely understands what kind of news reports readers need more. ** For example, the Honolulu Civil Beat held "pop-up newsrooms" similar to pop-up events in its area to strengthen communication with readers.
(2) With the rise of "short video journalism", the audience's attention has shifted, and traditional news concepts have been impacted
In 2023, the data statistics agency "PressGazette" (PressGazette) released a ranking list of 25 media information companies established since the millennium, of which Facebook topped the list, followed by TikTok. The influence of social media is self-evident. The rise of short video platforms represented by TikTok has had a profound impact on the development trend of the journalism industry.
TikTok is quickly becoming one of the largest content platforms and traffic bases in the world. Not only are a large number of young audiences gathered on the platform, but audiences of different age groups are gradually turning their attention to short videos instead of graphic content or serious news reports that news media are good at. Simultaneously with the diversion of audience attention are advertising revenue and venture capital, which also flow to TikTok and Instagram platforms that young users pay more attention to.
Not only that, but a new news format is emerging: "TikTok journalism". When major news events such as the new crown epidemic and the conflict between Russia and Ukraine occurred, people discovered that the main source of information was no longer the news media, but TikTok. A large amount of real-time and first-hand video content is spreading rapidly on TikTok. TikTok has gradually transformed from an entertaining short video platform to a comprehensive content platform including audio and video content, and has become an important way for Internet users to obtain information. For many young audiences, it's not that they don't watch the news anymore, they just don't read the news in the news media. Similar phenomena are also evident in China.
When the main carrier of news changes from text to video, this is a challenge for most news media. Some traditional media are actively seeking changes and trying to integrate into the short video news ecology. According to statistics from the "2022 Digital News Report" released by the Reuters Institute of Journalism at Oxford University, about half (49%) of mainstream media organizations regularly publish content on TikTok. As a representative of established media, the "Washington Post" specially hired a third-party team to produce content for its own TikTok account; the "Los Angeles Times" formed a content team called "404" to conduct experimental content production based on the preferences of young audiences.
Some native news outlets have also sprung up from TikTok journalism. For example, NowThis, which started with short video news, already has 8.5 million fans; Spanish content company Ac2ality takes "telling news in one minute" as its core concept, and has accumulated 3.9 million fans on TikTok since its launch in 2019.
The rise of TikTok journalism means a shift in the focus of audience attention. On the one hand, the impact on the news industry is reflected in the loss of advertising revenue and the deterioration of the media's living environment. Although efforts have been made to integrate into the short video news ecology, the inherent incompatibility between traditional news reports and video media has made this transformation ineffective. restricted. On the other hand, the widespread impact of short video news has impacted traditional news concepts. Values such as "objectivity" and "authenticity" that are regarded as the standard in the news industry are no longer emphasized. Fast, sensational, and visual impact have become new production standards. Data such as likes and retweets have become new indicators of news quality. "Yellow news" has gained more and more traffic and audiences, and the living space of traditional news has been further squeezed.
From the perspective of the audience, people are gradually accustomed to obtaining news and information through channels such as short videos, which is also affected to a certain extent by the increasingly prominent phenomena of "news fatigue" and "news avoidance". News avoidance is driven by both cognitive and emotional factors: the cognitive aspect is manifested in the perception that certain topics or events are reported too much, and reading these news will lead to a sense of fatigue, and it is difficult to obtain information increments, resulting in "news overload"; The emotional aspect refers to people actively avoiding news that triggers negative emotions, such as reports on epidemics, violence, and natural disasters.
In 2017, 29 percent of respondents said they "often or sometimes avoid the news," according to a report by the Reuters Institute of Journalism and Oxford University, a figure that rose to 32 percent in 2019. After the outbreak of the new crown pneumonia epidemic in 2020, people's demand for news surged briefly, but the phenomenon of news avoidance rebounded quickly, with 59% of people saying that they "sometimes or always actively avoid news". The audience's information receiving habits and changes in mentality have become factors that the news industry has to consider, and have also become obstacles to the transformation of news media.
(3) Shutdown and layoffs have become the norm, and journalists actively seek changes
Three years of the new crown epidemic have had a huge impact on the global economy, and the journalism industry cannot stay out of it.
The closure of news organizations became the norm. Foreign media, including BuzzFeed, VICE and other digital media, have closed their news businesses, and print publications such as The Livonia Observer have substantially ceased publication. The domestic situation is also not optimistic. The "Media Blue Book: China's Media Industry Development Report (2022)" jointly released by the School of Journalism and Communication of Tsinghua University and other institutions shows that the impact of the new crown pneumonia on some areas of the media industry is still continuing, and the advertising revenue of traditional journalism continues to decline. Budget for periodicals, newspapers and other media. Domestic newspaper advertising and distribution revenue fell sharply, and the TV advertising market was weak and declining. From 2020 to 2023, dozens of newspapers, including "City Pictorial" and "Southeast Express", announced their suspension or suspension of publication.
The income of journalists has dropped significantly. Economic uncertainty has affected the jobs of about two-thirds of journalists, according to the Press Gazette. More than 80% of the respondents are full-time journalists, most of them (71%) have an annual income of less than $100,000, and the average freelance writer's fee is less than $300.
Layoffs set the tone for media organizations. According to incomplete statistics, since 2020, dozens of media have announced layoff plans. BuzzFeed cut off a certain percentage of employees due to its plan to use AI to generate quiz content. On April 20, 2023, the founder of BuzzFeed once again announced the closure of its news business, laying off about 180 employees, involving content, technology, administration and other departments. 15% of the total. According to Forbes statistics, since January 2023, more than 30 newspapers and media organizations have carried out layoffs of varying degrees. The latest occurred on June 7. unique challenges” and will lay off 74 newsroom employees.
The global economic downturn and technological shock are causing the news industry to face a double crisis, and the living conditions of journalists are worrying, which is also closely related to the application of new technologies by news organizations. The introduction of automatic reporting and automatic editing systems has liberated part of the manpower, but also caused some practitioners to become redundant. The iteration of media technology has not brought about progress in labor relations, which is especially evident in the content industry. In the first half of 2023, a strike action continues to be staged in Hollywood, affecting many on-air dramas. Behind the strike is the transformation of the production method and broadcasting form of drama series by the streaming media platform represented by Netflix, which squeezes the living space of screenwriters. The substitution effect of technological mechanisms also impacts journalism.
Faced with limited living space, many news organizations and journalists began to shift their focus to social media and short video platforms. For example, opening channels and publishing video content on TikTok and YouTube to attract younger audiences while increasing revenue through ad-sharing models. On the other hand, journalists share knowledge and insights through social platforms such as Twitter and LinkedIn, and build personal brands and influence.
Emerging: AIGC is setting off a
New technology and industrial revolution
AIGC, which uses AI to automatically generate content (AIGeneratedContent). It is not a new thing. It can be traced back to 1957, when Lejaren Hiller and Leonard Isaacson completed the first computer-generated musical composition in human history. AI-generated models and AI-generated works continue to appear, but 2022 is truly the first year of AIGC's outbreak. The significance of AIGC is not just that technology generates content, but that AI has the ability to generate and create like humans. Benefiting from the unlimited creative potential and future application space, AIGC is setting off a new technological and industrial revolution, pushing artificial intelligence to usher in the next era.
(1) Large model is the cornerstone of AIGC outbreak
With the advent of deep learning in 2010, the development of artificial intelligence has advanced to the third climax, and the large model has brought this climax to a new stage. In 2017, Google released the landmark Transformer algorithm in the article "AttentionisAllYouNeed". Although it is still a continuation of deep learning, it made the deep learning model parameters exceed 100 million. Transformer replaced RNN and CNN and entered the era of large models. This is undoubtedly is a major milestone.
Transformer is a neural network model based on the self-attention mechanism. It was originally used to complete text translation tasks between different languages. The main body includes the Encoder and Decoder parts, which are responsible for encoding the source language text and converting the encoded information into the target language. text. Then based on the Encoder and Decoder, the development of the large model has roughly embarked on three roads: the first is to abandon the Decoder part, and only use the Encoder as the pre-training model of the encoder. The most famous representative is the Bert family; the second is to abandon the Decoder part. The Encoder part is based on the GPT family of the Decoder part; the third is the Google T5 large model route used by both the Encoder and the Decoder.
AI large model, also known as pre-training model or foundation model, is a model trained based on a large amount of data and has a large number of parameters, which can be adapted to a wide range of downstream tasks. These models, based on the ideas of transfer learning and recent advances in deep learning, as well as large-scale applied computer systems, exhibit surprising emergent capabilities and significantly improve the performance of various downstream tasks. In view of this potential, the large model has become a paradigm change in the development of AI technology, and many cross-domain AI systems or product services will be directly built on the large model. Specifically in the field of AIGC, AI large models can achieve multi-task, multi-language, and multi-mode, and will play a key role in the generation of various content. According to the basic types, pre-training models include natural language processing (NLP) pre-training models, computer vision (CV) pre-training models, and multimodal pre-training models. These three types of models have broad application prospects in journalism and other fields.
Why is it said that the large model is the cornerstone of the outbreak of AIGC? It is because the large model has triggered a qualitative change in AIGC's technical capabilities. Although various generative models have emerged in an endless stream in the past, the high threshold for use, high training costs, simple content generation and low quality are far from meeting the flexible, high-precision, and high-quality needs of real content consumption scenarios. The large model solves many of the above landing problems. For example, ChatGPT can provide high-quality text content generation services for people from different countries, different cultural backgrounds, different professional fields and age groups at the same time, which was unimaginable before. ChatGPT also demonstrates the magical ability brought by large models beyond text generation itself. ChatGPT, GPT-4, Bard, PaLM, LLaMA, etc. have brought the current prosperity of large models, and also brought the dawn of AGI.
In general, AIGC's outbreak in 2022 will benefit from large-scale model technology. The AIGC large model, which has the characteristics of versatility, basicity, multi-modality, multiple parameters, large amount of training data, and high-quality and stable generated content, has become a "factory" and "assembly line" for automated content production.
(2) Industrial ecology is the guarantee for the development of AIGC
The previous slow development of the AI industry is closely related to the lack of a more mature industrial system. Any mature industry has a relatively complete upstream and downstream industrial ecosystem. For example, in the automotive industry, there are only a few manufacturers of core components such as engines and gearboxes in the world, but there can be many consumer-oriented automobile manufacturers. In the previous AI industry, from basic model research and development to online sales of products and services, every company seems to have to cover the entire industry chain, and it is difficult to balance cost input and profit return.
The development of the industry is in a difficult situation. Previously, the lack of versatility of the AI model was the core problem, but now based on the large model, the AIGC industrial ecosystem has been initially formed, presenting a three-tier structure of upper, middle and lower.
**The first layer is the upstream base layer, which is the AIGC technical infrastructure layer built on the basis of the large model. **Due to the high cost and technical investment of large models, they have high barriers to entry. Taking the GPT-3 model launched in 2020 as an example, AlchemyAPI founder Elliot Turner speculated that the cost of training GPT-3 may be close to 12 million US dollars. Therefore, the main institutions currently entering the pre-training model are leading technology companies and scientific research institutions.
In the field of AIGC, American infrastructure companies (in the upstream ecological niche) include OpenAI, Stability.ai, etc. Because of the technical support of the basic layer, the downstream industry can develop like mushrooms after rain, forming the current AIGC business flow.
**The second layer is the middle layer, that is, vertical, scene-oriented, personalized models and application tools. **The pre-trained large model is the infrastructure. On this basis, it can quickly extract and generate scene-oriented, customized, and personalized small models to realize industrial pipeline deployment in different industries, vertical fields, and functional scenarios. Advantages of on-demand use, high efficiency and economy. Based on the large model, Model-as-a-Service (MaaS) becomes a reality, which realizes the transformation of AI from "manual workshop" to "factory mode". AI large models have stronger versatility and intelligence. MaaS provides safe, efficient, and low-cost model use and development support for downstream applications. It can be applied in industries on a large scale and empowers applications in various industries. Bring about the improvement of the production efficiency of the whole society. OpenAI CEO Sam Altman once clearly pointed out that the middle layer is the core position of future AI entrepreneurship.
For example, based on the open API interface of ChatGPT, many large models or application tools used in the financial and medical fields have been produced. JasperAI relied on GPT-3 to automatically generate creative marketing content, and turned from scratch into a unicorn in 18 months. Also, after StableDiffusion was open-sourced, there have been many secondary developments based on open-source models, training specific styles of vertical domain models have become popular, such as the famous Novel-AI generated by the two-dimensional painting style, and various styles of character generators, etc. .
**The third layer is the application layer, that is, content generation services such as text, pictures, audio and video for C-end users. **In the application layer, it focuses on meeting the needs of users, and seamlessly connects the AIGC model with the needs of users to achieve industrial landing. NotionAI based on the GPT-3 large model is such a product, which can meet users' professional text content generation needs. Take StableDiffusion open source as an example. It not only opens up programs, but also has trained models. Successor entrepreneurs can better use this open source tool to dig out more abundant resources with the computing power threshold of C-end consumer-grade graphics cards. The content ecology plays a vital role in the popularization of AIGC among a wider range of C-end users. Now there are more and more tools for C-end users, including web pages, locally installed programs, mobile applets, group chat robots, etc., and even content consumption services that use AIGC tools to customize and generate maps.
At present, from the infrastructure layer company that provides large-scale models to the application layer company that focuses on building AIGC products and application tools, AIGC has grown a prosperous ecology, technology innovation has triggered waves of application innovation, and technology empowers thousands of industries. As the integration of the digital economy and the real economy continues to deepen, and the digital scenarios of the Internet platform become more and more abundant, the overall human demand for the total amount and richness of digital content continues to increase. As a new type of content production method, AIGC has taken the lead in achieving major innovation and development in news media, e-commerce, film and television, entertainment and other industries with high digitalization and rich content demand, and its market potential is gradually emerging. At the same time, in the process of promoting digital-real integration and accelerating industrial upgrading, AIGC applications in various industries such as finance, medical care, and industry are also developing rapidly.
(3) Scenario application innovation, embodied intelligence, and ability equal rights are the future orientation of AIGC
Scenario application innovation is the future development path of AIGC. Any emerging technology can only be widely used if it is applied in a specific scenario and generates economic and social value. At the same time, in a wide range of applications, technology can continue to iteratively innovate and develop. This forms the flywheel effect of "scene application and technology iteration". In 2022, the Ministry of Science and Technology successively issued the "Notice on Supporting the Construction of New-Generation Artificial Intelligence Demonstration Application Scenarios" and "Guiding Opinions on Accelerating Scenario Innovation and High-level Application of Artificial Intelligence to Promote High-quality Economic Development". Ten demonstration application scenarios such as ports and smart mines. It has become an industry consensus to promote the implementation of artificial intelligence technology through application traction. At present, OpenAI's strategy is also to try to establish an application ecology and apply large models to various industries.
Scenario application innovation also means that AIGC will be more vertical and lightweight in the future. First, although the big model is generalist, it lacks industry depth. Its future development trend may be "verticalization" in six aspects, including industry depth, enterprise personalization, capability specialization, scale miniaturization, deployment distribution, and ownership privatization. Second, in the future, AI will be embedded in all areas of social production and life, especially mobile devices and embedded devices, that is, localized deployment is required. At present, large models have high requirements for hardware computing power and memory, while mobile devices or embedded devices often have limited computing power. Therefore, model lightweight will be an important direction for the future development of AIGC.
Embodied intelligence is an inevitable form of AI development. Embodied intelligence means that AI is not only digital or virtual, but also has a physical form in the physical environment, such as robots or other devices that can interact with the real world. True intelligence and learning require interaction with the physical world, as most biological intelligences have evolved in direct contact with their environment. Embodied intelligence can better learn perception and behavior through interaction with the environment. Similar to this point of view, some scholars believe that ChatGPT will not be able to achieve super artificial intelligence in the future because they lack the ability to interact with the real world. Therefore, embodied intelligence is considered to be the key to general artificial intelligence, and "embodied intelligence" robots are the ultimate form of artificial intelligence
In July 2023, the team led by AI scientist Li Feifei released the latest achievements in embodied intelligence. They connected large models to robots and transformed complex instructions into specific action plans. Humans can use natural language to give instructions to robots. More importantly, by combining LLM (Large Language Model) + VLM (Visual Language Model), the ability of the robot to interact with the environment is further improved, and tasks can be completed without additional data and training.
Ability equality is the inevitable result of the development of AIGC. At present, the development of AIGC has given users more creative power and freedom. For example, ordinary people can use AIGC to create novels, music works, 3D content, etc., all of which can be generated on demand based on input words. Not only that, but in the future everyone may have their own "Jarvis" - a personal intelligent assistant like Iron Man. In 2021, Microsoft first introduced the concept of Copilot (copilot) on GitHub. GitHub Copilot is an AI service that assists developers in writing code. In May 2023, with the blessing of the large model, Microsoft will usher in a comprehensive upgrade of Copilot, launching Dynamics365Copilot, Microsoft365Copilot and PowerPlatformCopilot, etc., and put forward the concept of "Copilot is a brand new way of working". Work is like this, and life also needs "Copilot". Li Zhifei, the founder of Go Ask, believes that the best job for large models is to be a "Copilot" for humans. The AIGC large model may become everyone's intelligent assistant, so that everyone can enjoy the AIGC technology dividend.
In addition to "supply-side reform":
What does the AIGC bring to journalism?
The overall recession of the global economy, the substitution effect of new technologies, the impact of short videos, and the reduction in traffic from social media have caused the news industry to face difficulties. In this context, the emergence of the AIGC may be a ray of hope for news production and journalism as a whole. So, what new possibilities will AIGC bring to journalism? Could it be a way out of a difficult situation?
(1) AI-assisted news production is not new
Before discussing the changes brought about by AIGC, looking back at the development history of journalism, we can see that AI's involvement in journalism, especially news production, is not without precedent. Over the past decade or so, the wave of journalism innovations triggered by artificial intelligence can be divided into three stages: the stage of automated reporting, the stage of enhanced reporting, and the stage of generating reports.
**The first stage, the stage of artificial intelligence automated reporting. **At this stage, it is mainly to use AI's natural language generation (NLG) ability to automatically report news. Media organizations such as the Associated Press, Reuters, Bloomberg, and Agence France-Presse all have representative practices. Automated reporting uses programs to automatically generate text content, which has advantages in reporting efficiency and accuracy, but due to lack of thinking and empathy, it is difficult to write reports comparable to human reporters, so it is only applicable to specific fields, such as finance, Sports and other news types that can be templated.
In terms of applications, for example, the automatic news generation system called "AI News Production Line" developed by Reuters can generate news such as stocks, sports and weather; the Washington Post uses an automated writing robot called Heliograf, which can Generating simple news reports in the fields of science, politics and sports; the automatic writing system launched by NHK TV station in Japan was outstanding in the report of the Tokyo earthquake in March 2011. Products such as DreamWriter launched by Tencent in 2015 and Kuaibi Xiaoxin by Xinhua News Agency are representative practices of automated reporting in China. During the two sessions of the country in 2018, the "media brain" launched by Xinhua News Agency sorted out the hot words of the two sessions of the country from 500 million web pages, and generated and released the world's first machine-produced video news about the two sessions, which took only 15 seconds.
**The second stage, the stage of AI-enhanced news reporting. **This phase focuses on using machine learning and natural language processing (NLP) techniques to analyze data and reveal relevant trends. For example, the Argentine newspaper La Nación has been using artificial intelligence to support its data team since 2019, and then cooperated with data analysts and developers to establish an AI laboratory to further strengthen AI applications.
The application of AI in public opinion analysis is also an example of AI-enhanced news reporting. In the public opinion analysis process, AI can assist in tasks such as sentiment analysis, topic detection, forecasting and trend analysis, helping organizations better understand public opinions and attitudes in order to cope with complex public opinion and market environments. For example, the application developed by the Associated Press and NewsWhip can help professionals track the dissemination of content, analyze how the content will drive the social participation of members and customers, and adjust the content strategy to better meet the needs of users. There are also media that use the data capabilities of AI to optimize content. For example, Forbes launched the AI content publishing platform Bertie in 2019, which can generate more attractive headlines and automatically match pictures for the content of the report to optimize the communication effect; Washington Post "Also continue to explore the practice of incorporating AI into the business, such as launching the ForYou recommendation system, and using AI models to detect subscription tendencies and user loss.
**The third stage is the stage where generative artificial intelligence (AIGeneratedContent) participates in news production with multi-modal generation capabilities. **ChatGPT, Google Bard, Microsoft NewBing and other products are based on a large-scale language model (LLM) that can generate narrative text. Compared with the automated reporting stage that is only applicable to financial reports, sports reports, etc., AIGC can conduct longer , higher-quality report writing, and can imitate specific work styles according to instructions. AIGC's multi-modal generation capability also brings many new possibilities for the visualization of news reports. At present, the journalism industry is still at this stage, and relevant practices still need to be deepened. However, it is foreseeable that AIGC will affect news gathering, production and presentation, and then change the entire pattern of the journalism industry.
(2) AIGC will realize the "supply-side reform" of journalism
Multi-modal content such as Wensheng text, Wensheng picture, Wensheng audio and video, Wensheng code, etc., all belong to AIGC, that is, the category of artificial intelligence generated content. Traditional content production models, such as UGC, PGC, etc., mainly differ in the professionalism and composition attributes of the authors, but in essence, people are the main body to produce content, while AIGC uses AI to produce different forms of content.
AIGC's influence on journalism is mainly concentrated in the stage of news production. With the improvement of AIGC technical capabilities such as ChatGPT and the deepening of its application, its impact on the journalism industry will also deepen. The current application practice shows that the impact of AIGC on journalism mainly includes the following aspects:
**Firstly, the collection and processing of news information to optimize the production process. **
With the help of plug-ins such as plugins, ChatGPT can quickly capture and collect massive amounts of data, and perform automatic processing, such as quickly browsing texts and generating summaries for further analysis by reporters. This ability provides a possibility to improve the efficiency of information acquisition. In the data retrieval stage, reporters and editors do not need to read a large amount of full-text materials, but can use ChatGPT’s data analysis and semantic analysis capabilities to generate abstracts and quickly obtain core information to improve work. s efficiency. ChatGPT's language generation capability can also be used to translate cross-language texts, making it easier for reporters and editors to obtain materials and information in different languages. At the same time, AIGC tools can assist journalists to identify and organize interview audio and video content, improve productivity and optimize the creative process. According to our research, "document retrieval" and "translation content" are currently the two most commonly used AIGC by media practitioners, accounting for 54.8% and 44% respectively.
Using AIGC to enhance the ability to collect and process information will play an increasingly critical role in news reporting. Roula Khalaf, editor-in-chief of the Financial Times, pointed out that the newsroom should establish an AI technology team to assist reporters in data mining, content analysis and translation. AI's ability to mine stories.
**Secondly, the generation of news content improves the reporting efficiency. **
ChatGPT has strong learning ability and text generation ability. After networking, it can quickly collect Internet data to generate news content. Through the setting of prompt words (), ChatGPT can also generate news reports of a specific style. In addition, ChatGPT can be applied to generate interview outlines, article frames and titles, etc. It can also translate news reports into multiple languages, break language boundaries, and spread news to diverse audiences.
Some media have incorporated AIGC into the production process of news content. For example, BuzzFeed uses ChatGPT for quiz content generation; before Valentine’s Day in 2023, The New York Times created a Valentine’s Day message generator using ChatGPT. Users only need to enter a few prompt instructions, and the program can automatically generate a love letter; Germany Publishing group AxelSpringer and UK publisher Reach have also recently published articles written by AI on local news sites.
NewsGPT.com, the world's first platform for news reports generated entirely by artificial intelligence, has also been launched. According to the statement, the website has no human reporters, and NewsGPT scans and analyzes news sources from around the world in real time, including social media, news websites, etc., and creates news reports and reports. Its founder claims that NewsGPT is "not influenced by advertisers or personal opinions" and provides "reliable" news 24/7.
**Finally, the multi-modal presentation of news reports has given birth to news types such as "interactive news". **
With the improvement of technical capabilities, GPT-4 already has the ability to generate multi-modality. In addition to Wenshengwen and Wenshengtu, it may generate more media forms in the future. At the same time, with the help of AIGC tools such as Midjourney, it has also achieved multi-modal content such as text generation, pictures, audio, code, and 3D content, which has created new possibilities for the generation of news content. The "media convergence" and "all-media reporter" that the journalism industry once pursued are now seeing the light of day due to the emergence and application of AIGC. The multimedia report "Avalanche" produced by "New York Times" in 2012, including pictures, videos, data, 3D content, etc., took 6 months and a team of 11 people to spend 250,000 US dollars to complete. Modal generation capability will greatly reduce the production cost and threshold of similar content.
At the same time, thanks to ChatGPT’s real-time interaction capabilities, it can be used to develop dialogue robots for journalism, integrate them into news reports, answer readers’ questions in real time and provide supplementary information based on data. This may expand a content form of "AIGC Interactive News", emphasizing the interaction with readers, and presenting a complete news picture through continuous questions and answers. AIGC can also enhance technical forms such as "virtual anchor" and optimize the effect of news presentation.
In terms of advertising and marketing content, AIGC has also demonstrated strong generation capabilities, such as using ChatGPT to write advertising copy or using products such as Midjourney to directly generate advertising content to improve creation efficiency. In addition, ChatGPT can also be used to analyze data sets to help advertisers understand consumer behavior patterns and market trends in order to optimize the effectiveness of advertising. AIGC is poised to bring about a revolution in the world of digital marketing.
(3) Objectively understand the role of AIGC in journalism
Overall, the AIGC technology represented by ChatGPT has the potential to improve efficiency and even realize changes in news information collection, content generation, and multi-modal presentation. In the future, with the further improvement of technical capabilities and the deepening of its application in the journalism industry, AIGC will replace some conventional content production links, liberating reporters and editors from tedious work that consumes time and energy, and focusing on more creative work. However, in this process, the problem of manpower reduction caused by "technological substitution" is inevitable, so the survival status of journalists in the new technology environment deserves attention.
With its powerful content generation capabilities, AIGC is expected to realize a "supply-side reform" of the journalism industry. But in terms of actual application, it is still too early to "reform". Currently, tools such as ChatGPT are mainly used to improve the efficiency of content production, which is an "upgraded version" of automated reporting. Because it still does not have empathy, thinking, common sense judgment, etc. Basic ability, AIGC cannot really be used for writing in-depth reports, but is used in specific fields such as sports and stocks, as well as "leftovers" such as the generation of test content. Deputy director Cao Feng commented that ChatGPT is still unable to replace writing needs in high-demand and high-limit scenarios. It can also be seen from industry practice that after the ChatGPT fire, although many media organizations have made relevant attempts, they have not No authoritative media has really applied ChatGPT to the production process of news reports. Including our survey results, only 38.1% of news media organizations are actively using AIGC tools like ChatGPT.
There are several reasons for this, including:
** Content is poorly readable. **Although ChatGPT can quickly generate content based on prompts, its readability is poor. The generated content is more like an expository text, which is not thoughtful and interesting to read. News is a report of recent facts. Although readers want to quickly understand the dynamics of the environment around them, they prefer to read more readable news reports than boring "explanatory texts". Part of the reason for the poor readability is that ChatGPT lacks analytical and investigative capabilities, and cannot perform the same original expressions as humans, so it cannot provide an in-depth view of events, and can only stack in-depth "pictures of soup". On April 18, 2023, the official account "Daily People" published an article titled "This is our first manuscript written entirely by ChatGPT". The reporter entered prompt words, and the content was all generated by ChatGPT. However, no matter from the actual text or the feedback from readers, this article cannot compare with the level of human authors. Key words such as "dull", "elementary school students' composition", "routine sense", "stiff" and "translation accent" appear frequently in the comment area . The human author who cooperated with ChatGPT also expressed his feelings about this cooperation: "It is definitely not pleasant, and it can even be described as painful."
**Information sources are confusing. **The technical principle of AIGC is a large model, and the data set composed of massive data constitutes the model training samples of AIGC. However, these data often include books, media reports, academic journals, as well as self-media articles, advertising and marketing copywriting, and social media content. For professional media, the news reports they release should be responsible not only for the readers but also for the reputation of the institution. AIGC with confusing information sources is obviously not an ideal choice. As Julia Beizer, chief digital officer of Bloomberg Media, commented, the position of the media is to provide readers with fact-based information, but AI is not enough to be an accurate source of information.
** Made up information indiscriminately. **The concept of "machine hallucination" is used to describe AIGC's ability to "talk nonsense seriously". The word "hallucination" comes from the mental illness "Confabulation" in psychology, which means that individuals will answer questions by fabricating content out of fear of disappointing the other party or avoiding appearing stupid. Due to the setting of the program, tools such as ChatGPT must give answers to users' questions. If the training data set does not contain this question or the data set is wrong, ChatGPT will fabricate a wrong answer. At the same time, it lacks basic common sense and judgment, so it cannot realize that the answer given is wrong. If it is applied to news reports, it needs to be matched with manual proofreading and verification, which in turn increases the workload of humans. In 2023, the American technology news website CNET.com once launched dozens of articles generated by AI. Although the website editor claimed that the articles had been "checked and edited" before publishing, readers soon discovered that there were a large number of these articles. Fundamental mistakes, and half of them have problems of plagiarism and plagiarism.
Therefore, we need to objectively understand the role of ChatGPT in journalism. It is still too early to say that AIGC will revolutionize or even replace journalism. As a content industry, the news industry's demand for excellent talents will never change, and in-depth content based on first-hand interviews will become more and more important. As Madhumita Murgia, artificial intelligence editor of the Financial Times, puts it, although generative AI tools can synthesize information and edit it, they cannot output original content or have analytical capabilities. Can replace someone with original ability".
Sword of Dachmoth:
Will the AIGC be the death knell for journalism?
For the journalism industry, AIGC will set off a supply-side reform in the content production link. However, given the current level of AIGC technology, "reform" is far from coming. The AIGC has been incorporated into journalistic production practices rather limitedly and has not really begun to be of value. Therefore, it is actually too early to discuss the AIGC's challenge to the journalism industry. However, technology has been iterating. From the perspective of technology development history, we cannot underestimate the transformative effect caused by any technology. When the more advanced AIGC is incorporated into the journalism industry and widely used in the future, what challenges will it bring to the journalism industry? This is something we need to think about.
(1) Destroying the field effect of news production and impacting news concepts such as "objectivity"
AIGC's involvement in the content production link of the journalism industry will inevitably bring about destructive effects while improving efficiency.
ChatGPT is applied in the news production process. After a news event occurs, the program captures, analyzes and summarizes the relevant information, and quickly produces a collage of content, which maximizes the efficiency. However, as far as the journalism industry is concerned, multiple forces originally in the news field will have an impact on the content of the news report. Therefore, the birth of a report is not just a reporter's personal inspiration, but the product of the game balance of multiple forces. The result of the institutionalized operation of the news media. During this process, journalists also accept the discipline of journalistic professionalism to ensure the balance and authenticity of the reports as much as possible. But when the generating subject becomes ChatGPT, this "field effect" of news production gradually disappears.
Correspondingly, as Professor Wu Xiaoning of South China University of Technology mentioned in the paper "The Impact and Challenge of the ChatGPT Information "Revolution" on the Journalism Industry", in this process, the importance of news facts in historical texts has increased. Since the principle of ChatGPT is to use existing content as a training data set, the longer the influence of a phenomenon or event, the more relevant content, and the easier it is to be captured and integrated into the news content produced by the machine. In the same way, if certain news figures and news events have higher popularity, they are more likely to be captured and re-presented by artificial intelligence, which may form an "information polarization" effect and form an artificial intelligence-made "Information Cocoon".
At the same time, the information capture process itself involves legal and ethical issues, such as whether AIGC captures network content and uses it as a training data set in compliance with legal requirements? Should the subjects of the captured content (especially content creators such as journalists) be compensated financially? In February 2023, the image provider Getty sued StabilityAI on the grounds of "copyright infringement". These issues, at least for now, are still in the fog stage.
In addition, the ChatGPT-style news generation model will impact the existing news concept. Journalism professionalism emphasizes the dimensions of authenticity, objectivity, and publicity. These concepts are a set of operational norms gradually formed in journalism practice to ensure that news reports do not deviate from the truth. In the traditional journalism industry where people are the main body of production, journalists are disciplined by professionalism and professionalism, and pursue these concepts in their personal production practices. However, ChatGPT has no subjective consciousness and cannot understand the meaning behind these news concepts, and these concepts cannot be converted into a "language" that ChatGPT can understand as a string (prompt words).
There is a view that ChatGPT gets rid of the subjectivity of the individual subject and seems to be able to report more objectively and fairly. As NewsGPT advertises, this website will present news objectively and truthfully. But the problem is that the algorithm itself still has values, and the algorithm will also extend the discrimination in the real world. This is an unavoidable problem that is more difficult to solve than people as the subject. Professor Hu Yong from the School of Journalism and Communication of Peking University pointed out that the "objectivity" of journalism is endorsed by the reputation and word-of-mouth of people and institutions, but the "objectivity" of algorithms excludes any institution. The logic behind it is that technology is neutral Yes, there is no human bias, so objectivity can be guaranteed. But the problem is that technology is never neutral and lacks human judgment, so it is not the savior of "objectivity".
It is worth noting that the impact of ChatGPT on news production is also reflected in the irregular use of ChatGPT by practitioners, which can easily lead to problems such as plagiarism and unclear sources. According to our research, most (81.9%) media organizations have not issued specifications and guidelines for the use of tools such as ChatGPT. This is a practical issue that needs attention.
The impact of ChatGPT on news production will also be reflected in the employment replacement issues brought about by new technologies. This phenomenon is happening intensively due to ChatGPT's higher content production efficiency, which can replace human reporters on certain types of reports. For example, after BuzzFeed announced that it would use ChatGPT to assist in the generation of quiz content, it immediately announced its layoff plan. At the same time, in the "Hollywood Strike" movement that will occur in May 2023, how to prevent AI from replacing the work of human screenwriters has also become the core appeal of those participating in the movement. While these two examples do not point directly to journalism, this phenomenon will soon occur as ChatGPT is more deeply used in news production.
(2) "Hijacking" traffic, AIGC changes the content distribution pattern
At present, the proportion of information generated by AIGC is still low, but with the widespread promotion of AI-generated content and the in-depth application of AIGC technology, the field of content distribution will face a major impact.
In the digital age, a large portion of the traffic of online news media comes from search engines, and generative artificial intelligence is gradually becoming the main source of information for search engines. Microsoft's Bing browser integrates ChatGPT and is upgraded to NewBing; Google also announced that it will give priority to displaying content generated by artificial intelligence (such as its Bard) in search results. According to Google's test in March 2023, Bard only provided basic answers and summaries, but did not include links to news sources.
For search engines, this is a natural "market behavior" because it can directly present sorted search results, greatly improving the efficiency of users' information retrieval and optimizing user experience. However, once a pattern develops in which search engines allocate more traffic to the results generated by generative AI, more in-depth, long-form news content will be ignored.
Not only does this impact traffic to news outlets, it can significantly dent news outlet revenues. As more and more users get the desired content directly from the search page instead of clicking into the homepage of the news media, the living space of the news media that relies on advertising revenue sharing will be compressed. The revenue model centered on advertising will face a huge impact, and at the same time, the subscription revenue of the media will also be directly damaged.
Social media has also been affected. In the first half of 2023, the collapse of digital media such as BuzzFeedNews and VICE has confirmed the importance of social media. Once such traffic sources are cut off, the media that rely on it will be hit hard. News media such as "New York Times" and "Wall Street Journal" also set up accounts on social media platforms such as Twitter and Facebook to distribute content. When AIGC content floods into social media, similar "news bot accounts" will also appear. Taking away users' attention, users tend to choose to obtain quick and easy-to-obtain news summaries, thereby affecting the exposure of news media content.
(3) The Birth of Audience 4.0: From "News Consumer" to "News Producer"
For the journalism industry, AIGC will not only change the content production method, but also reconstruct the production relationship.
The reason is that, as an underlying technical capability, AIGC has a relatively low threshold. As long as network problems and account problems are solved, not only journalists can use it, but ordinary users can also use it. For the former, due to its high level of specialization, considering factors such as readability, production time and cost, the degree of acceptance of AIGC technology may not be deep. As for the latter, that is, ordinary audiences, they are more willing to use related technologies because they do not have similar "professional baggage".
In this case, ordinary people can also generate news information by using AIGC's generating ability. For example, for a certain news event, let ChatGPT quickly generate a news report explaining the cause and effect, or let ChatGPT generate a summary of a series of recent news, so that you can quickly understand the news. In addition, content such as news comments can be directly generated.
In this process, audiences are no longer just consumers of news information, but creators and producers of news information, turning from passive to active, thereby realizing the transformation of identity subjects. Looking back at the history of technological development, the emergence of the Internet has achieved a round of transformation. In the era of Web 2.0, the application of personal blogs (Blog), social media and other media forms has enabled ordinary people to obtain the "right to publish", that is, they can express their various opinions on the Internet. This has reversed the monopoly of traditional media on publishing rights in the pre-Internet era. Due to the extremely high cost of establishing a media organization, a newspaper or a TV station, it has formed a high threshold for information release, and it is difficult for ordinary people to have the opportunity and sufficient capital to establish their own channels. With the help of the Internet and mobile devices, everyone They have all become "news reporters", recording and publishing anytime and anywhere.
If the Internet has changed the pattern of content distribution, then the AIGC technology represented by ChatGPT has realized the "civilianization" of content production. With the help of AI, ordinary people can cross the professional threshold and become content producers comparable to professionals. Generate customized news content according to your own needs. With the help of social media, the cost of distribution is also negligible.
The research field classifies the "audience". The audience as the main body of daily dialogue is "Audience 1.0", and the audience as media content readers and attention commodities is "Audience 2.0". In social media where "everyone is a journalist" In this era, audiences who can record and publish at any time become "Audience 3.0". Then, entering the era of AIGC, with the help of AI, we can obtain an audience comparable to professional production capacity, and directly enter the era of "Audience 4.0".
The implications for journalism are profound. After the audience has the ability to collect and produce content, they can consume content more independently, reduce their dependence on news media output, and further reduce the latter's influence and "gatekeeper" status. The boundaries of the journalism industry will become increasingly blurred. How to differentiate from ordinary creators, strengthen professional boundaries, and how practitioners can deal with the crisis of professional identity will be challenges that the journalism industry must face.
(4) Crisis of trust in journalism triggered by the prevalence of fake news
AIGC has democratized content production, but it may also lead to the proliferation of rumors and fake news.
As the subject of content production, journalists are restricted by their media organizations and production mechanisms on the one hand, and constrained by news professionalism on the other hand. In the process of news production, they will pay attention to following various principles to ensure that news reports can Balanced, objective and authentic. Authenticity is the most basic requirement for publicly released news reports, including authenticity of facts, authenticity of details and authenticity of sources.
However, after the production subject is generalized, these limitations will no longer exist, and AIGC has the potential to become a tool for generating fake news and rumors. In February 2023, a "press release" about "Hangzhou Municipal Government will cancel traffic restrictions" was circulated on the Internet, and it was later discovered that the owner of a community used ChatGPT to generate it, and was forwarded by other owners with screenshots, resulting in the spread of wrong information. Similar incidents include the "Notice of Hangzhou Municipal Government on Adjusting Property Market Policies" circulated on April 18, 2023. The news stated that Hangzhou will implement a new property market policy in May, which was later confirmed to be fake news generated by ChatGPT. These fake news may bring extremely high political and economic risks, and damage the interests of relevant subjects. For example, in May 2023, a fake news written by generative AI "Warning of Major Risks of HKUST Xunfei" attracted widespread attention , leading to a sharp drop in HKUST Xunfei’s stock price.
In these incidents, AIGC has become the right-hand man of rumormongers. Its generation ability reduces the cost of dissemination and production of false information. If it is not controlled, the unverified false information generated by it will seriously pollute the information ecosystem. cause serious social impact.
AIGC's ability to create websites could also be used to spread fake news. With ChatGPT, anyone with basic coding skills can create a fake news website. This will also pollute the information ecology and cause great risks. At the same time, due to the characteristics of AIGC, after false news flows into the content market, if it is not screened, it may continue to form the corpus for large-scale model training, leading to further spread and strengthening of rumors, resulting in more serious and continuous consequences. The dissemination of fake news will affect the audience's recognition and trust in the news, which may overwhelm the facts, create confusion, and even bring about a new round of crisis of trust in journalism.
AIGC Era
Six Possibilities for the Development of Journalism
The application of new technologies often brings about disruptive changes. As media scholar Joshua Merowitz said: The intervention of any kind of media will create a new environment. Although AIGC has not yet been used on a large scale in news reports, in the face of the menacing AIGC wave, the news industry cannot stay out of it, and is bound to be involved in it, and even be completely reshaped.
From the perspective of historical development, as an observer and recorder of social development trends, the journalism industry does not resist new technologies, but instead integrates its capabilities into its own development to achieve self-innovation. This report believes that with the improvement of AIGC's technical capabilities and the continuous deepening of its application, the journalism industry will have the following six possible directions:
(1) Large media-specific models will be developed and applied
At present, the application of AIGC in journalism is still shallow. The key reason is that its information sources are unknown and its content is uneven. There are articles from authoritative journals, articles from self-media and marketing accounts, and a lot of fake news and fake news. It is because the current large models mostly use general-purpose training databases, so the quality of the presented content varies. These are the difficulties that hinder the application of news reports that focus on rigorous details, accurate information, and clear sources of information.
On the other hand, news reports have certain expression norms and discourse habits. In this case, it may become a trend to develop a dedicated large-scale model for the news industry. Its training data sets are all from news media reports, and the source can be traced to ensure that the information is true and accurate, the source is clear, the bias is reduced, and the content presentation is more in line with the professional expression norms of journalism.
At present, the cost of large-scale model training is gradually decreasing, and large media organizations may have their own exclusive large-scale models. This trend may not be limited to the journalism industry. For industries with clear industry boundaries and requirements for information sources and content presentation (such as the legal industry), it will be a development to develop dedicated large models instead of using off-the-shelf general large models. direction. There have been many practical examples in this regard, such as the "CCTV Media Large Model" jointly released by Shanghai AI Lab and China Central Radio and Television on July 20, which combines massive audiovisual data from the media and advanced algorithms and technologies from the lab. Improve the quality and efficiency of audio-visual media production.
(2) Fact checking and content proofreading will play a key role
Fact checking and content proofreading play a pivotal role in the traditional news industry, and almost all traditional newsrooms have a dedicated proofreading department (copydesk). However, with the accelerated process of media digitization, the importance of verification and proofreading has gradually decreased. A very clear example is that when the media has undergone large-scale layoffs in recent years, verification and proofreading departments are often the hardest hit areas, which is enough to show the neglect of verification and proofreading functions in the digital media era.
However, with the application of AIGC, the role of fact checking and content proofreading will become more and more critical. Similar positions will continue to play the role of "gatekeeper" to proofread and verify the content and details generated by AIGC, so as to avoid AIGC's random fabrications and prevent uncontrollable phenomena such as "machine hallucinations". In the face of increasingly advanced technology, the media should also strengthen cooperation with academic institutions and technology companies to improve the ability to identify wrong content.
At the same time, since the operating principle of AIGC is to reassemble and collage the content in the training data set, for the journalism industry, the originality of reporting is the bottom line that must be defended. Therefore, the accusation of verification and proofreading also includes "duplicate checking" of AI-generated content, deleting or marking the source of non-standard referenced content, avoiding the risk of public opinion caused by "plagiarism", damaging the reputation of the institution, and preventing media Ethical anomie and legal and moral issues.
(3) AIGC usage ethics and norms in journalism will be established
As a professional field, journalism has its own professionalism, ethics and normative requirements. For AIGC, a new technology form, relevant usage ethics and norms should also be established to form a unified principle within the profession, which is easy for practitioners to follow. These ethical norms include not only basic principles, such as "content generated using ChatGPT must be marked to ensure readers' knowledge", "content generated using ChatGPT must be manually checked and proofread before release", but also some specific ones. Operational regulations, such as in a report co-created by humans and AI, the content created by AI must not exceed a certain proportion, etc., in order to minimize the chaos caused by the application of AIGC. The "Ten Basic Principles of Journalism" applicable to the AIGC era is about to come out.
At present, the media has begun to promote such practices. For example, the technology media "Connection" has formulated relevant regulations, clearly defining the purpose and workflow of using AI to ensure content quality. Norms are not constraints, and reasonable norms will help technologies better integrate and exert their value. The main body for establishing norms may be industry associations, and each news organization will also form its own relevant norms and requirements based on its actual operating conditions. In addition to the code of ethics, it is equally important to help practitioners better understand and use AIGC's instruction manuals and courses. How to use AIGC to assist one's own news reporting practice will become one of the key capabilities of future journalists.
(4) News stratification, authoritative professional news reports will be more important
In the era of AIGC, the importance of authoritative and professional news reports will become increasingly prominent, and reshaping professionalism will become an important mission and a way out for news media organizations. AIGC has greatly improved the efficiency of content generation. However, there is a difference between machine-generated text and human-written content. Although the former is fast and has a complete framework, it cannot replace "good" news reports, and the latter will always have an audience market. The "good" mentioned here includes excellent writing, high readability, and strong empathy... These factors together constitute the conditions for touching readers.
AIGC intervenes in news production, and can quickly generate a report with complete elements when a news event occurs, which will meet the basic information needs of the audience. However, for the in-depth excavation of events and the supplement of background information, human reporters still need to go deep into the scene and conduct first-hand interviews and investigations. Therefore, the types of news will further differentiate in the future. On the one hand, real-time event reports and information reports will be completed by AIGC. In this field, the space for human reporters will become increasingly narrow. On the other hand, authoritative professional news reports and In-depth reporting will become more important and get more attention.
Correspondingly, the connection between media organizations, journalists and readers will become increasingly critical. One of the problems with AI as the main body of production is that it cannot establish an emotional connection with readers. In most cases, readers can clearly realize that AI is AI, a system without emotion and consciousness, which will weaken readers’ trust in the content degrees, and that’s where the opportunity for human journalists lies. Strengthening the connection with readers and building the brand of the organization and the personal brand of journalists will become key issues.
(5) There will be a "localized news" shift in the journalism industry
Due to the training principle of the AI large model, general-purpose text constitutes the main body of the training data, and the amount of text based on local content is small. Even if it is included in the training data set, it is easily overwhelmed by other types of information, so AIGC is not good at generating localized content. good. At the same time, the audience's attention to localized reports has not weakened. Therefore, the journalism industry in the era of AIGC may have a trend of localization.
The neglect of local news has become increasingly evident since the advent of digital media. Due to the flatness and low threshold of the Internet, the potential audience of a website is theoretically Internet users all over the world. For online media, in order to increase the traffic and exposure of website content, they often adopt a global strategy in content production and presentation, expand the scope of attention as much as possible, and report important events happening around the world. This tendency has also affected traditional media in turn. More and more local newspapers gradually expand the proportion of national reports in news gathering and editing.
At the same time, the reporting of localized news was gradually neglected. This is also an important reason for the audience to have a "news avoidance" emotion. The audience's demand for localized news is not met. Many times, the audience only wants to know what is happening around them, and does not want to pay too much attention to distant news events. Many media outlets have noticed this trend and are returning their focus to localized reporting. This shift will continue in the AIGC era, with more and more news outlets focusing on localized news reporting.
(6) AIGC Application Deepening Promotes News Type Innovation
The journalism industry has been relatively positive about embracing new technologies. The journalism industry is good at applying various new media forms to news reports to achieve richer presentation effects. For example, with the help of big data and algorithm technology, data journalism has emerged, featuring the visual presentation of objective data; as another example, with the help of multimedia technology, the "New York Times" conducted a comprehensive report on the avalanche that occurred in Tunnel Creek in the Cascade Mountains in Washington State. Reporting, the digital special report "SnowFall" (SnowFall) was launched, including text, pictures, video, data content and other media forms, which is considered to "redefine news reporting".
Similarly, by absorbing the characteristics and advantages of AIGC technology, new types of news will also emerge. One of the more likely innovations is "intelligent interactive news", that is, the main body of the report focuses on the core of the news event, and readers can interact at any time through the dialog box attached to the report page to understand the background information of the news, the cause and effect of the event and Historical context, and even the progress of the latest events, etc., the interaction between the audience and news reports will be enhanced like never before. Of course, this is only one of the possibilities. With the continuous deepening of the application of AIGC in the news industry, more imaginative news types and formats may appear in the future.
Conclusion:
Will the AIGC replace journalism?
German scholar Staubel summed up three stages of technological evolution: first, "invention", second "innovation", and finally "institutionalization", that is, the formation of culture. In a nutshell, "invention" is creation from scratch, and "innovation" is the utilization and improvement based on invention. As far as the current situation is concerned, AIGC is still in the stage of invention, and is moving towards the stage of innovation integrating with various fields. From the perspective of the history of technological development, it takes a long process for any technology to be accepted, adopted by society and really play a role. We should neither underestimate the change that the AIGC may trigger, nor overestimate the speed with which it will be achieved.
AIGC is promoting innovation in news collection, production, and presentation, but it is still too early to "disrupt" and "change". In our survey, most practitioners (50.5%) also believe that for journalism, tools such as ChatGPT are more of an auxiliary role, and only 10.5% believe that these tools are quality improvement tools. The most fundamental impact of AIGC on the journalism industry is that it has triggered a change in the way news is produced, thereby realizing the reconstruction of production relations. Specifically, AIGC has improved the efficiency of news production and lowered the threshold for news production. Using AIGC technologies such as ChatGPT, audiences can generate customized news information and comments based on their own information needs. As a result, traditional audiences have completed their identity transformation, from passive information consumers to active news producers, which will change the pattern and existing cognition of the journalism industry. This is the trend that journalism should be most wary of and needs to deal with.
Of course, advanced technology may change the way of production, but it cannot change the place of responsibility. For journalism in particular, humans will always be the moral actors and ultimate gatekeepers behind AI, even if all articles are generated by the AIGC. From this perspective, the responsibility of human beings will be more important. It will also become more and more important to strengthen the responsibility of the main body, strengthen the verification, and form the application ethics and norms of AIGC.
The term "news" not only refers to the "news reports" we can read, but also refers to the journalism industry and the news traditions it carries, including values, operating norms, ethical principles, and so on. As an unconscious subject, AI has never been able to inherit and follow these traditions, which are the basis for the existence and continuation of journalism.
ChatGPT will not replace journalists, just some of their jobs. Experienced journalists have high sensitivity, insight and empathy for news events, and can extract news value and put it into writing with fluent words. These subjective characteristics are the abilities that ChatGPT cannot replace. With the waves washing away, excellent journalists and authoritative news organizations will become more and more important. The origin of instrumental rationality is bound to stick to value rationality. For the journalism industry, strengthening professionalism and authority, emphasizing investigative reporting and explanatory reporting will be a way out in the AIGC era.
Many people think that ChatGPT has already appeared, so let GPT write articles and even replace journalism. But this view obviously ignores the complexity of journalism and the significance of its existence. The real journalism industry is "the lookout at the bow", safeguarding the public interest and expressing the demands of the people. This is the responsibility of the journalism industry and the starting point for the struggle of generations of journalists. Technical tools cannot understand this passion, nor can we attempt to transfer responsibility and professionalism to ChatGPT line by line. AIGC can never replace journalism at this point.