About the author
Xiaohua Lu is a chair professor at the School of New Media and Communication, Tianjin University.
Summary
The essence of intelligent content generation tools such as ChatGPT is an intelligent content generation system rather than a "chatbot"; It is an intelligent content generation tool trained according to the rules set by humans with a position and controlled to a certain extent, and cannot generate content purely objectively based on user problems; They are not imaginative, critical and creative, but they can replace humans in doing certain jobs and help humans become more efficient and productive. The content generated by intelligent tools began to influence human cognition, decision-making, and behavior, which meant that the way and logic of news communication activities were generated significantly changed. The content generated by intelligent tools has strong persuasiveness due to the pertinence of the answer to questions and the knowledge of the content presentation, which may replace the dominance of human communication subjects in the dissemination of intellectual content to a certain extent. Intelligent content generation technology is applied to the field of news communication, and the situation that human production content and artificial intelligence generated content jointly affect human beings will undergo a major transformation in the operation paradigm of media and other professional organizations, forming a new operation paradigm of collaborative production of content between humans and artificial intelligence.
keyword
Intelligent content generation; Influence; generate logic; Run the paradigm
ChatGPT, an intelligent content generation tool that can generate content based on user questions, presents landmark major advances in understanding the intent of questions, accuracy of generated content, and ease of use. Users have formed a huge involvement effect on their feelings and the spread of trial results, which not only enabled ChatGPT to accumulate hundreds of millions of users within two months of its launch, professionals and institutions in many fields to try it, but also many Internet companies and technology companies concentrated on publishing their own research results or entering further trials in more than ten days. What exacerbates concerns about intelligent content generation tools is that the images generated by users using the subsequent release of intelligent image generation tools that resemble "news pictures" are widely disseminated, enough to confuse people's perceptions. After Trump posted on social media on March 18, 2023 that he expected to be arrested next Tuesday, a user used the Midjourney V5 image generation tool released on March 16, 2023 to generate multiple photos of "Trump arrested", which were quickly forwarded for multiple rounds.
The rapid development of intelligent content generation tools raises a series of important questions that are worth studying and must be answered. What is the essence of a smart content generation tool like ChatGPT? Does the content generated by intelligent tools be discussed, shared, and disseminated, mean that the logic of influence generation has changed significantly? Does the availability of intelligent content generation tools mean a major paradigm shift in the operating paradigm of journalism? This series of questions is worth pondering and exploring answers.
What is the essence of intelligent content generation tools such as ChatGPT?
ChatGPT is an intelligent content generation tool driven by the artificial intelligence technology launched by OpenAI on November 30, 2022, which can have a continuous dialogue with humans on any topic by understanding and learning human language, and can generate corresponding content according to the dialogue, and even write emails, video scripts, copywriting, translations, code, papers, etc., with "high-quality dialogue, complex reasoning, thought chain (CoT), zero/few sample learning (contextual learning), cross-task generalization, code understanding/generation, etc." aspects of better performance and more "impressive" capabilities than previous intelligent content generation tools (Zhou Jie and Zhang Junping, 2023). Its essence is to train intelligent content generation tools with positions and a certain degree of control based on the rules set by humans; An intelligent content generation tool that generates content based on a corpus of connected existing knowledge, information, and problems, without imagination, speculation, and creativity; It is a productivity tool that can help humans improve efficiency or replace some of their jobs.
The essence of ChatGPT is, first of all, an intelligent content generation system. It is not what some articles and discussions call a "chatbot". "Chatbot" is defined in a way that limits people's perception of intelligent content generation technologies and the impact they can have, as well as their use of intelligent content generation technologies. Most of the answers to the user's questions by chatbots are pre-written to the database and retrieved according to the results of the user's speech understanding. Traditional search engines search for articles, data, posts, pictures, images, etc. that already exist on the Internet, and do not generate new content. Instead of simply processing existing content, ChatGPT intelligently generates content based on user problems with algorithms trained by manually annotated datasets. Intelligent content generation tools include artificial intelligence systems with strong natural language understanding and text generation capabilities such as ChatGPT, and intelligent image generators such as Dall-E 2 that generate images based on written text. Dall-E was launched by OpenAI in January 2021 and is said to be named after painter Dalí and Wall-E. DALL-E 2 was released in April 2022 and was put in public beta in September of the same year. Users only need to simply enter some text descriptions, and the system can display the scene described by the text in the form of pictures. According to the official website of OpenAI, as of September 28, 2022, more than 1.5 million users have used DALL-E 2, creating about 2 million images per day; As of November 3, 2022, more than 3 million people are using DALL-E 2, creating approximately 4 million images per day. Both the number of users and the number of images generated have doubled.
In fact, a few years ago, there were already some writing robot programs to assist in writing texts in finance and other fields, mainly using the data released by the specification to generate manuscripts according to a certain format template. In July 2014, the Associated Press announced that it would use writing software to write 150 to 300 words of company performance from this month. However, compared with ChatGPT, an intelligent content generation tool based on large model training, it is difficult to compare with the accuracy of user intent understanding, the pertinence, completion and ease of use of the generated results. Similarly, five years ago, I investigated and tracked the company Wibbitz and video production platform Wochit, which can generate interactive videos based on manuscripts, and wrote in "What new communication changes are being generated by intelligent content generation?" The article introduces the research situation five years ago (Lu Xiaohua, 2023), and it can be clearly seen that its size is too small, and the technical route has been surpassed by later large models and training methods. As Bill Gates calls "the world that will change our world" "as important as the invention of the internet," ChatGPT "uses language models (LMs) to train huge neural network models on massive amounts of data," and "existing research shows that the larger the model size and the amount of data, the better the performance." When models and data scale to a certain point, models gain emergence." As a result, ChatGPT "achieves a huge performance improvement on zero-shot learning tasks, with contextual learning capabilities that small models do not have." In addition, it adopts training strategies such as code pre-training, instruction fine-tuning, and reinforcement learning based on human feedback to "further improve reasoning ability, long-distance modeling, and task generalization" (Zhou Jie and Zhang Junping, 2023) [1]. The intelligent content generation technology represented by ChatGPT is a technology group with great potential, change and driving force, which will make people accustomed to great changes in the way of information acquisition, content production and dissemination, and will give birth to a variety of intelligent content generation products, which will have a strong impact and even impact on many fields and industries.
Secondly, ChatGPT is not purely objective to generate content based on user problems, but is actually a positioned, controlled intelligent content generation tool trained according to the rules set by humans. This knowledge is a prerequisite for observing, piloting, and even using intelligent content generation tools such as ChatGPT. The reason why ChatGPT is an intelligent content generation tool with a position is because OpenAI uses supervised learning and human feedback reinforcement learning (RLHF) training methods to train and optimize ChatGPT, so that its generation results are more in line with human expectations, and it also sets some rules so that ChatGPT does not provide substantive answers to questions asked by users that violate these rules. Therefore, "it can answer follow-up questions, reject inappropriate requests, challenge wrong premises, and admit its mistakes" (Zhou Jie and Zhang Junping, 2023).
According to the results of the trial shared on social media, these rules include positions involving violence, discrimination, crime, etc., as well as major issues. Algorithms and models are designed by humans, and adjusting certain parameters and rules can affect the position of the generated content. Therefore, "ChatGPT, as an intelligent information tool, can not only be used to process objective information, but also hides the corresponding possibility of being used to implement subjective cognitive confrontation" (Xu Xin and Liu Weichao, 2023).
Third, ChatGPT generates content based on a corpus of connected existing knowledge, information, and questions, but it is not yet imaginative, critical, and creative. After all, artificial intelligence is a tool designed by human beings according to a certain purpose of use and evolutionary logic, after all, it is to simulate, extend and expand human intelligence, and there is no possibility that artificial intelligence can replace human intelligence. Asking questions about intelligent content generation tools such as ChatGPT is an important manifestation of human imagination, critical thinking and creativity, not to mention human emotions, empathy, empathy and so on. The designers of ChatGPT cleverly use various problem strings and problem groups that embody human intelligence to train the "intelligence" of ChatGPT, which also makes it more "smart" than past Internet applications.
Finally, smart content generation technologies and applications such as ChatGPT are useful tools that can replace humans in doing certain jobs and help humans improve efficiency and productivity. It can not only replace some informational, simple and inductive auxiliary work, help write code, check possible problems with computer programs, and even assist people to engage in some creative work. For example, ask questions to intelligent content generation tools and use the content they generate to enrich concerns and clues to expand the space for thinking and imagination.
ChatGPT is an intelligent content generation tool that is used by humans while also influencing human development and cognition itself.
Will intelligent content generation change the logic of influence generation?
The content generated by artificial intelligence tools step by step according to user problems is obtained, discussed, shared and disseminated by people, which has a significant impact on people's information extraction, cognitive formation, opinion expression, and decision-making behavior, which means that the content generated by artificial intelligence tools begins to affect human cognition, decision-making and behavior, which means that the logic of influence generation has undergone new major changes.
Influence is the ability to change others' attention, cognition, judgment, decision-making, and behavior. It includes both the coercive influence formed by cultural customs, legal rules, hierarchical power, force, etc., and the non-coercive influence formed by the sharing of information, knowledge, insight, etc. in social interactions. In news and communication activities, the content production and dissemination activities of media and other communicators have a much greater impact on the public and decision-makers than ordinary individuals in social communication activities. Even if individuals use social platforms to become influential communicators, the influence of professional organizations such as Internet platforms and media is still far greater than that of individuals.
The application of major changes in science and technology to the field of news and communication will not only change the form of media and communication, but also make major changes in the mode and logic of generating influence of news and communication activities. Historically, there have been at least six major changes in the way news campaigns generate influence and the logic of generating it.
First, printing technology realizes the wide and continuous dissemination of printed matter such as books in a wide area and in different time and space, amplifying and extending the content value and vitality of printed matter and books, so that large-scale communication occupies an important share in influence generation and changes the logic of influence generation. The scope of dissemination of recorded content such as handwritten copies and inscriptions is relatively limited. Large-scale communication supported by printing promotes the dissemination of knowledge, changes people's perceptions and behaviors, and changes social processes. China's Tang Dynasty woodblock printing and Song Dynasty movable type printing have greatly promoted the wide dissemination of knowledge and Han culture. In Europe, printing made Martin Luther's Ninety-Five Theses affixed to church doors widely disseminated, accelerating the Reformation.
Second, after the industrial revolution, modern media with specialized division of labor, organized operation and large-scale production appeared, and the content and communication activities of media collection, processing and dissemination occupied a major position in influence generation, forming an influence generation logic that played a major role for a long time. The reason why newspapers appeared after the Industrial Revolution was not only based on modern printing presses, but also on industrial production methods. Modern newspapers not only lay the basic mode of media operation with collection, editing and distribution, but also form an industrial communication mode with timing, quantitative, targeted, classification and selection as keywords, a media organization form based on professional division of labor and a series of rules and constraints, and a basic concept and knowledge framework to meet the needs of readership by classification. Subsequent media such as news agencies, radio stations, and television stations developed based on the operation mode, basic concepts, and knowledge framework formed by newspapers. Even mobile communication is newer or more efficient in different keywords of industrial communication mode.
Third, the public uses the Internet and social media to become the main body of communication, not only producing content but also cooperating with Internet platforms and media to produce content, so that the public's production of content and participation in communication activities occupy an increasing share in influence generation, and compete with media influence, making major changes in the logic of influence generation. The Internet has gradually become the technical foundation of human information dissemination, carrying and giving birth to various forms of websites, social networks, and mobile social platforms, and evolving new communication tools such as blogs, podcasts, SNS, and wikis. The mobile communication form began with SMS, and the mobile communication form after that was mobile TV. The new media of SMS reminds people to discover the new functions of mobile phones, and develop mobile phones, which are the most closely connected and sticky information terminals with people's inexhaustible imagination. As a result, the Internet has become mobile. The development of mobile communication has made matching mobile demand a basic rule, and the evolution path of communication form has been defined from the dimension of on-demand content distribution anytime, anywhere. With the help of various communication tools of the Internet and mobile Internet, the public has become the main body of communication, and the influence that no social subject and social manager can ignore, which not only makes major changes in the communication pattern, but also profoundly changes the way of political operation, economic operation and social operation.
Fourth, multi-dimensional, multi-directional, multi-round, and diversified re-dissemination based on social platforms has become the main driving force for the formation of communication influence, and the re-transmission effect has replaced a communication as a key factor in influence generation in a sense, and occupies an increasing proportion, profoundly changing the logic of influence generation again. One of the directions of Internet development is the enhancement of sociability, and the Internet application with social functions not only promotes the form of social communication to become the mainstream, but also the re-dissemination and multiple dissemination formed by likes, forwards, comments, etc. account for an increasing share in information dissemination. Re-dissemination occupies an important position in the influence generation mechanism, making various social subjects, not just communicators, the prerequisite for effective communication. Sharing and various re-dissemination based on social relationships have become the main form and force of information dissemination, and even formed a communication mode of "mobile knowledge, home screen viewing, social recognition, and empathy driven", so as to extend and amplify the change of re-communication to the logic of influence generation to other communication-related fields. Under this influence generation mechanism, it is difficult for a transmission to produce the desired impact of the communicator without the help of retransmission. Correspondingly, an important choice for media in the integration transformation is to replace the simple circulation and other one-time communication indicators as a measurement tool with re-dissemination effect indicators.
Fifth, algorithm push has become one of the main mechanisms of content distribution, so that the proportion of distribution links in influence generation continues to expand, and the real-time matching personality information requirements realized by algorithm push profoundly change the influence generation logic from a new dimension. Different administrative regulations and normative documents have different names for recommendation algorithms. On August 28, 2020, the Ministry of Commerce and the Ministry of Science and Technology adjusted and released the "China Export Prohibited Export Restricted Export Technology Catalog", which is called "personalized information push service technology based on data analysis". On January 4, 2022, the Provisions on the Administration of Internet Information Service Recommendation Algorithms issued by the Cyberspace Administration of China and other four ministries and commissions divided Internet services using algorithm recommendation technology into application generation synthesis, personalized push, ranking and selection, retrieval and filtering, scheduling decision-making and other algorithm technologies to provide information content to users. From another perspective, algorithms are strategic mechanisms that describe and solve problems in a systematic way based on data. More than a decade ago, with the intervention of algorithms in the information distribution of mobile platforms as the main symbol and entering the era of intelligent communication, although the Internet platform using recommendation algorithms did not produce content, it gathered more than 600 million daily active users in eight years, which has great influence. Intelligent communication that can realize algorithm distribution, personality push, and accurate matching continues to develop, which not only makes real-time matching of people's individual needs a reality, but also further improves the weight of distribution links in influence generation, and even the recommendation algorithm is given more and more content distribution decision-making power, thereby more profoundly changing the underlying logic of content dissemination and influence generation.
From the content production of traditional news activities accounting for the main weight in influence generation, to the continuous expansion of the weight of distribution links in digital news activities in influence generation, recommendation algorithms occupy more and more content distribution decision-making power, objectively squeezing human control over information distribution. This change in the logic of influence generation has far-reaching implications and has led to questioning and research on many related issues. Real-time matching, rather than general satisfaction or delayed gratification, has promoted the continuous optimization of personalized information push service technology based on data analysis, which not only realizes real-time matching of people's individual needs, but also is moving towards the ability to match people's personalized experience requirements in real time. This is also the basis for the realization of ideas such as the metaverse.
Sixth, artificial intelligence tools enter the field of communication according to the content generated by user problems, and occupy an increasing proportion in influence generation, in fact, once again historically changing the logic of influence generation. It is at least manifested in: First, the content generated by artificial intelligence tools has a profound impact on human information acquisition, cognitive formation, and opinion expression, which means that it is no longer only the content collected, processed and distributed by humans that affects people, it is no longer just the information and knowledge extracted by humans that affect people, and the content generated by artificial intelligence tools according to user problems is also profoundly affecting people. Second, the content generated by artificial intelligence tools not only presents the pertinence to the problem, but also has strong persuasiveness because it shows knowledge, which may replace the control of human communication subjects over knowledge content dissemination to a certain extent. The competition for the right to interpret and speak is the focus of competition in human society, not only in the fields of international politics, but also in the field of news and communication. The emergence of intelligent content generation tools not only increases the new competitors for the right to explain and speak, but also becomes a competitive tool for the right to explain and speak in international politics. Third, the efficiency, ease of use, convenience, and high matching of intelligent content generation tools with the needs of use and other factors have continuously increased people's usage rate and dependence, which will continue to deepen the change of influence generation logic.
3. How will intelligent content generation change the operating paradigm of the news and communication industry?
The paradigm is what Thomas Kuhn calls "an accepted model or pattern" (Thomas Kuhn, 1962), "refers to the theoretical system accepted by the members of the scientific community, and is a way of thinking to grasp the object of study" (Xue Jinghua and Chen Guangyu, 2023). The so-called operation paradigm is a basic mode of operation, and it also implies a basic way of understanding related things. Before the emergence of intelligent content generation tools, it was generally human content production, media, publishing institutions and other professional organizations formed around human content production, and the main body of content production process was people, and the content production process was formed around the collection, processing, distribution, dissemination and other activities of journalists. The modern media that emerged after the industrial revolution and continued to this day with specialized division of labor, organized operation, and large-scale production is premised on people as the main body of production and control, content production organization and content production activities are led by people, and the technology and equipment used are only tools used by people. The era of digital technology is still "content is king". Even if digital technology empowers ordinary people to make everyone speak out, content recommendation algorithms and other in-depth application to content distribution and move towards intelligent communication, content is still basically produced by people, it is still the foundation of the media, and it is still the basic bearer of values and communication intentions. Content is king remains the basic survival principle of mainstream media. Therefore, the implicit premise of the cognition that "content is king" and the corresponding operating paradigm is that humans are producing content.
Once intelligent content generation tools enter the realm of professional content production, they can trigger far-reaching operational paradigm shifts. With the application of digital technologies marked by big data and artificial intelligence to news gathering, production, distribution, reception and feedback, the logic of content production and operation has undergone important changes. Not only has AI-assisted content production appeared, but the volume and availability of AI-produced content such as writing robots has increased dramatically. The content generated by ChatGPT enters the dissemination, and the combination of intelligent content generation tools and industry application scenarios will further promote intelligent content generation tools into the field of news communication, so that news activities will move from the traditional operating paradigm of content production with the help of machines, mainly by people, to a new operating paradigm of large-scale content production jointly carried out by people and artificial intelligence. The analysis here of the paradigm shift that will occur in large-scale content production operation by humans and artificial intelligence is not to say that the emerging operating paradigm is superior to the original operating paradigm, objectively speaking, the relationship between them, as Thomas Kuhn put it, is "not only logically incompatible, but also practically irreducible" (Thomas Kuhn, 1962). This kind of digital news activity of human and artificial intelligence cooperating to produce content not only provides digital journalism with a research object that cannot be ignored and cannot be abandoned, but also puts forward a series of serious topics, which require digital journalism and other disciplines to separately and collaboratively answer how to construct the theoretical framework, ethical rules, policies, regulations, and regulatory methods of large-scale content production between humans and artificial intelligence.
Historically, every change in the tools of labor has brought about changes in the mode of production and with it changes in the social form. The important driving force for the transmutation of reporting forms, media forms, and communication forms is the development of information technology, the introduction of new content production tools and communication technology means and the effective combination of news and communication activities. This combination of new technology and tools with certain job needs may change the operating paradigm. In many industries and fields, including journalism, the use of certain technologies and tools has led to changes in operating modes many times.
Large-scale content production by people and artificial intelligence is not only content production with the help of machines, but also not only co-production by people and artificial intelligence, but also the continuous production, automatic generation, and direct content production and dissemination of artificial intelligence may occur. As far as content quality control is concerned, there are several fallacies derived from hidden concerns. First, in order to adapt to the characteristics of mobile media and changes in people's needs, people are accustomed to using a variety of means of expression to produce digital content; Digital content itself uses a variety of expression means and forms of expression, but does not necessarily enhance the expression effect, and sometimes detracts from the expression effect. However, digital content producers are not necessarily aware of how this impairment occurs and how it can be curbed. After all, digital content is often a collection of expressions, such as text, audio, video, and data. Each means of expression may be appropriate individually, but the fallacy of synthesis can occur when applied to a digital content product. The so-called synthesis fallacy means that the means used appropriately respectively, but the effect produced by synthesis, when used together, may not be appropriate; Locally reasonable and efficient choices add up to a fallacy. Second, the production of digital content is continuous, automatically generated, and accumulative, which may lead to the accumulation of small errors in the generated content and constitute a big error, which may be called the cumulative fallacy. Third, when the production of digital content is continuous, very few abnormal points may lead to implicit deviations in the generation results. Fourth, the results of the operation of the algorithm of continuous deep learning may deviate from the preset or be difficult for the designer to explain. This requires digital journalism and other disciplines to carry out research, deep understanding, systematic research, and discovery of laws to guide how to effectively control the production of digital content participated by artificial intelligence in practice, and use artificial intelligence more controllably in digital content production and dissemination based on the principle of explainability and traceability, and guide people and artificial intelligence to collaborate on large-scale content production.
The entry of intelligent tool-generated content into the field of knowledge production and dissemination means that intelligently generated content may enter the human cultural inheritance system. Therefore, people need to face a series of major problems involving the fundamentals of human civilization, and need to face a series of special problems about how humans and artificial intelligence coexist. All of this requires us to deeply analyze, think systematically, and explore answers.
exegesis
[1] 译自Zhou, J., Ke, P., Qiu, X. P., et al. (2023). ChatGPT: Potential, Prospects, and Limitations. Frontiers of Information Technology & Electronic Engineering. https://doi.org/10.1631/FITEE.2300089.
(Cover image from the Internet)
The reference of this article is omitted, and the original article was published in the 4th issue (204th issue of 2023) of the "University of Journalism". Please check the original text for quotations; For reprinting, please contact the editorial office of this journal.
Long press to identify the QR code
Follow us
National News Core Journal
The core journal of the Statistical Database of Academic Papers of Chinese Social Science Journals
Chinese first source journals for the Social Science Papers and Citations Database
University of Journalism
Phone: 021-65641289
Address: School of Journalism, Fudan University, No. 440 Handan Road, Shanghai
Email: [email protected]
WeChat: Fudan_xwdx
Online Submission System URL:
http://devjava.odb.sh.cn/fudanNews/client/contribute.html
![Affected by ChatGPT and other positive news, Microsoft and Google's latest financial reports have increased significantly[fig]](data:image/gif;base64,R0lGODlhAQABAIAAAP///wAAACwAAAAAAQABAAACAkQBADs=)