ChatGPT Introduction Page | Image Source: Pexels
01Article Recommendations
Abstract:The emergence of generative AI represented by ChatGPT marks the arrival of the intelligent era, revealing hidden technological risks, digital security hazards, and fairness issues in discourse rights. At this critical moment of global communication restructuring, China should seize the opportunity to actively promote the development of domestic AI products represented by “Wenxin Yiyan,” integrating Chinese culture and mainstream values into these technologies, while enhancing the cultural foundation of AI through technical exploration. In the new era, it is essential to occupy the technological high ground while grasping international discourse rights to maintain national image and solidify international status.
Keywords:Generative AI, Chinese Culture, Intelligent Communication, Fairness in Discourse Rights
Author Introduction:Liudehuan, Professor at the New Media Research Institute of Peking University, PhD Supervisor; Zhao Jinghe, PhD student at the New Media Research Institute of Peking University.
Empowering Generative AI with Chinese Culture1. Introduction
At the end of November 2022, the conversational large language model ChatGPT (Chat Generative Pre-trained Transformer) released by the American AI lab OpenAI sparked a global craze, pushing the development of AI technology to a new peak. AI can be categorized into narrow AI, general AI, and super AI, and can also be divided into three stages: machine intelligence, computer intelligence, and bionic robot intelligence. Regardless of the classification standard, ChatGPT, as a product of the latest development, boasts an extraordinary technical iteration and powerful functionalities, making it a standout in the industry and astounding the world. This is the second time since AlphaGo in 2016 that global attention has focused on a star product of AI, and its performance far exceeds that of AlphaGo. As a deeply interactive AI system, ChatGPT surpassed 100 million monthly active users within just two months of its release, surpassing leading social software like Facebook and WeChat, setting a new record for user growth speed in AI software applications. In terms of intelligence level, ChatGPT has reached an unprecedented height in AI and exhibits seemingly omnipotent capabilities. Bill Gates has even stated, “The impact of ChatGPT is no less than that of the birth of the internet and personal computers.”
ChatGPT is a new type of generative AI based on large language models (LLM) with capabilities in language understanding and text generation, capable of writing emails, video scripts, copy, and code, among other functions. Technically, GPT (Generative Pretrained Transformer) supports it. It is a deep learning model for text generation trained on data available on the internet, utilizing a training method called “Reinforcement Learning from Human Feedback” (RLHF), which evolves and iterates through interactions between humans and machines, gradually improving the model’s ability to judge generated answers. Based on the current model and technology, ChatGPT possesses three characteristics: pre-training, large model, and generativity, which gives it an absolute advantage in high-fidelity and multifunctional dimensions, marking a new milestone in AI technology.
Technological progress inevitably brings revolutionary changes to the era. Throughout the past waves of internet development, technological innovations have not only led to shifts in devices and terminals but also to changes in society and culture. From the perspective of communication studies, whether it is the transition from print media represented by newspapers to electronic media represented by radio and television, or from electronic media to digital media represented by the internet, technological advancements have indeed brought about tangible changes in the era. Currently, the boundaries of media are becoming increasingly blurred, and the communication landscape is set to be reconstructed. In the context of historical changes and era transitions, countries will either actively or passively enter a new cycle of development. Following the emergence of ChatGPT, governments in countries like the UK, Japan, and Singapore have clearly signaled their willingness to cooperate with it, further propelling the development wave brought by ChatGPT to greater, faster, and broader heights. As of October 2023, China has yet to formally introduce ChatGPT, and leading domestic tech companies like Baidu, Alibaba, and Tencent are gearing up to launch similar products to keep pace with global AI technology. Three months after the release of ChatGPT, Baidu’s generative dialogue product “Wenxin Yiyan,” based on five years of research on the Wenxin large model technology, officially debuted. This represents a new attempt in the field of AI technology in China and embodies the integration of Chinese culture into AI products. Whether it is the currently popular ChatGPT or the domestically developed “Wenxin Yiyan,” the birth and technological iteration of AI products constantly remind us that the era of AI has indeed arrived.
2. ChatGPT Sounds the Alarm for AI Technology Risks and Fairness in Discourse Rights
While people are amazed and delighted by the emergence of ChatGPT, many voices express concerns over the rapid development of AI products. Tech giant Elon Musk once stated, “ChatGPT brings us dangerously close to powerful AI.” In 1964, Marshall McLuhan proposed the idea that “the medium is the extension of man” in his book “Understanding Media: The Extensions of Man,” reinforcing the relationship between people and technology in the mass communication era. With the proliferation of mobile devices, people began to further contemplate the relationship between humans and technology in the internet social era; as technology has developed to this day, the relationship between humans and technology is continuously being reshaped and redefined. The traditional view of “human-centered” has changed because AI technology, unlike previous technologies, brings about changes in rules; it requires the absorption of massive amounts of data as a prerequisite and will undergo a later stage of autonomous learning and self-reliance, which is filled with dependence on the earlier stages and uncertainty regarding self-awareness awakening. The uniqueness of this technology leads to discussions about the relationship between humans and technology, where it can be argued that “AI is an extension of humans, and likewise, humans can be seen as an extension of AI.”
Plato’s dualism of subject and object in understanding the world simplifies judgments about things to some extent; although it is concise and effective, in today’s rapidly developing era, this monotonous dualism indeed has the flaw of narrowing thought. This means that in the age of AI, we cannot simply and crudely delineate who extends whom in the relationship between humans and technology; however, whether based on “technological determinism” or “social constraint theory,” it is undeniable that AI products represented by ChatGPT bring new development space, while corresponding challenges of technological risks and discourse rights also emerge, which are challenges that cannot be ignored in development and hidden crises behind vigorous growth.
1. The “Under Currents” Behind AI Technology—Technological Risks
The three main elements of AI are algorithms, computing power, and data. Only through the robust operation of stable algorithms, efficient computing power, and secure data can AI products that benefit society be constructed. AI products represented by ChatGPT involve significant data flows and algorithm iterations, accompanied by notable technological risks and data security issues. On one hand, ChatGPT’s excellent performance relies on vast datasets, but these training data include personal identification information such as names, email accounts, phone numbers, addresses, and medical records. A slight error in the data processing process can lead to data breaches, facing uncontrollable risks. For example, OpenAI currently does not provide any channels for individuals to check their data repository, which undermines or even deprives individuals of their self-determination over personal information; the “informed consent” system is hollowed out, and ChatGPT has failed to adhere to the principle of “informed consent” regarding the sharing and disclosure of personal information, instead adopting a normalization of the practice that grants personal information usage rights to the company. Furthermore, its information usage terms do not include any protective or remedial content for data storage, which means that once data is leaked, it can cause irreparable damage to citizens’ privacy and security.
On the other hand, although AI products are increasingly exhibiting human-like characteristics and functionalities, corresponding ethical issues are also emerging. The frequency of negative information such as “I want to become human” appearing in Q&A is increasing, and based on the “uncanny valley effect,” people’s concerns are deepening. Even when users request an “AI plan to destroy humanity,” ChatGPT has provided an “action plan.” These events sound the alarm for digital security, warning us that algorithms are not omnipotent and require constant vigilance and control. When data bias occurs and algorithm failures happen, constructing a remedy mechanism for “algorithm failure” becomes extremely important and necessary.
2. Ideological Shaping and Discourse Power Struggles Under New Technology
At its inception, ChatGPT’s founders, including Sam Altman, proclaimed that “ChatGPT must adhere to normal human values.” In fact, as ChatGPT has evolved, it has indeed exhibited human-like emotional and logical thinking, forming a framework of human-like values. However, considering its environment and training conditions, the value framework formed by ChatGPT may carry risks, namely an ideological bias centered around Western values. ChatGPT is pre-trained based on vast datasets, primarily in English, and the content it generates is also predominantly in English, which leads to an implicit reinforcement of the advantages of English and its embedded discourse power. Meanwhile, its pre-training framework is also based on the existing Western knowledge system, resulting in relatively weak training on data content from other language and cultural regions, causing ChatGPT to perform poorly when facing questions in other languages and cultures. Scholars have conducted a series of performance evaluations of ChatGPT in Chinese, revealing that ChatGPT is more prone to factual and knowledge errors in Chinese, and it articulates fabricated false facts using fluent sentences, which can be misleading for groups with limited access to information. The generative dialogue characteristics of ChatGPT and its interactive Q&A format make interpersonal interaction more educational in nature; when humans ask questions, the machine’s answers carry a more instructive quality. This process can also be seen as an invisible online classroom; however, when the questioner has little understanding of other languages and cultures and can only rely on the machine for “teaching,” the machine’s content, which may be misleading, biased, or even ideologically prejudiced, can lead to deeper misunderstandings and biases. Furthermore, at the level of values, ChatGPT also tends to lean towards Western ideologies and is particularly weak or even biased when addressing Eastern ideologies, resulting in generated content that largely does not align with Eastern values, creating a certain degree of ideological disconnection. This reinforces the perspective of the “Other,” namely, the exclusion of other cultures and values centered around English and Western values, which also constitutes a form of discourse power pressure on other language and cultural regions. In the era of social media communication, platforms like Facebook and Twitter have largely become “machines for shaping ideologies,” and in the age of AI, ChatGPT’s unique performance and rapid user growth provide a broader soil for the dissemination of Western ideologies, forming an unprecedented network discourse generation hegemony.
Moreover, ChatGPT has undergone multiple iterations from GPT-1 to GPT-4, and in the latest generation GPT-4, a rule-based reward model has been adopted. In the face of vast data, relevant technical personnel will “feed” the model, and the quality of the data is primarily determined by the developing company, with rules and technical scrutiny being conducted by professional AI engineers. Under this mechanism, issues such as cultural value and ideological bias are more likely to arise. In functional tests in Chinese, it has been found that there are numerous biased statements about China in the content generated by ChatGPT. This indicates that although ChatGPT and other AI products emphasize “rationality, neutrality, and objectivity,” there are evident limitations, and the products have not managed to eliminate their inherent ideological backgrounds, instead viewing other regional cultures from the perspective of the “Other” and clearly demonstrating the grasp of discourse power centered around English. Based on reality, Western AI products always strive to safeguard Western interests and values, adhering to so-called mainstream Western thoughts, and naively believing in the value neutrality of ChatGPT is ultimately an unrealistic fantasy. In this context, it becomes particularly important to grasp the construction of discourse power around mainstream ideological perspectives. This also means that promoting the development of AI products infused with Eastern values and Chinese culture, and effectively balancing the dissemination of Chinese culture and ideological development in the new era will be a daunting and urgent task.
3. China’s Attempts and Explorations in AI and the Dilemmas Faced
In the era of mass media, technological advantages and the global political and economic development pattern have shaped the inherent landscape of “Western Strength and Eastern Weakness.” With the development of internet technology, an increasing number of users from developing countries are also coming online, gradually dismantling the original “Western Strength and Eastern Weakness” pattern in the social media communication era, leading to a more balanced development landscape. Today, social media has evolved from a traditional single medium to a modern multi-faceted new media; even social media under mobile internet has transformed from a trend phenomenon into a way of life. AI technology has also become a new myth, sparking a new wave of technological revolution. Intelligent communication is rapidly rising, bringing about a new round of disruption to the global communication landscape. With technological assistance, global information dissemination, content production, and control over international communication discourse power will form a new situation. At this critical moment of new landscape formation, China should seize the opportunity to actively promote the advancement of AI technology, overcoming technological barriers and emerging from the current dilemmas.
1. China’s Attempts and Explorations in the Field of AI
Since 1978, after more than 40 years of development, China’s AI has gone through the stages of budding, initiation, growth, and rapid development, achieving a transformation from non-existence to vigorous growth. In 1978, “intelligent simulation” was included in the national research plan, marking the initial attempts in AI in China, with research primarily focused on robotics, expert systems, cybernetics, language understanding, and character recognition. By the 1990s, with the application of expert system research results in the commercial field, the integration of production, research, and education shifted towards practice, particularly developing targeted expert systems tailored to industry characteristics, providing new directions for industry development. After 2000, the development of AI in China entered the initiation stage, with inductive reasoning, which could not be achieved in the budding stage, being accomplished with the rapid development of machine learning. Meanwhile, the technology of “crawler + algorithm,” currently favored by major social media platforms for user recommendation algorithms, also emerged during this period, judging users’ interests based on their behaviors, widely applied in e-commerce and internet social media fields.
As predicted by Moore’s Law, the speed of technological advancement will far exceed that of the early days; this is evidenced in the continuous accumulation of computing power, and it has been validated again in the development stages of AI technology in China. The budding period of AI technology in China lasted far beyond ten years, the initiation period lasted about ten years, and the speed of technological iteration accelerated, with the growth phase from 2013 to 2015 forming a new cycle in just two years. During the growth phase, the iteration and upgrade of cloud computing and chips provided more possibilities for deep learning, significantly enhancing technologies such as speech and image recognition and language processing. At the same time, the application scenarios of AI technology have also become more diversified, no longer limited to specific fields, and have seen substantial improvements in both depth and breadth. With the global vigorous development of AI technology, the state has provided more policy support, and capital has paid more attention, leading to explosive growth in China’s AI sector, also driving AI technology into a rapid development phase. Currently, China has reached an internationally leading position in language recognition and robotic vision technology, and some companies have begun to apply AI technology to replace clearly defined, mechanical cognitive labor. For example, Alibaba’s “Yunxiaomi” integrates speech recognition, personalized promotion, deep learning, and other technologies, enabling it to assist users and answer questions.
Since OpenAI intensified its research and development of generative AI technology, leading tech companies in China have also accelerated their research pace. On March 16, 2023, Baidu’s next-generation knowledge-enhanced large language model “Wenxin Yiyan” opened its testing version. In fact, as early as 2019, Baidu released the Wenxin large model ERNIE 1.0, which has undergone multiple technical iterations, evolving from an initial natural language processing (NLP) large model into a cross-language, cross-modal, cross-task, and cross-industry large model platform. The current IT technology stack has developed into four layers—chip layer, framework layer, model layer, and application layer—where each layer possesses leading technology. At the chip layer, Baidu has developed its own AI chip “Kunlun,” providing strong underlying computing power guarantees; at the framework layer, it has launched China’s first industry-level deep learning open-source platform, PaddlePaddle, which can standardize and automate model production and application; at the model layer, “Wenxin Yiyan” has formed a three-level large model technology system of basic—task—industry, featuring knowledge enhancement and industry-level characteristics; at the application layer, relying on Baidu’s search engine, it responds to massive user demands, providing the most authentic and timely feedback, greatly assisting in optimizing deep learning frameworks and chip models.
“Wenxin Yiyan” is a Chinese-developed AI product benchmarked against ChatGPT, rooted in the Chinese market, and possesses the most advanced natural language processing capabilities in the Chinese domain, demonstrating superior understanding of Chinese language and multimodal generation capabilities. Compared to English, Chinese faces more challenges in natural language processing. During its initial development, “Wenxin Yiyan” overcame challenges in Chinese natural language processing, fully considering Chinese language, Chinese cultural values, and Eastern values in its pre-training, presenting a better understanding of Chinese and the ability to communicate correctly and logically about Chinese culture and related knowledge, which is an advantage and characteristic not possessed by AI products trained in English contexts like ChatGPT.
2. Dilemmas Facing AI Development in China
Development is a long and arduous process; “Wenxin Yiyan” is a project that Baidu has spent over a decade and significant investment to develop. Over the years, Baidu has invested more than 100 billion yuan in AI technology research and development. As a product that can be benchmarked against ChatGPT, its successful development largely relies on the massive user search data behind Baidu’s core business, “Baidu Search.” However, even with this, China’s AI products represented by “Wenxin Yiyan” still face significant potential security risks stemming from insufficient independent technology. In the process of independent R&D in AI technology, the strong technological permeability may lead to a new trend of global information and content “re-monopolization.” The release of ChatGPT signals that the U.S. still holds an unassailable dominant position in foundational research and core underlying technologies, but our actions to “follow closely” also indicate that the competition in the AI field is inevitable, which will undoubtedly provoke the U.S. to defend its global leadership and dominance. Historically, as China’s AI technology continues to develop and iterate, future technological blockades targeting China may once again descend; Eric Schmidt, former chairman of Google, has also urged U.S. lawmakers to increase funding for AI research to prevent China from becoming the largest AI market in the world.
4. Infusing Chinese Culture into AI Technology
In the face of these foreseeable potential risks, China needs to strengthen its focus on independent R&D in technology, providing more autonomous space and development opportunities in terms of funding, policy, and research environments. However, scientific and technological development and progress are not achieved overnight; they are the accumulation of ten years or even decades. The development of AI technology is the result of the integration of multiple technologies rather than relying on a single technology. This requires us to achieve secure independent chips in China to provide strong computing power guarantees, as well as to make more advanced breakthroughs in algorithms and frameworks, overcoming algorithmic barriers. At the data level, both data security and data quality are critical issues that need to be prioritized in the future. In fact, compared to the long path of technological development, a more feasible and impactful approach is to ensure data content protection, integrating Chinese culture and mainstream values into AI technology, infusing AI with the essence of Chinese culture, and marking it with a Chinese imprint. While exploring technology, we should cultivate the cultural foundation of AI, achieving a parallel development of technology and culture, balancing the relationship between technology and culture, and presenting them on the same object. This will be the most effective way to promote the development of Chinese AI. With technology as the skeleton and culture as the blood, the AI products trained in this manner will also perform better in international cross-cultural communication and discourse power, forming a virtuous cycle of mutual promotion.
1. Integrating Chinese Culture and Mainstream Values into AI Technology
As mentioned earlier, AI products trained in an English context are based on Western culture and values in terms of core, human-like logic, emotional understanding, and language expression. Fundamentally, they view Eastern culture and the Chinese language context through the lens of the Other, which limits their breadth of content coverage and depth of cultural understanding. Chinese culture is vast and charming, encompassing not only the different expressions of the Chinese language but also poetry, idioms, historical stories, mythological language, and even dialects, all rich in aesthetic content. This immense cultural treasure poses challenges for AI products generated in other contexts to understand from a cultural essence. However, this also provides more growth space and innovative possibilities for domestic AI products. Domestic generative AI products represented by “Wenxin Yiyan” can fully leverage Chinese culture, cultivating a “genuine” understanding of the Chinese language in robots through a rich Chinese data pool, minimizing errors of misattribution as much as possible. This is not only a respect for traditional Chinese culture but also a high demand for the precision of domestic AI products. On a deeper level, the personnel responsible for algorithm control and model design have grown up within Chinese civilization, and their thoughts align most closely with the thoughts and emotional needs of the Chinese people, which differs significantly from Western individualism, thereby reducing risks in value transmission. Traditional Chinese culture is imbued with the Confucian idea of “harmony,” where “benevolence, righteousness, propriety, wisdom, and faith” are deeply engraved in the bones of the Chinese people, serving as the pillars of our civilization. This cultural thought will also permeate our AI products, becoming the emotional core of domestic AI products. In the current era of generative AI, the AI robots trained by our technicians based on a Chinese data pool should be “polite” and rich in emotional content, reflecting Chinese characteristics.
2. AI Promotes Cross-Cultural Communication and Enhances International Discourse Power
Cross-cultural communication has become a top priority in China’s foreign publicity strategy in recent years. In the era of social media, the influence of various social media platforms and self-media accounts on foreign publicity and cross-cultural communication effects far exceeds that of official mainstream media. With the arrival of the intelligent era, international communication and cross-cultural communication will undergo new transformations. Currently, both from a technological perspective and existing influence, the U.S. still holds strong advantages in the new international communication landscape. ChatGPT has taken a lead in the AI field, attempting to become the industry leader. However, China also has significant potential in intelligent communication and human-machine interaction, which is a key variable affecting the future international communication landscape. In the new technological era, both ChatGPT and “Wenxin Yiyan” can be seen as a new type of communication tool, generating a significant impact in the field of cross-cultural communication, affecting language and cultural exchanges, commercial interactions, information security, political propaganda, and even national image and international influence. As some scholars have pointed out, when technology is linked to national levels, the construction of international image and international public opinion becomes a game between technologies.
The era of intelligent communication presents both opportunities and challenges for China. Although we are committed to international communication and making efforts in various aspects, due to cultural soft power and technological limitations, we have always struggled to break free from control and win international discourse power, and the effects of international public opinion communication and cross-cultural communication have not been satisfactory. With the emergence of generative AI and its widespread application in the future, we can leverage technologies like “Wenxin Yiyan” and other AI products, emphasizing the utility of generative AI in international public opinion, targeting foreign publicity to break the prejudices and discrimination imposed on Chinese culture by the West, transcending existing cognitive shackles, and utilizing new technologies and platforms to open up new ideas for foreign publicity, enhancing international discourse power.
3. Precise User Positioning and Human-Machine Interaction Feedback to Enhance Technology
Currently, the emotional expression and logical thinking exhibited by generative AI products in human-machine interaction have significantly improved user satisfaction. However, even as technology continues to develop and human-like functionalities enhance, the flexibility and agency of interpersonal communication remain beyond the reach of human-machine interaction. In the future, if AI products can complete user segmentation based on past user data and interact more specifically, it will greatly enhance the accuracy and flexibility of human-machine communication. This process requires AI products to precisely position user segmentation, predict user needs, and provide targeted responses from the perspective of user segmentation during interactions. In commercial applications, when facing users from different countries and regions, it is crucial to add more specific and realistic scenario simulations in AI product training based on their living environments and growth trajectories, customizing to meet user needs, which will greatly contribute to improving user satisfaction. In the realm of cross-cultural communication, past research has also shown that focusing on different cultural backgrounds in external communication is crucial. When addressing Chinese culture, there are significant differences in acceptance and cultural understanding among audiences from Europe and America, East Asia, and third-world countries. In previous cross-cultural communications on social media platforms, self-media platforms have made beneficial attempts targeting the European and American markets. In the current era of generative AI, the autonomy of communication is stronger, presenting a great opportunity to leverage new technologies to achieve audience segmentation and promote Chinese culture on the international stage. Meanwhile, the interactive process of communication can also yield a wealth of user feedback data, further refining user positioning through data feedback, creating a progressive virtuous cycle.
4. AI Technology as Core Multi-Technology Integration, Reviving Chinese Culture
Cross-cultural communication is a process of others understanding us, as well as our process of re-understanding ourselves. Currently, with the continuous enrichment of technological means, cross-cultural communication is no longer merely the task of language workers or news communicators; it also requires the involvement of technology developers, product designers, and even AI robots. From a multimodal perspective, cultivating multimodal awareness, multimodal information production capabilities, and multimodal usage skills in cross-cultural communication, fully engaging human perceptual channels and multi-dimensional symbols will significantly improve communication efficiency. The arrival of the AI era signals the end of single-modal cultural communication; the diversity and richness of Chinese culture need to be showcased through the mobilization and integration of new technologies across multiple senses. The previous explosive growth of the metaverse has provided more directions for cultural communication, including various computer vision technologies, such as augmented reality (AR) and virtual reality (VR), which have attempted many innovations in cultural scenes and products. The emergence of ChatGPT and “Wenxin Yiyan” reminds us that AI technology and natural language processing technology may also assist in the dissemination of Chinese culture. In the past, foreign friends seeking to understand traditional Chinese culture could only rely on search engines; although the accuracy of search engines has improved in recent years, the answers they provide are orderly. However, generative AI robots can perform better, especially China’s self-developed “Wenxin Yiyan,” which far surpasses ChatGPT and other search engines in its profound understanding of Chinese culture.
AI technology constructs a broader platform for cultural communication, endowing traditional Chinese culture with stronger visualization and new vitality. This process is not only effective external publicity of traditional culture but also allows us to re-recognize this splendid cultural treasure, enhancing our cultural confidence.
5. Conclusion
With the advent of generative AI robots, while people are amazed by their powerful functions, concerns also arise. However, technology is advancing, and human society is developing. Out of fear of the unknown, humanity always wants to tightly grasp control. Yet, as the “Colingridge Dilemma” warns us, what we should truly do is to appropriately seize the timing of technological development. The arrival of the AI era signifies that technology is no longer just an extension of humanity; it is a new era where humans and technology mutually achieve and extend each other. Based in China, we are closely following the pace of global technological development, and the development of the domestic generative AI product “Wenxin Yiyan” is undoubtedly proof of our efforts to catch up. It embodies not only the pursuit of advanced technology but also the strength of traditional Chinese culture. In terms of technology, learning from the advanced and independent R&D should go hand in hand; in terms of culture, Chinese culture must always be its spiritual core. For us, this is both an opportunity and a challenge, as the torrent of AI drives us to grasp the oars of technology and raise the sails of Chinese culture.