What is the relationship between AI large models and traditional culture?

AI large models and traditional culture the relationship?

On June 21, under the guidance of the UNESCO Office in China, the China Cultural Heritage Information Consultation Center (National Cultural Heritage Administration Data Center), and the Chinese Academy of Cultural Heritage, Tencent SSV Digital Culture Laboratory and Tencent Research Institute jointly launched the “Exploration Plan 2023”.

The Exploration Plan 2023 is aimed at innovative technology groups such as artificial intelligence, digital twins, and immersive perceptual interaction, focusing on traditional cultural application scenarios such as cultural relics, intangible cultural heritage, and traditional cultural arts. It collects and selects innovative technology solutions. Tencent invests tens of millions of dollars and leads nearly 10 core technology teams to promote the co-creation and implementation of “culture + technology” solutions.

Interview Experts:

Shen Hao: Professor and Ph.D. Supervisor of the Communication University of China, Chief Scientist of the National Key Laboratory of Media Convergence and Communication Media Big Data Center, Vice President of the China Market Research Association, Invited Expert of the Exploration Plan 2023.


Tian Hui and Liu Jianghong from the School of Cultural Industry Management, Communication University of China.

Sun Yi from the Social Value Innovation Research Center of Tencent Research Institute.

1. What kind of sparks can be created by the collision between “AI large models + traditional culture”?

Regarding the collision between AI large models and the field of traditional culture, I believe there are four key points to consider. The first is the “form and content” of the fusion product. The second is the cultural attributes and artistic quality of the generated product. The third is the copyright of the generated product. The fourth is the labor employment in the cultural-related industries.

If we look at AI-generated content based on these principles, there are still the following situations:

First, it is necessary to recognize that AIGC has the problem of “form over content”. Especially when facing tasks that require logical response or involve professional knowledge or the latest information, the content generated by AI large models is likely to be “false information” lacking fact-checking.

Second, currently AI large models can create content at an extremely fast speed, leading to a massive increase in the quantity of cultural products. But can these generated products maintain cultural attributes or artistic aesthetics? This is worth discussing. For example, many digital collections generated by AI large models are only patterned collages of cultural elements, greatly discounting their artistic aesthetics and cultural value.

Third, copyright is the lifeline of cultural industrialization. However, there is a problem with assessing the contribution of “natural person creators” and “artificial intelligence creators” in the cultural content generated using AI large models. In the “Midjourney case” ruling, the U.S. Copyright Office refused to protect the content generated by ChatGPT-like products. Currently, there is no clear legal provision in China regarding the attribution of copyright for generated works.

Fourth, currently generative AI may bring about unemployment issues, but it can also create new professions such as “prompting engineers” and break through the threshold of the cultural and creative industry, bringing about a trend of “popularization” or “democratization” of development. For example, AI generation tools such as Stable Diffusion and Midjourney enable anyone with imagination to use natural language to create cultural and artistic works that were previously difficult to achieve at the technical level, achieving a vision of “creative democratization” and “everyone can create”.

Based on the above advantages and disadvantages, we bring the thinking of the collision between AI large models and traditional culture into the application scenarios of cultural relic digitization. AI large models are the combination of “big data + high computing power + strong algorithms” and can store and learn a large amount of cultural resource data for unique content presentation. Therefore, based on the national cultural digitization strategy, the cultural resource elements can be put on the chain and go online, and the cultural data information can be imported into the model to achieve, for example, the exploration and tracing of the history of Sanxingdui.

AI large models can help improve the efficiency and quality of traditional cultural digital twins. However, at present, AI large models can only answer simple questions and are difficult to generate story sequences. Therefore, how to use AI to generate Chinese cultural themed image, audio, and text story sequences, how to explore and understand the inherent value of culture, and based on this, to carry out innovative creation and demonstrate advanced narrative skills, are still challenges.

In the future, if we imagine the application of generative AI in cultural heritage exhibition scenes, viewers only need to scan the QR code to interact with AI, and they can obtain virtual scenes and creative content generated by AI based on historical documents and research databases, providing viewers with immersive communication and unique interactive experiences. I think this is still feasible.

II. How to use AI large models to improve the quality and efficiency of cultural resource element management and application?

In 2022, China is promoting the implementation of the “National Cultural Digitization Strategy” and advancing cultural digitization. The collection of cultural resources is a prerequisite. Although Chinese technology companies have a massive amount of internet data, the data types are not comprehensive enough, and the standards for the online and on-chain of cultural data are not unified. AI large models can enhance the management and application efficiency of cultural resource elements through data replication, learning, and training.

Due to the enormous resource consumption, OpenAI will provide higher access priority, shorter response time, and higher versions of models to users through subscriptions, in order to release computing power resources as much as possible. Projects that consume substantial computing resources will also adjust their business strategies for charging. In order to quickly enter the race, large domestic enterprises are also competing to launch their own large models, which brings tremendous energy consumption and carbon emissions pressure.

Therefore, while embracing technology, it is also recommended to approach the development and use of large models rationally. Different large enterprises can share data and integrate resources to some extent. While steadily evolving, it is also necessary to avoid unnecessary consumption brought about by excessive competition. At the same time, the innovative development in the field of generative AI also means achieving breakthroughs in key technologies of software and hardware, which requires the support of industrial policies.

III. How to create AI industry large models with “Chinese characteristics” for application in the cultural field?

According to statistics, there are already more than 70 so-called AI large models in China. However, from the perspective of technological applications, many large models in China have not undergone revolutionary iterations, and it needs to be considered whether they can create or form uniqueness in the field of cultural sub-categories. ChatGPT is revolutionary and disruptive in many industries, especially in the cultural field, where it touches the core area. It may also bring some related risks, such as the weakness of AI large models in Chinese text comprehension compared to English, and the presence of deviations in the translation process. Therefore, it may cause cognitive biases for Chinese native users and difficulties and misunderstandings in understanding Chinese culture for non-Chinese native users.

In addition, the current AI models abroad involve implicit unfair factors such as race, gender, nationality, and political affiliation during coding and training, further leading to social bias and stereotypes, resulting in risks of extreme social emotions and ideological risks. If China wants to create AI models with unique Chinese characteristics, it needs to reverse these biases and provide value at least at these levels.

It is worth noting that from the perspective of data resources, English is still the global lingua franca, and there are corresponding English versions of data and information from various countries. The amount of training data for English models is much higher than that for Chinese. Although China has a large amount of data and user base for the development of artificial intelligence, the rich traditional culture has not been fully digitized, resulting in limited Chinese language data available for training large-scale AI models. For example, in order to train GPT, OpenAI hired Kenyan workers for data annotation and produced a large amount of training data. In contrast, China has relatively fewer data resources and higher costs, which to some extent affects the development of large-scale models.

In terms of technological differences, China is currently facing restrictions from the United States in terms of technical content in related industries, and the development and innovation of hardware required for training large models have encountered certain bottlenecks. From the perspective of the entire industry ecosystem, the AI industry ecosystem in the United States is more mature, covering the entire chain from data resources to algorithm innovation. China has just started to enter the field of large-scale models and needs to further strengthen cooperation and innovation upstream and downstream of the industrial chain.

4. Will the new content generated by AIGC encourage and inspire cultural innovation?

AIGC mainly generates new content by copying, learning, and training existing content, using algorithms and data analysis. It may face limitations in creativity and innovation. Therefore, AIGC mainly focuses on authentic restoration and innovative presentation to stimulate cultural innovation, providing new media and approaches for cultural inheritance.

From the perspective of cultural protection, after “learning” a large amount of traditional cultural corpus, AIGC can digitally restore the content of certain traditional cultural forms that are on the verge of extinction, providing better protection for these traditional cultural contents or forms. The content generated by AIGC can serve as a new medium and provide new ways for the inheritance of traditional culture, such as virtual cultural exhibitions, interactive games based on cultural themes, etc. These advantages can attract more young people to participate in cultural inheritance.

Within the scope of “fair use,” AI can inject some new forms of expression into traditional culture. For example, there are already models on the Internet that combine QR codes with ink paintings with cultural characteristics to generate scannable images. This not only enhances the practicality of QR codes but also injects new forms of expression into traditional culture. Similar ideas include using AI technology to digitize works with traditional cultural characteristics for the protection and preservation of valuable resources of traditional culture, and using AI translation technology to translate literary works into different languages to promote the international dissemination and communication of traditional culture, and so on.

The essence of the connection between cultural works and the audience is the intrinsic value empathy. AIGC cannot replace the uniqueness of human creativity, but it can serve as an excellent “assistant” by providing creators with more diverse ideas and entry points for cultural innovation through its expertise in “brainstorming” style content generation. In short, when it comes to AIGC and cultural innovation, we need to maintain a certain level of rational restraint to ensure that AI training does not weaken the sustainable driving force of artistic creation and does not hinder the long-standing civilization of humanity.

5. What paradigm changes or opportunities and challenges will AI large models bring to professionals in cultural and creative institutions, cultural content creators, cultural technology companies, etc. in the future?

In the aforementioned “Midjourney case,” AI painting has already begun to compete with human painters and illustrators, threatening their professional development. In the film and television industry, some post-production companies have already covered the entire process of AIGC technology application, driving AI participation in derivative product design and development, film and television promotion, etc. The animation and gaming industries and their practitioners are also actively exploring changes based on AIGC technology. In the long run, the trend of AI replacing specific positions in the cultural industry and even subverting the industry’s production ecology is unstoppable.

It should be emphasized that although the popularization of AIGC means a reduction in the threshold of content creation, it does not mean the subversion of content industrialization. AI generators are a kind of “catalyst” for knowledge, and the generation of specific content can be fully generated or dominated by AI. However, the generation of original and original professional content still relies on top creators with professional knowledge, technical experience, and industrial capabilities.

Large models can digitize and restore cultural heritage through technologies such as deep learning, enabling better inheritance and utilization of cultural heritage. For professionals in cultural and creative institutions, the development of large models will promote digital transformation and innovation in cultural creativity. The application of AI large models will also change the way traditional cultural institutions play their roles in collection, exhibition, and education, requiring practitioners to possess relevant technology and innovation capabilities.

AI large models can not only bring inspiration to cultural content creators, but their continuous development will also provide more creative tools and methods. While providing convenience, it also comes with new challenges because the use of large models may result in a lack of uniqueness and originality in creators’ works, requiring creators to maintain their individuality and innovative abilities during the creative process.

By using large models to develop intelligent cultural products and services, cultural technology companies can provide personalized cultural experiences and customized cultural products, improve user experience and the quality of cultural consumption, and bring more business opportunities and innovation space.

6. If AIGC’s creation is only “innovation in arrangement methods,” can it still enjoy copyright protection?

Whether the products generated by one-click generation of “orchestrated innovation” based on large models should enjoy copyright protection is still subject to debate. For example, in academic paper writing, prestigious institutions such as the journal “Science” have stated that they do not accept ChatGPT as authors of papers. AI is temporarily not widely recognized by the public as creators or co-creators. There is still no consensus on whether AIGC should be attributed to human creativity or machine creativity. Blockchain technology may be a better choice for rights protection and copyright protection, but currently there is no public chain in China. However, one thing is certain: whether the content is generated by AI can be detected through algorithms, similar to “using magic to defeat magic”.

7. How should the authenticity of artificial intelligence technology be regulated in the field of mass communication and culture?

This involves the application of artificial intelligence technology in image, video, audio processing, and synthesis, including image editing, face synthesis, and video synthesis, etc. After these technologies mature, they may generate some content that is indistinguishable from reality.

As for the regulation of authenticity, on the one hand, we may need to develop more advanced artificial intelligence technology to detect AI-generated images and videos, such as developing authenticity detection algorithms to label this content. On the other hand, we may also need the constraints of laws and policies to prohibit and combat the spread of false information.

In order to ensure that the application of artificial intelligence technology in the field of mass communication and culture is authentic and trustworthy, which is an important part of ensuring the benign development of society and cultural heritage, we need to strengthen data management and privacy protection to prevent data misuse and leakage. It is necessary to popularize and promote AI technology to the public, improve the public’s scientific and media literacy, and encourage people to treat AI technology and its applications in the cultural field more rationally. At the same time, we need to strengthen social supervision and public participation, promote the openness and transparency of AI technology, and avoid issues of interest transfer and information asymmetry.

(Acknowledgements for this interview: Bao He and Liu Shuxiu from the Communication University of China)