AI Meets Ancient Chinese Civilizations: Fudan, Shanghai Zhi Yuan, and Shanghai Chuangzhi College Launch Multimodal Large Model for Early Chinese Culture}

Editor | Xin Yue

In recent years, AI has demonstrated its powerful ability to analyze massive data, revealing underlying scientific laws and gradually transforming research paradigms. We seem to be witnessing AI-driven breakthroughs in scientific discovery after discovery.

If we look into history, what sparks might collide when AI encounters ancient Chinese civilizations, leveraging its pattern-recognition prowess?

On July 26, at the 2025 World Artificial Intelligence Conference (WAIC 2025), the world's first AI model focused on early Chinese civilization was launched during the Galaxy Enlightenment Science and Intelligence Open Cooperation Forum.

In response to President Xi Jinping’s call to "accelerate the building of a culturally strong nation," Fudan University, Shanghai Institute of Science and Intelligence (referred to as Shanghai Zhi Yuan), and Shanghai Chuangzhi College jointly developed the Early Chinese Civilization Multimodal Model.

This model covers disciplines such as archaeology, cultural relics, ancient Chinese history, historical geography, historical literature, classical Chinese philology, Chinese language and literature, and minority languages and literature. It integrates rare historical materials like oracle bones, bronze inscriptions, local chronicles, and maps, forming a vast knowledge system of early Chinese civilization. Its goal is to pioneer intelligent pathways for research into early Chinese history and build a digital foundation for cultural inheritance and innovation.

Wu Libo, Assistant President of Fudan University, Chairman of Shanghai Zhi Yuan, and Vice President of Shanghai Chuangzhi College, explained: "As a project focused on the origins of early Chinese civilization, its core mission is to answer two questions: Where does our Chinese civilization come from? How has it evolved? Finding answers to these questions is crucial for telling China's story well, inheriting Chinese culture, and strengthening cultural confidence today."

Collision of 'Technological Rationality' and 'Humanistic Spirit'

Compared to natural sciences, humanities and social sciences deal with complex social and cultural phenomena, requiring consideration of multi-dimensional and multi-layered data. AI not only processes vast amounts of data efficiently but also excels at recognizing hidden patterns within that data.

This makes deep collaboration between AI and humanities and social sciences both possible and inevitable.

The early Chinese civilization multimodal model prioritizes establishing a solid academic foundation for its focus on early Chinese history.

Fudan University boasts strong expertise and deep historical roots in humanities and social sciences, especially in history, archaeology, and Chinese language and literature. The university has long engaged in excavated texts and ancient script research, aiming to uncover historical ideas and cultural values.

The model is built upon Fudan’s authoritative, systematic, and cutting-edge knowledge base, enabling its application in professional research scenarios.

With domain experts involved in data construction and knowledge review, AI scientists designed algorithms and technical solutions, while engineering teams developed the system and platform—multi-disciplinary collaboration and complementary strengths.

This is a collision of "technological rationality" and "humanistic spirit." As Qu Yuan, a distinguished professor at Fudan and director of Shanghai Zhi Yuan, said: "This cross-disciplinary combination is the most vital."

Building a Powerful Intelligent Engine for Scholars

The early Chinese civilization multimodal model centers on interdisciplinary integration and cutting-edge technology, deeply analyzing the origins and evolution of Chinese civilization. Zhu Siyu, researcher at Fudan’s AI Innovation and Industry Research Institute and AI scientist at Shanghai Zhi Yuan, stated: "We aim to assist archaeology and classical texts verification through data and models, achieving multi-modal data fusion to better serve national projects like the Chinese civilization source exploration."

The model integrates resources from archaeology, ancient texts, historical geography, linguistics, and genetics, covering topics such as human origins, agricultural origins, and the development of Chinese ethnicity.

Early Chinese civilization multimodal database

Data is the foundation of large models. The scale of data determines potential, and quality determines performance.

Fudan University provides high-quality humanities and social sciences corpora, integrating authoritative heterogeneous data sources such as archaeological artifacts, historical documents, ancient scripts, geographic information, and genetic data, achieving cross-modal, cross-disciplinary, and cross-temporal alignment and correlation.

To solidify the data foundation, the project recruited graduate students and established a professional annotation and data construction team, ensuring academic rigor and consistency in data collection, cleaning, and annotation. Under expert guidance, they built a robust multimodal database supporting the training of the early Chinese civilization multimodal model.

Evaluation Set for Early Chinese Civilization

To comprehensively and objectively evaluate the model’s capabilities, the team developed a three-dimensional layered evaluation framework based on "discipline × question difficulty × research scenario," ensuring broad coverage and scientific design.

In the discipline dimension, the evaluation covers key humanities and social sciences fields related to early Chinese civilization, ensuring professional and broad question coverage.

In difficulty, the framework distinguishes basic factual recall, evidence integration, and critical analysis of scholarly debates, covering the full spectrum from basic cognition to advanced reasoning.

In scenario, the evaluation aligns with real research needs, supporting tasks like efficient retrieval, precise translation, unstructured text/image reading, fact verification, multimodal feature extraction, and deep reasoning.

The system contains over 10,000 high-quality questions, ensuring a comprehensive and precise assessment of the model’s ability to meet complex humanistic research needs.

Multimodal Generative/Understanding Model for Early Chinese Civilization

Based on the database and evaluation system, the team developed a multimodal generative/understanding large model for early Chinese civilization, integrating massive multimodal data.

The model adopts a humanistic research-oriented multimodal architecture, breaking disciplinary boundaries and enabling complex knowledge network construction, pattern discovery, and phenomena explanation, significantly enhancing interpretative power.

It uses archaeological, ancient texts, inscriptions, and geographic data through modules like "Chinese Early Civilization Multimodal Spatiotemporal Data Alignment," "Multimodal Completion and Generation," and "Causal Logic Inference of Origin, Formation, and Development," forming a cognition engine for early Chinese civilization research.

Early Chinese Civilization AI Agent Platform

Finally, the team built an AI agent platform for early Chinese civilization, integrating humanities research methods and knowledge production mechanisms. The platform supports multi-step reasoning and task planning, ensuring all outputs are based on reliable sources and complete evidence, with traceability and verifiability, supporting professional humanities research and applications.

For example, inputting an image of a prehistoric pottery shard from the Baodun site in Xinjin County, Sichuan, can retrieve detailed features and perform related analysis with artifacts from the Sanpo site.

It is noteworthy that the model’s responses are highly professional and precise, enabling the platform to serve academic research, cultural dissemination, and education sectors effectively.

In summary, the team has built a comprehensive support system from data to intelligent applications, covering the entire research chain of early Chinese civilization, aiming to promote research efficiency and paradigm shifts through AI technology.

Towards AI4SSH

The multimodal large model for early Chinese civilization is a significant practice of the "AI for Social Sciences and Humanities (AI4SSH)" concept, requiring deep integration of humanistic insights and engineering technology.

AI4SSH, proposed by Fudan University, emphasizes data-driven and mechanism-driven research. In March, the first comprehensive report on AI and humanities—The Future is Here — Blue Paper on AI and Humanities Development—was published, led by Fudan’s National Development and Intelligent Governance Laboratory in collaboration with Shanghai Zhi Yuan and Deloitte China. It states that AI is driving humanities and social sciences into a "dual-driven" research paradigm.

Wu Libo, the chief editor, hopes the report marks a new beginning: "We want to refine the humanities, which is a key task in the paradigm shift."

The early Chinese civilization multimodal model is a turning point in this paradigm shift, driven by AI4SSH. The team uses cutting-edge AI to create a powerful intelligent engine that makes Chinese civilization research more accessible, efficient, and insightful.

There are many uncertainties in exploring the origins of Chinese civilization, such as unrecognized oracle bones. Wu Libo notes: "Past research lacked deep temporal and spatial alignment, leaving gaps in the full story of Chinese civilization. The multimodal model aims to break disciplinary barriers, connect knowledge across fields, and form a complete narrative of Chinese civilization’s origins."

In the long term, this model will provide forward-looking, systematic support for the inheritance and promotion of Chinese civilization and represent a significant step in AI-driven transformation of humanistic research paradigms. We look forward to more brilliant achievements in humanities and social sciences driven by AI4SSH and breakthroughs through interdisciplinary integration."