AI from China Benefits the World | A "Smart Key" Helps Unravel the Mysteries of Earth

GeoGPT effectively integrates multi-source heterogeneous data from geoscience research.

In July 2025, GeoGPT is showcased at the AI for Good Global Summit.

In early July 2025, the Palexpo in Geneva, Switzerland, was filled with crowds and lively discussions as it hosted the 6th AI for Good Global Summit.

At the summit, GeoGPT, a geoscience foundational model spearheaded by Zhejiang Lab, drew the attention of global experts and was honored with the Outstanding Innovative Practice Case Award 2025 issued by the International Telecommunication Union.

As a public technological good designed to serve earth science researchers worldwide, GeoGPT has woven itself into the global innovation network through open-source sharing and international collaboration, a vivid example of China's active engagement in "AI+" international cooperation.

AI+Research: Forging a New Paradigm for Geoscience Research

Over the course of the Earth's 4.6-billion-year history, countless secrets about the origins of life, biological evolution and climate change lie buried. Probing the deep interior of our planet and unraveling its mysteries is the mission and pursuit of geoscience researchers.

Yet, today's geoscience research is facing a set of challenges, including heterogeneous data from multiple sources, difficulties in effectively linking massive volumes of long-tail data, and obstacles to knowledge integration arising from disciplinary silos. AI might be the key that can break through these impasses.

GeoGPT was developed in response to the real-world demands of earth science research. As a public technological good developed specifically for geoscience research, GeoGPT aims to drive paradigm shifts in geoscience research. The model attracted the participation of over 400 geoscience experts across 25 institutions in China and abroad.

GeoGPT integrates Deep-time Digital Earth data with intelligent algorithms, offering a wide range of capabilities: document parsing and data extraction, domain-specific knowledge graph construction, personalized research assistance, geological map recognition and Q&A, and research idea generation.

In April 2025, GeoGPT officially opened to users worldwide. To date, it has served 55,000 geoscience researchers across 145 countries and regions.

The value of openness can only be validated in real research settings. GeoGPT's applications in geoscience research, spanning fields such as "dark data" governance and paleontological identification, have already demonstrated how AI technology can be transformed into shareable, reusable and collaborative scientific infrastructure. With the support and facilitation of the Zhejiang Deep-time Digital Earth International Research Center, GeoGPT has forged collaborative projects with leading teams worldwide.

In the field of paleontology, the GeoGPT team, in collaboration with a team led by Professor James Ogg of Purdue University, has for the first time enabled high-throughput automated extraction of structured data from nearly 30,000 pages of unstructured PDF files spanning the 50-volume "Treatise on Invertebrate Paleontology," a globally recognized authoritative work in paleontology. This would have taken decades of manual effort to transform into a globally accessible FAIR (Findable, Accessible, Interoperable and Reusable) knowledge base, providing an unprecedented data infrastructure for studying mass extinction events, evolutionary drivers and future environmental change.

Michael Stephenson, former executive chief scientist of the British Geological Survey and founding president of the Deep-time Digital Earth international big science program, has devoted himself to advancing the application of AI in earth science, particularly in paleontology. Building on GeoGPT, Stephenson developed an AI agent for paleontological classification that covers all genera in the "Genera File of Fossil Spores and Pollen," with more than 4,000 genera in total, far exceeding the 100 to 300 genera that an average palynologist could reasonably master. The tool helps palynologists identify and classify fossil spores and pollen, compressing identification time to less than 1% of what conventional methods require. It has since been deployed on the zero2x platform.

In April 2026, Professor James Ogg of Purdue University introduces the collaboration with the GeoGPT team.

A researcher from Zhejiang Lab conducts an offline GeoGPT training session for African partners in Kenya.

Shared Global Governance: Opening New Horizons for AI Governance Cooperation

For AI to function as an international public good that genuinely benefits humanity, simply being useful is not enough; it should also answer whether it is trustworthy, reliable and controllable, and how it can foster responsible innovation.

In late 2024, an independent, international GeoGPT Governance Committee was formally established to oversee ethical compliance throughout the model's development and to safeguard its integrity, data privacy, fairness, and scientific transparency.

The committee comprises seven international experts from the United States, the United Kingdom, France, Canada, Brazil and other countries. Under the supervision and guidance of the Governance Committee, GeoGPT has opened access to users worldwide and made its model weights and training data source list publicly available. Training has also been realized on multiple open-source foundation models, including DeepSeek, Qwen and the 021 Science Foundation Model.

At the committee's annual meeting in April 2026, members noted that GeoGPT's technical roadmap as a knowledge-and-reasoning engine had grown increasingly clear, with multimodal capabilities, map-interaction features and agent functions all showing strong potential. Meanwhile, scientific application cases were beginning to demonstrate real research value and bright prospects, and an early-stage "AI+Geoscience" system was taking shape.

This model of open capabilities and open rules elevates GeoGPT's international collaboration beyond mere tool promotion; it is an open science practice of collective development, joint oversight and shared benefits. It reflects the international community's trust in China's open science initiatives, while offering a replicable model for dismantling the technological barriers and data silos in the AI field.

Empowering Global Research, Sharing the Latest Fruits of AI Technology

After uploading a geological map, remote-sensing data and a scientific document, the model helped analyze the data, correlate information and make knowledge-based inferences. In September 2025, at a scientific workshop held in Kenya, Zhejiang Lab researcher Mohamed Jaward Bah was demonstrating GeoGPT to an audience of African geoscience professionals.

"We've been working with our African partners for more than two years now, and we hope to expand our training reach to over 20,000 people in the future," Mohamed Jaward Bah said.

Shadrach Sheriff, a geographic information system analyst from Nigeria, said after using GeoGPT: "Traditional geological map generation tools involve complex workflows and steep learning curves, and they typically demand systematic training and extensive experience that not every geoscientist possesses. AI models are transforming this domain through prompt-based geological map generation."

By building a secure, reliable open-source platform, GeoGPT lowers the barriers to technological innovation and application, enabling countries and regions with relatively limited research resources to access cutting-edge scientific infrastructure, join global research collaborations and share the fruits of AI development.

From serving global geoscience research and promoting shared governance by international experts to sharing technology and resources with the world, GeoGPT stands as another microcosm of how Chinese solutions are advancing open, accessible AI and engaging in global science and technology governance. (Ding Lei)