At a state-of-the-art laboratory, a researcher scribbles a chemical formula on the surface of a smart glass.

Instantly, with a digital command, the robotic "scientist" empowered by artificial intelligence (AI) launches into executing a sequence of tasks, including material retrieval, reagent addition, reaction initiation, monitoring, and testing, with the precision of a culinary expert preparing a dish.

This vivid scenario unfolds within the cutting-edge automated laboratory at XtalPi's Shanghai headquarters. Here, AI is revolutionizing the field of new drug discovery, infusing the process with unprecedented efficiency and innovation.

China is on the cusp of an industrial revolution powered increasingly by basic research, where AI technology is rapidly emerging as a key catalyst in advancing the frontiers of science.

In recent years, the prowess of machine learning and the emergence of generative AI technologies for tackling complex mathematical challenges have been combined to usher in a transformative approach to scientific inquiry: AI for Science (AI4Science).

This innovative paradigm is redefining the boundaries of research, enabling deeper exploration and discovery across various scientific disciplines.

Last year, the Ministry of Science and Technology, together with the National Natural Science Foundation of China, launched a project to harness the power of AI to advance the frontiers of scientific research.

The project is designed to deploy sophisticated AI models and algorithms to drive breakthroughs in genetic studies and stimulate significant progress in the field of biological breeding.

New models

Earlier this month, OpenLAM, a Chinese open-sourced large language model (LLM), introduced a suite of seven innovative model solutions tailored to address complex challenges in areas such as alloy development, pharmaceutical small molecules, solid-state battery technology, semiconductors and high-temperature superconductivity.

AI technology has also made significant strides in enhancing the accuracy and efficiency of weather forecasting and revolutionizing the way that people predict and prepare for meteorological events.

A Shanghai-based team has developed an AI weather forecast system named FengWu, based on a multi-model and multi-task deep learning architecture, capable of generating high-precision global meteorological forecast results for 10 days in just 30 seconds.

Compared to traditional physical models, FengWu reduces the rate of error by 19.4 percent and extends the forecast range to 10.75 days based on reanalysis data, according to the Shanghai Artificial Intelligence Laboratory, one of its developers.

"The FengWu model exemplifies the transformative potential of AI within the realms of Earth sciences and beyond," said Ouyang Wanli, the lab's lead scientist.

Pangu Weather, another cutting-edge model developed by Huawei, is also making waves in the field of meteorology with its precise weather forecasts.

Its predictive accuracy, ranging from a period of one hour to seven days, has surpassed that of several meteorological centers in Europe and the United States for these same periods.

According to the Central Meteorological Observatory of China, the model showed impressive performance in the trajectory prediction of Typhoon Mawar in June 2023 and was successfully used in forecasting Typhoon Doksuri last year.

Moreover, in the vast expanses of the universe beyond our planet, AI is proving its capabilities, contributing to our understanding and exploration of the cosmos.

In a pioneering collaboration with the National Astronomical Observatories, Tencent, a Chinese tech giant, is spearheading a star exploration initiative that utilizes AI to boost the efficiency of pulsar searches conducted by China's Five-hundred-meter Aperture Spherical Radio Telescope (FAST).

Remarkably, this AI-assisted approach can complete data processing work in three days that would otherwise have taken a year.

The initiative has already detected two fast radio bursts via the processing of celestial survey data, thereby contributing to studies that aim to deepen our understanding of the universe's origins.

New discovery

A research team in China's eastern coastal city of Yantai has successfully trained an artificial intelligence system for catalyst screening. From an extensive pool of more than 14,000 potential candidates, they have pinpointed four molecular formulas, all of which have yielded highly satisfactory outcomes.

"Using conventional approaches, a seasoned analyst could perform approximately 20 experiments annually. To explore such a vast array of candidates, it would have required the efforts of 20 analysts over 35 years," said Liao Zengtai, chairman of Wanhua Chemical.

"However, with the adoption of AI, this extensive endeavor was completed in a mere six months," Liao added.

The chemical supplier stands out as a pioneering enterprise that has mastered comprehensive manufacturing technology across the entire production chain of nylon 12, a polymer with excellent mechanical and corrosion-proof properties.

Another team from the Tianjin Institute of Industrial Biotechnology under the Chinese Academy of Sciences has leveraged AI algorithms to design a de novo type of enzyme called cytochrome P450, which is known as a "universal catalyst" in the field of industrial biotechnology.

The catalytic capability of this enzyme has seen a 1.3- to 3.5-fold enhancement, according to the researchers.

AI technologies are instrumental in the intelligent optimization of chemical synthesis pathways, thus enabling the automation of lab procedures and expediting the discovery of new materials through sophisticated high-throughput screening, said Yang Jinlong, a professor from the University of Science and Technology of China.

Zhang Linfeng, the president of the AI for Science Institute in Beijing, cited an example of the enigmatic growth of lithium dendrites, which can render lithium batteries inoperative and remains an unsolved puzzle.

This is largely due to the limitations of conventional techniques in capturing the rapid, millisecond-scale transformations involving hundreds of thousands of atoms, according to Zhang.

Zhang's team and scientists from Peking University have worked with CATL, China's largest automotive lithium-ion battery maker, to develop new battery materials.

"The use of AI, particularly large language models, offers a promising avenue to achieve both precision and efficiency in this challenging domain," Zhang explained.

Zhang is also the founder of DP Technology, a Beijing-based startup that utilizes AI for molecular modeling to offer industrial design across a spectrum of industries, including biopharmaceuticals, renewable energy, and chemical engineering.

DP Technology has been listed in CB Insight's AI 100 ranking in 2024. This exemplifies the significant shift of China's AI4Science from the confines of research labs to the broader market, emerging as a highly sought-after investment opportunity.

MindRank, a Hangzhou-based biotech firm, utilized its own AI drug discovery to develop a small-molecule medication. The AI tool assisted it in achieving an impressive feat, which saw the research team identify a preclinical candidate among nearly 100 molecules in just eight months.

The drug MDR-001 has now received clinical trial approvals in both China and the United States, targeting conditions such as obesity and type 2 diabetes.

AI4Science is set to exert a deep and lasting influence on the realm of basic sciences, with ripple effects extending to subsequent technological innovations and the field of smart manufacturing, according to a report by Crystal Stream Capital, a Beijing-based venture capital investor.

"AI4Science has enormous potential and will comprehensively transform the process from scientific research to industrial application," said E Weinan, a professor from Peking University.

"It represents the best opportunity in the history of China's sci-tech innovation," he added.