Artificial Intelligence for Science (AI4S) is rapidly becoming a strategic priority and competitive focus for the world's leading scientific nations. Some scholars believe that AI could produce the equivalent of 10,000 Einsteins within this century, while others argue that AI4S is merely an auxiliary tool that can never possess original creative capability. How exactly will AI transform human scientific research? At the 2026 Science Intelligence Conference, a Global Times reporter posed this question to experts from agriculture, life sciences, artificial intelligence, and earth science.
In just a few months of work, what used to take months can now be completed in the time it takes to drink a coffee. Previously, when scientists faced a research problem, they typically needed to review extensive literature, design experiments, run repeated trial-and-error cycles, and then extract patterns from limited experimental results. Today, this process is being fundamentally rewritten by AI. Multiple scientists at the conference told the Global Times that AI's most immediate contribution is a dramatic leap in efficiency.
A research team led by Song Yang, associate researcher at the Chinese Academy of Geological Sciences, collaborated internationally with the Commission for the Geological Map of the World (CGMW) and the Deep-time Digital Earth International Big Science Program (DDE) to complete the integration and compilation of a 1:5 million scale global geological map in 2024 — the first time such a map and its accompanying database were unified at this scale. Traditional geological maps are primarily supervised and completed by national geological survey bureaus, and differing national standards make data coordination extremely difficult. The emergence of AI has provided the perfect tool to establish standardized workflows. Today, leveraging AI's semantic alignment and vectorization capabilities, what used to require months of work can now be completed in the time it takes to drink a cup of coffee. The efficiency of quality inspection data updates for geological maps has also greatly improved, with the update cycle shortened to just a few months. Song Yang told the Global Times that the team is now updating the map scale from 1:5 million to 1:1 million and achieving multi-resolution presentation, with data volume growing geometrically. Without AI, we would never have dared to propose this goal.
Helping to cultivate the 'Perfect Rice'. Behind the efficiency gains are a new generation of AI models and experimental platforms rapidly becoming the new infrastructure of scientific research, reshaping how Chinese scientists conduct research.
In the view of academician Zhong Kang of the Chinese Academy of Sciences, AI is transforming plant breeding from finding a needle in a haystack to designing on demand. Like assembling a computer, you install the best graphics card you want, the best sound card you want — each function implemented by a module. Breeding works the same way: one gene controls grain weight, another controls disease resistance, another controls low-temperature tolerance. With AI, scientists can combine the best genetic modules to create perfect varieties that are simultaneously high-yield, high-quality, pest-resistant, and stress-tolerant.
Zhong Kang told the Global Times that in the future, with the help of AI, cultivating a perfect rice variety is within reach. I estimate that in the next 10 years, we will be able to achieve near-perfection in rice breeding.
The ability to have 'aha moments' remains scarce. Today, countries are accelerating their AI4S deployment. China has released opinions on implementing the 'AI+' action, listing AI-enabled science and technology as one of six key initiatives; the United States launched the Genesis Project to integrate national research infrastructure and leverage AI to accelerate frontier scientific breakthroughs; the EU has released a dedicated strategy to pool resources for AI R&D and its application in science.
However, the academic community remains sharply divided on the prospects for AI4S development. Some scholars believe AI will produce 10,000 Einsteins within this century, while others insist AI4S is merely an auxiliary tool that can never achieve original innovation.
Li Guojie, academician at the Chinese Academy of Engineering and computer science expert, noted that the fields most likely to achieve major breakthroughs in AI4S over the next five years include drug discovery, materials science, engineering design, earth science, and mathematics. By 2030, a significant proportion of new drug R&D projects will use AI-assisted tools, but the proportion of approved new drugs discovered and designed primarily by AI may still be limited. By around 2035, this proportion could rise to over 20%.
Nevertheless, AI still has notable weaknesses in assisting scientific research. Many experts interviewed told the Global Times that AI excels at summarizing and finding patterns but struggles to propose new scientific questions — and the latter is often the key to scientific exploration.
Xu Xun, chief researcher at BGI Group, shared a five-level framework for AI capability in life sciences: L1 is knowledge synthesis; L2 is understanding scientific data and improving bioinformatics processing efficiency; L3 can independently complete scientific discovery — humans propose hypotheses, AI performs reasoning and closed-loop verification — equivalent to doctoral student level; L4 not only analyzes but independently produces data, requiring a fully automated laboratory to support it; L5, the highest level, involves AI controlling the theoretical framework of an entire scientific branch — the gatekeeper of that field. Xu Xun said current AI is at the L3 level: it can complete discoveries but cannot independently pose original questions, and asking questions remains the irreplaceable core value of humans.
Professor Tang Qiuhong of the Department of Earth System Science at Tsinghua University noted that AI's mastery of existing hydrological knowledge is roughly equivalent to a master's to first-year doctoral student level, but in proposing truly frontier and detailed scientific questions, it generally cannot match trained master's or doctoral students. He also mentioned that AI's most obvious current weakness is its lack of understanding of the physical principles of the real world, meaning the correctness of AI reasoning cannot be guaranteed, and scientists need to spend more effort verifying it.
In Chen Yuntian's view, AI excels at creating unprecedented new combinations from existing human information, but that kind of out-of-the-box, eureka-style original breakthrough is still something AI lacks.
Song Yang expressed concern about excessive reliance on AI in scientific research. When students encounter problems, they are increasingly accustomed to throwing them to the model first. This, while helpful, is actually preventing students from truly growing into experts.
Professor Qin Tao, vice president of Beijing Zhongguancun College, told the Global Times that AI4S is no longer just a single-point exploration within universities or research institutes — it is becoming an important arena for international technological competition. He observed that in the past, research, education, and industry tended to follow a relatively serial process: students first learned knowledge, then did research with mentors, and after graduation entered industry. But AI technology is accelerating this process, making scientific research increasingly directly address real industrial problems.
