Top scientists debate in Shanghai: AI empowers scientific research, but beware of "the dangers of using AI"

AI empowering scientific research is the direction of paradigm change in research that the scientific community is most focused on today. On October 30, at the 2026 World Laureates Forum held in the Lingang New Area of Shanghai, this paradigm change became a hot topic of discussion among scientists from various countries. In the field of materials science, Nobel laureate in physics Konstantin Novoselov said that AI has brought a revolutionary impact to the design of new materials, and can predict what new functions will result from atomic-level structural modifications. Robot laboratories commanded by AI will also become capable assistants for materials scientists. In the field of integrated circuits, Chen Yunji, co-founder of Cambricon, said: "We are now using chips to accelerate AI, and at the same time using AI to design chips. Further ahead, these two things will form a loop—AI designs better chips, and better chips in turn train smarter AI." On how to properly handle the relationship between humans and AI, Fields Medalist Laurent Lafforgue offered this advice: "AI cannot save anyone the effort required for learning. One 'danger' of using AI to learn is that the knowledge acquired always remains on the surface. To integrate knowledge into your mind, you must put in effort and patience. Some things, even if boring, must be done yourself." Using AI to modify the structure of two-dimensional materials In 2010, Novoselov shared the Nobel Prize in Physics with his advisor Andre Geim for successfully exfoliating single-atom-layer graphene. With the rise of large models, he began applying AI to the field of functional intelligent materials. At the World Laureates Forum · Future Energy Conference, he showed the audience a diagram of "machine learning for single-atom catalysis," in which individual atoms formed various shapes of two-dimensional material structures. "We can use AI to modify the atomic structure of two-dimensional materials and predict what new functions such modifications will give the materials," Novoselov said. At the same time, he also pointed out the limitations of "AI designers"—real-world materials are more complex than materials designed by AI. "For example, the idealized materials such as pure iron and pure silicon used by AI in predictions do not exist in the real world." Novoselov speaks at the 2026 World Laureates Forum. To ensure the practical accessibility of new material development, Novoselov proposed incorporating "material synthesis" into the research and innovation chain from the very beginning, connecting the "material synthesis—material performance" pathway, observing intermediate products in the synthesis process through experiments, and then using theoretical methods to model these chemical reactions. "Then, I hope you will feed all these data to AI, so that AI can predict: which type of synthesis parameters should be selected to obtain the material with specific properties." These remarks by the Nobel laureate expressed what he has learned in recent years from leading his team in exploring AI-empowered materials research. "Welcome to come find me after the session," he told the Chinese scientists attending the forum. It is understood that Novoselov is a foreign academician of the Chinese Academy of Sciences and a recipient of the Chinese Government Friendship Award, and his cooperation with China's science and technology community has lasted more than ten years. He has worked with Chinese universities, research institutes, and industrial platform teams to tackle key problems in directions such as two-dimensional materials and quantum devices, achieving many results. Now, in the burgeoning field of AI-empowered materials research, he is expected to find more Chinese partners. Large amounts of capital pour into the AI materials track Can AI replace scientists? Jeffrey Snyder, a professor at Northwestern University in the United States and a top scholar in the field of thermoelectric materials, gave a negative answer in his speech. In his view, AI is like a doctoral student who has just enrolled and currently cannot replace experts in many fields. How fast the "AI doctoral student" grows depends to a large extent on the quality of the data input by the research team. "If you input garbage, AI will also output garbage." Although it cannot replace scientists, AI has greatly improved R&D efficiency. "In the past, developing a new alloy took about 20 years; now it has been compressed to 6 months to 3 years," Snyder said. In the new materials design stage, AI inverse design is gradually becoming mainstream. This method means that the R&D team first determines the desired performance indicators, then works backward to search for and generate material structures and compositions that meet those performance requirements, and selects the material production process. Snyder introduced that the material used in Apple smartwatches is the product of inverse design. "The watch surface looks like stainless steel or aluminum alloy, but it is actually a special alloy inversely designed by Apple. Its performance indicators such as strength, corrosion resistance, reliability, ease of processing, and polishability are all determined by specific parameters." Today, with the rapid development of large models and agents, AI's ability to design new materials is becoming stronger and stronger. This American scientist showed a financing list of technology companies, all of which are startups on this frontier track: CuspAI raised $570 million, Periodic Labs raised $300 million, Citrine Informatics raised $81 million... The massive influx of capital is driving these companies to grow rapidly and become a transformative force in materials science and its industries. The World Laureates Forum is held at the Lingang Center. In addition to large models and agents, robot laboratories also play an important role on this frontier track. Snyder said that the foundation of such laboratories is materials informatics, and their biggest feature is autonomy, that is, they can autonomously carry out materials science experiments under the command of AI. Promoted by Novoselov, the National University of Singapore and the University of Toronto jointly built the "Materials Data Foundry."It is a robotics laboratory that enables automated robotic platforms to work in concert with large models—the robots automatically synthesize and characterize new material formulations predicted by the large model, and automatically send experimental data back; the large model iteratively optimizes based on the experimental results and completes the next round of design, forming a closed loop of "AI proposes a plan—robots conduct experiments—experimental data feeds back into AI." The Nobel laureate predicts that future materials laboratories will be like unmanned factories, operating autonomously 24 hours a day without interruption, and together with "AI designers" will become capable assistants to materials scientists. AI designs mass-produced chips for large models In the field of integrated circuits, "AI designers" have also emerged. Speaking at the World Laureates Forum · Artificial Intelligence Parallel Forum, Chen Yunji said that getting AI to design chips is not easy, for two reasons: first, there are too many options—to design a computer CPU, the number of candidate solutions is unimaginably large; second, not a single mistake is allowed—a chip must perform billions of operations per second, and its accuracy must be close to 100%. In response, Chen Yunji's team's approach is not to make the AI "get it right in one go," but to have it automatically check and repair after drawing, revising hundreds of times per second. Chen Yunji speaks at the 2026 World Laureates Forum. Following this approach, in 2021 they produced the first AI-designed chip, "Qimeng 1," whose performance is roughly equivalent to an Intel 486 from the 1990s. By 2025, "Qimeng 2" was released, with performance equivalent to the mainstream mobile phone CPUs of more than a decade ago. In other words, in just four years, AI's chip-designing ability has gone from "it runs" to "roughly good enough." As early as 2008, Chen Yunji's team began pondering one question: could a chip be made specifically for AI? At the time, the mainstream view was that this was unreliable—AI technology was changing rapidly, and by the time the chip was made, the algorithms might already be outdated. After analyzing more than a hundred AI algorithms, they found that no matter how the algorithms changed, the "basic actions" required at the underlying level were actually just a few. On this basis, they developed an instruction set specifically for artificial intelligence. Subsequently, Huawei's first-generation AI chip adopted the "chip brain" blueprint they designed, paying about 200 million yuan in licensing fees. In 2020, the project "A New Paradigm for Deep Learning Processor Architecture," completed under Chen Yunji's leadership, won the second prize of the National Natural Science Award—the first and so far only National Natural Science Award in China's processor chip field. Chen Yunji said: "We are now using chips to accelerate AI, while also using AI to design chips. Further ahead, these two things will form a cycle—AI designs better chips, and better chips in turn train smarter AI." It can be foreseen that this cycle will accelerate chip iteration in the future and drive large models to move faster toward the "ultimate goal" of artificial general intelligence.