"AI for Science", favored by the Nobel Prize, is helping CATL develop batteries | Focus Analysis

Written by | Wang Fangyu

Edited by | Su Jianxun

The announcement of the 2024 Nobel Prize in Chemistry has excited a group of domestic AI entrepreneurs and practitioners.

"The Nobel Prize in Chemistry being awarded to AI for Science saves me a lot of time explaining to others what we do," Jia Haojun, the founder and CEO of Deep Principles, told 36Kr.

In the past two days, he has received more than 100 consultation calls from investment institutions. This AI for Science startup, which was established just over half a year ago, has encountered an opportunity for the industry to gain wider recognition.

On October 9 Beijing time, the Royal Swedish Academy of Sciences awarded the 2024 Nobel Prize in Chemistry to three scientists dedicated to the field of AI for Science. Among them are the CEO and senior scientist of Deepmind, an AI company under Google. The AlphaFold series developed by the two has made a landmark contribution to protein structure prediction and is regarded as The most representative achievement of AI for Science.

Image Source: The Official Website of the Nobel Prize

AI for Science (AI4S), that is, artificial intelligence-driven scientific research, is a new scientific research paradigm. The decision of the Nobel Prize in Chemistry largely affirms the long-term value of this research paradigm and even the entire field.

In fact, before receiving the support of the Nobel Prize, AI4S has already had specific practices in multiple fields such as life sciences, drug research and development, material research and development, energy, semiconductors, environmental science, and industrial simulation, which can help shorten the research and development cycle and reduce the research and development cost.

The largest downstream market of AI4S is currently in drug research and development, and its applications are very extensive. Global leading pharmaceutical companies including Pfizer, Johnson & Johnson, Merck, and AstraZeneca are all actively laying out. According to MedMarket Insights, the global AI pharmaceutical industry market size has reached 1.293 billion US dollars in 2023. Public information shows that there are already more than 100 AI pharmaceutical enterprises in China.

After achieving results in the field of drug research and development, Material research and development is becoming the next position that AI4S is going to capture.

In December 2023, CATL, the world's largest lithium battery leader, announced that it will establish an international research and development center in Hong Kong. Chairman Zeng Yuqun stated that this research and development center mainly focuses on AI4S to "explore new energy materials, systems, and application solutions".

This event is considered by industry insiders to have a benchmark significance, marking the recognition of the value of AI4S by leading enterprises in the energy materials field.

"The research and development of new materials using AI4S is still in an early stage, and the market size is still limited, but the general trend of this technological development is already relatively clear, and the future growth space is huge," Ding Zhebo, the managing partner of Woyen Capital, which focuses on new material investment, told 36Kr.

"In the near future, with the help of AI tools, combined with laboratory automation equipment and high-throughput screening platforms, it will Greatly improve the research and development efficiency of chemical new material practitioners, and AI and laboratory robots will become very important assistants for scientific researchers."

AI "Liberating" Scientists, From Medicine to Materials

Material research and development is becoming the next frontier of AI4S. A distinct signal is that many leading domestic AI pharmaceutical manufacturers are "Cross-border" entering the field of material research and development.

In mid-August, XtalPi, the first AI pharmaceutical stock listed in Hong Kong, announced a cooperation with GCL Group to provide customized services for the research and development of new energy materials for GCL. According to its semi-annual report disclosure, this cooperation lasts for 5 years with a total cooperation amount of approximately 1 billion yuan.

At the end of August, DeepModeling Technology, an AI4S startup that has received multiple rounds of financing, signed a contract with Dongyangguang Materials to establish an AI4S new material research and development joint laboratory. Prior to this, it had announced a strategic cooperation with CATL to jointly build a joint laboratory.

The AI pharmaceutical industry is highly competitive, and many AI pharmaceutical companies have started the painful period of layoffs and pipeline adjustments due to poor performance this year . Entering new application fields is a way for manufacturers to create new growth points.

Currently, the domestic materials industry is also facing many challenges, which can be summarized as "advanced basic materials are uneven, key strategic materials are subject to others, and frontier new material technologies need to be breakthroughs". And AI-assisted research and development can help shorten the time from material discovery to application and improve research and development efficiency.

Especially in the research and development of new materials with revolutionary significance in their respective industries, such as solid-state battery materials and photovoltaic perovskite materials, AI4S is considered by many experts as a breakthrough in research and development.

Taking all-solid-state batteries as an example, one of the difficulties in its research and development is to develop a stable electrochemical material system. Toyota in Japan has the largest number of patents in the field of solid-state batteries globally and has tried tens of thousands of electrolytes in batteries in the past 30 years, but has not yet successfully mass-produced them. However, AI4S is expected to help solve this problem.

Ouyang Minggao, an academician of the Chinese Academy of Sciences, also pointed out at a conference in January this year: "The core of the technical competition in the next decade of lithium batteries lies in materials, and artificial intelligence is changing the research and development paradigm of materials, which will greatly accelerate the research and development speed of all-solid-state batteries."

In addition to CATL and GCL Group, more and more leading chemical, energy, and material enterprises are actively embracing AI4S for the research and development of new materials.

Liao Zengtai, the chairman of Wanhua Chemical, a leading domestic chemical enterprise, said in a sharing in March this year that through AI technology, Wanhua Chemical can quickly screen out 156 from 14,000 schemes, and further optimize to obtain 4 effective schemes, greatly accelerating the research and development progress.

However, due to an earlier start, more applications and cases of AI-assisted material research and development are still overseas.

Including Internet giants such as Microsoft and Google, in order to sell cloud services in combination, they have long started to develop industry models and PaaS platforms of AI4S. The downstream application parties are mainly large chemical and material giant enterprises.

Ding Zhebo told 36Kr that the use of AI by the German veteran material enterprise Merck to develop new OLED luminescent materials and the use of AI tools by the chemical giant Dow to optimize the development of non-metallocene catalysts for polyolefins are relatively successful typical applications.

Where there is demand, there is supply. 36Kr has learned that as a third-party service provider, in addition to listed companies represented by XtalPi, startups represented by DeepModeling Technology and Deep Principles, The AI Labs of Internet companies such as ByteDance, Tencent, and Baidu are all exploring or laying out AI-assisted material research and development.

High Difficulty in Material Discovery, AI4S Has a Long Way to Go

Both are application fields of AI4S, but the development of AI material research and development is about 10 years later than that of AI pharmaceuticals. The former has developed into a market with a scale of over tens of billions, while the latter's current scale is almost negligible.

In Jia Haojun's view, although the underlying principles of AI material research and development and pharmaceuticals are similar, the technical applications are quite different, which is closely related to the characteristics of the materials industry.

"AI4S studies scientific problems and must also abide by the laws of physics. In the field of AI pharmaceuticals, the biological properties of protein macromolecules determine that the number of their combinations is relatively limited, and more research focuses on exploring the spatial folding structure. However, the elemental composition and coordination structure of chemical materials are more variable, and the number of candidate spaces for permutations and combinations is higher, Resulting in greater difficulty in early discovery and optimization."

The lack of high-quality data is also one of the problems that AI material research and development needs to face.

"The biggest problem facing AI material research and development is still high-quality data," Ding Zhebo said. A large amount of raw data in the biomedical field is publicly disclosed through literature reports and is relatively easy to obtain and can be used for model training. However, the new chemical material industry is relatively closed, and the raw data involving materials and catalysts are often the trade secrets of enterprises and are not open to the outside world.

In this regard, Zeng Yuqun also frankly admitted in an interview that AI4S (used for battery material research and development) currently does not have particularly good models, structures, and algorithms, and there is still a long way to go.

AI research and development also faces limitations of objective conditions in actual implementation, especially the shortage of talents.

AI research and development of materials is an interdisciplinary field between materials science and computer science, requiring proficiency in both aspects. However, domestic chemistry talents generally do not possess programming skills, and enterprises lack cross-disciplinary talents who understand both materials and computers.

Sun Weijie, the CEO of DeepModeling Technology, mentioned in an interview in May this year that the company has a total of nearly 300 employees, and more than 80 are interns. Because "AI4S requires a strong interdisciplinary ability, We simply cannot recruit from the market."

Forward-looking technologies are often not mature enough, with many bottlenecks, and require more time and cost for exploration.

However, from the active layout of leading energy and material giants such as BASF, Dow, CATL, and Wanhua Chemical to the favor of the Nobel Prize in Chemistry, the long-term value of this field has undoubtedly been affirmed.

"Organic chemists often use a method called retrosynthetic analysis to find the synthetic route of the target molecule. This method won the Nobel Prize in the 1990s. But I believe that in another 20 years, almost all retrosynthetic analyses can be completed by AI, including the recommendation of reaction routes, and even the recommendation and optimization of reactants and reaction conditions. AI-driven new material research and development will revolutionize the entire chemical new material industry." Ding Zhebo said.