The 2024 Nobel Prizes in Chemistry and Physics were awarded to AI-related research projects, a landmark event that not only highlights the maturity of AI technology but also foretells its revolutionary reshaping of the established paradigms in scientific research. In the fields of chemical and materials development, many companies have keenly recognized this wave of technological change and are actively engaging in it, either through independent exploration or collaboration with tech companies, leveraging AI technology to empower scientific research and development, aiming to propel the industry from traditional models towards a more intelligent direction characterized by "predictive design" and "precise creation."

Application of artificial intelligence technology in the field of chemical materials research and development

At present, AI technology has shown great application potential in the field of chemical material research and development, and many companies have achieved significant results.

BASF introduces the high-performance supercomputer Quriosity, integrating AI into the molecular and compound simulation calculation process. This move significantly enhances computational efficiency, enabling rapid screening of polymer structures and accelerating the development of new molecules and compounds. Calculations that once took a year to complete can now be finished in just a few days, and it can also uncover potential correlations that are difficult to detect with traditional methods, opening up new avenues for research and development.

Dow Chemical and Microsoft have reached a cooperation, deeply integrating Azure AI and machine learning technologies into the research and development of materials such as polyurethane. The AI model they have built is like an intelligent brain, capable of analyzing and screening millions of formula combinations within just a few seconds, and providing highly targeted optimization suggestions. Laboratory exploration work that originally took 4 to 6 months can now be completed in just 30 seconds, with efficiency improved by about 200,000 times. This significantly shortens the time to market for new material differentiation solutions, allowing the company to gain a competitive edge in the market.

Wanhua Chemical has achieved a significant breakthrough in the catalyst screening process with the help of AI technology. Facing more than 14,000 candidate options, the AI algorithm quickly screened out 156 potential choices, which were then further optimized to 4, precisely recommending molecular synthesis experiments, greatly shortening the R&D cycle and achieving a qualitative leap in research efficiency.

CATL, on the other hand, has taken a different approach by integrating material mechanisms, big data analysis, and AI algorithms to accelerate the development of battery materials such as electrolytes, cathodes, and coatings. Through this innovative model, not only has the R&D cycle been shortened by 30%, but R&D costs have also been reduced by 30%, effectively improving product performance while enhancing the company's economic benefits.

XtalPi leverages quantum physics simulation, AI algorithms, and cloud computing technology to build an intelligent drug discovery platform, shining brightly in the field of drug development. In the early stages of a project, this platform can generate millions of virtual molecules and quickly screen out key candidate molecules. During the collaboration with Pfizer on the development of the oral COVID-19 drug PAXLOVID, the combination of AI prediction algorithms and experimental validation successfully identified the superior drug crystal form in just 6 weeks, a process that would traditionally take several months or more, fully demonstrating the efficiency of AI technology in the field of drug development.

From these cases, it is not difficult to see that the deep integration of AI technology and multidisciplinary knowledge can quickly screen out feasible options from a vast number of solutions and further optimize them, significantly shortening the experimental and R&D cycle and improving R&D efficiency. At the same time, it can more accurately achieve material design, performance prediction, and process optimization, providing scientific and reliable basis for R&D decision-making, accelerating the discovery and application of new materials, and injecting strong momentum into the innovative development of chemical materials.

challenges faced by artificial intelligence technology in the field of chemical material research and development

Despite the broad prospects of AI technology in the field of chemical material R&D, it still faces many severe challenges in practical application.

01 | At the data level, it mainly faces challenges such as data scarcity, heterogeneity, and quality defects

Currently, a large amount of valuable and effective data is dispersed and stored within enterprises, mostly in unstructured form, making the circulation and integration of data very difficult. Especially in the research and development of new materials, due to the lack of historical data as support, AI technology faces a serious "cold start" problem. In addition, relying solely on experiments to obtain data is not only time-consuming and labor-intensive, but also extremely costly, severely restricting the widespread collection and application of data.

In the process of new material design, it is necessary to integrate various types of data, such as molecular structure [simplified molecular input line entry specification (SMILES)], spectral data (infrared, Raman), and process parameters (temperature, pressure), etc. At the same time, it is also required to achieve cross-scale data correlation from micro-level data to macroscopic properties, which poses extremely high demands on data processing techniques and integration methods. However, the lack of uniform standards among different institutions in aspects such as material composition labeling (e.g., mixed use of mass fraction and molar fraction) and experimental condition recording leads to low efficiency in data fusion and model training.

During the experimental process, various errors inevitably occur, such as equipment measurement errors and batch errors, which seriously affect the accuracy of the data. In addition, different laboratories often have significant differences in the definition of the same phenomenon, making it difficult to ensure consistency in data labeling, greatly reducing the reliability and usability of the data. Data quality is also reflected in the imbalance of the data; in chemical material R&D data, the proportion of data for some high-performance or special materials is extremely small, making it difficult for models to fully learn from the minority class data during training, thus affecting the prediction and analysis of rare but important material characteristics.

02 | At the algorithm and model level, challenges such as the contradiction in model interpretability, the spatiotemporal gap in multi-scale modeling, and the bottleneck of few-shot learning are faced.

Deep neural networks, although capable of achieving high accuracy in predicting material properties, have a very low degree of interpretability regarding their internal physical mechanisms, forming a typical "black box model dilemma." The reason for this is that most existing AI models are primarily data-driven and lack effective embedding of fundamental physical laws such as the conservation of mass and thermodynamic laws, leading to prediction results that may contradict scientific common sense. Therefore, how to ensure model complexity while improving its physical interpretability has become a key challenge that urgently needs to be addressed.

Material research and development need to span a time dimension of 12 orders of magnitude, from femtosecond-level molecular dynamics to year-scale aging experiments, while also linking quantum computing with reactor-level spatial features. Although there are some model frameworks that attempt to bridge this gap through multi-scale theoretical modeling, they are still constrained by computational resources and algorithmic efficiency in practical applications.

In the new material R&D scenario, the available data volume is often very limited, usually less than 100 sample quantities, which makes the generalization error of traditional models larger. For unverified systems, the prediction failure rate of zero-shot exploration is even higher. Although techniques such as transfer learning provide some ideas for solving this problem, data noise and domain differences still significantly affect the transfer effect of the model.

03 | At the talent level, the insufficient integration of interdisciplinary knowledge, the imperfect talent cultivation system, and the issues of talent attraction and retention cannot be ignored.

Chemical material research and development involves multidisciplinary knowledge such as chemistry and physics, while AI technology requires professional knowledge in fields like computer science, mathematics, and statistics. There is a significant difference between these two knowledge systems, leading to an extreme scarcity of versatile talents who are proficient in both chemical materials and AI technology. Furthermore, there is a clear knowledge barrier between AI algorithm experts and chemical material domain experts, making communication and collaboration difficult, which also hinders the deep integration of algorithm models with chemical material research and development.

Currently, many talents engaged in AI technology lack practical project experience in chemical material R&D, and their understanding of the R&D process, requirements, and pain points is not deep enough. At the same time, practical platforms and projects in the field of AI R&D for chemical materials have just begun, and talents have not yet accumulated enough experience or enhanced their capabilities through practice, which to some extent restricts the application and development of AI technology in this field.

The competition for high-end talents in the AI field is extremely fierce. Compared to popular industries such as the internet and finance, the chemical materials industry faces greater pressure in attracting and retaining talent due to relatively harsh R&D environments and lower compensation levels, resulting in a more severe phenomenon of talent loss.

Thoughts and suggestions on accelerating chemical materials R&D using AI technology

Accelerate the application of AI technology in the R&D of chemical materials, which can be addressed from three key aspects: data, algorithms and models, and talent.

01 | At the data level, strengthen data integration and sharing, establish and improve data standardization, enhance data quality, and mine data to achieve value-added

A unified data management platform should be established within the enterprise to effectively integrate the scattered data resources of various business departments and subsidiaries, breaking down data silos and achieving centralized storage and sharing of data, allowing for the free flow of data within the enterprise. At the same time, actively cooperate with external research institutions and universities to establish a data sharing mechanism, widely obtaining more dimensional external data, enriching data sources, and providing sufficient data support for AI model training.

Develop unified data standards and specifications covering aspects such as material composition labeling and experimental condition recording to ensure consistency and comparability of data from different sources, facilitating subsequent data integration and model training, and enhancing the usability and value of the data.

Build a comprehensive data quality evaluation system to strictly assess and monitor the accuracy, completeness, and consistency of data. Strengthen data cleaning and preprocessing to remove noise and erroneous data. At the same time, optimize experimental design and operational processes to reduce errors at the source and ensure data quality.

Make full use of data mining technology to extract potential patterns and knowledge from massive historical data, providing valuable references for the development of new materials. By predicting new market demands and R&D directions through data analysis, it offers strong support for corporate strategic decision-making, achieving the maximization of data value.

02 | At the algorithm and model level, enhance model explainability, multi-scale optimization modeling, and break through small sample learning technology

Develop AI models that effectively incorporate physical laws, chemical principles, etc., making the model's prediction results scientifically based and explainable. At the same time, strengthen the verification and evaluation of the models to ensure their reliability and accuracy, providing reliable support for R&D decision-making.

Conduct research on multi-scale modeling technology, establish cross-scale models from micro to macro, and achieve the fusion and analysis of data at different scales. By optimizing the algorithms and computational methods of the models, improve the computational efficiency and accuracy of the models, reduce error accumulation, and enhance the practicality and reliability of the models in the R&D of chemical materials.

Actively explore learning methods suitable for small sample data, such as transfer learning and meta-learning, to improve the model's generalization ability and predictive performance under small sample data. Strengthen research on data augmentation techniques, expand small sample datasets through data augmentation methods, enhance the training effect of the model, and effectively address the challenges brought by small sample data.

03 | At the talent level, build an efficient talent cultivation system, interdisciplinary integration training, focusing on attracting and retaining talent

Establish an interdisciplinary talent cultivation system, strengthen the integration of chemical materials specialty with computer science, mathematics, statistics, and other specialties, and cultivate compound talents who are proficient in both chemical materials and AI. Encourage employees to actively participate in interdisciplinary learning and training, enhance the comprehensive quality and interdisciplinary capabilities of employees, and provide a solid talent guarantee for the application of AI technology in the enterprise.

Strengthen the cooperation between universities and enterprises, establish internship bases and practical platforms, provide more opportunities for university students to engage in actual projects, and cultivate students' practical abilities and innovative awareness. At the same time, strengthen the internal talent development within enterprises, and improve employees' AI technology levels and application capabilities through internal training, project practice, and other methods.

Formulate competitive talent policies to enhance the appeal of the chemical materials industry to talents. Provide talents with a good working environment and broad development opportunities, give them challenging project tasks, and stimulate their innovative vitality. Strengthen corporate culture construction, enhance the sense of belonging and loyalty of talents, retain excellent talents, and build a stable and high-quality talent team.

Only by actively responding to challenges and taking practical and effective measures, accelerating the application of AI technology in the research and development of chemical materials, and enhancing the innovation capabilities and core competitiveness of enterprises, can companies stand out in fierce market competition and lead the innovative development of the chemical materials R&D industry.