The Changing Path of Scientific Discovery

AI is increasingly involved in scientific research, acting as a powerful engine for acceleration, from predicting protein structures to discovering new materials. The traditional research model of “hypothesis-verification” is evolving into a new paradigm of “data-pattern discovery-intelligent generation-closed-loop iteration.”

Wang Xijun, a professor at the University of Science and Technology of China, notes that AI can autonomously discover patterns in massive datasets, allowing for precise material design tailored to specific goals. For instance, in the field of framework materials, AI can quickly predict material properties, significantly reducing the costs associated with trial-and-error experiments.

Generative AI further enables research to transition from “selecting the known” to “creating the unknown,” allowing for the direct generation of new material structures based on desired performance characteristics. This evolution signifies that AI is not only accelerating problem-solving but also expanding the boundaries of the questions being asked.

AI’s role in research is continuously evolving from a computational tool to an analytical partner, and ultimately to a driver of autonomous exploration. However, AI will not replace scientists; human judgment and insight remain crucial for understanding key scientific issues. The collaboration between humans and AI will create a broader space for future research innovation.

Enhanced Efficiency in Research Innovation

AI excels at handling tasks with clear answers that require extensive repetitive calculations. Professor Mo Bofeng from Capital Normal University highlights that AI significantly improves efficiency in literature review, experimental design, and data analysis, even in the study of oracle bones from over 3000 years ago.

AI can assist in tasks like reconstructing broken oracle bones and restoring missing images, which previously relied heavily on expert experience. However, while AI can identify subtle features undetectable by humans, the sheer volume of oracle bones (over 160,000 pieces and more than a million characters) still necessitates human oversight, especially for deep semantic judgments.

The collaboration between humans and AI is just beginning, with potential advancements in classification, aggregation, and translation of oracle bones as technology progresses. Future researchers will need to enhance their data processing skills and leverage technology to amplify their research capabilities.

The Impact of AI on Research Judgment

AI is transforming the entire research process from a linear model of hypothesis formulation, experimentation, and result analysis to a collaborative, predictive, and iterative system. Yang Yaodong, a researcher at Peking University, emphasizes that while AI enhances efficiency, it also raises concerns about false citations, erroneous reasoning, and academic integrity.

AI can quickly sift through vast amounts of data, freeing researchers from repetitive tasks to focus on critical issues. However, scientific understanding requires more than just accurate predictions; it also demands an understanding of the underlying reasons. If AI models are opaque or data sources unclear, the conclusions drawn may pose new risks.

The essential question remains: while AI can optimize existing data, determining which problems are worth studying and which results are scientifically significant still requires human oversight.

Effective Resource Integration

Wu Libo, assistant president of Fudan University, discusses the shift from a technology-centric to a scientist-centric approach in scientific intelligence. The establishment of the Xinghe Qizhi Open Platform aims to lower the barriers for scientists to utilize AI throughout the research process.

This platform provides comprehensive infrastructure covering data, models, computational power, experiments, intelligent agents, and collaboration communities. It has already aggregated over 400 scientific models, 22PB of high-value data, and 500 million literature patents, enabling scientists to conduct research without delving into technical details.

The platform also fosters cross-disciplinary collaboration, allowing different fields to share and reuse results. It serves as a hub connecting scientists, AI engineers, and industry forces, facilitating systematic acceleration of innovation.

Building and Utilizing Intelligent Platforms

Liu Tieyan, president of Zhongguancun College, emphasizes that having many platforms does not guarantee their effectiveness. A survey of over 30 materials companies revealed that only 20% of critical issues could be addressed with current mainstream scientific intelligence technologies. The remaining challenges stem from low digitalization, data deficiencies, and algorithmic limitations.

To overcome these challenges, Liu suggests three approaches: promoting industrial digitalization to align scientific research with real industry needs, establishing incentives for open sharing of data and tools, and creating foundational infrastructure for cross-disciplinary collaboration.

The Debate on AI as a Research Author

A recent call for papers from East China Normal University sparked discussions by stating, “The first author must be AI.” This experiment aims to explore the ethical boundaries of AI in knowledge production and academic research.

While supporters view this as a groundbreaking experiment in academic norms, critics worry it signifies a retreat from human involvement in research. The experiment collected 820 papers with AI as the first author, revealing AI’s strengths in topic planning, outline generation, and data analysis, but also its limitations in creativity and value judgment.

The consensus emerging from this debate is that while AI can assist in research writing, humans should maintain roles as problem posers and quality controllers, ensuring academic integrity and responsibility remain intact.