AI can efficiently discover hidden patterns in data-rich fields such as climate science, ecosystems, life sciences, and complex diseases. Photo: TUCHONG

Analyzing user sentiment and predicting event trajectories through natural language processing; constructing multi-agent systems with varying policy parameters to simulate how policies shape social trends; mining vast troves of historical and real-time economic data to uncover market patterns and improve forecasting accuracy; employing generative AI models to review and synthesize vast bodies of social science literature—across these applications, AI demonstrates remarkable potential in the humanities and social sciences.

This has prompted some in academic circles to suggest that research in these fields is entering a “fifth paradigm” driven jointly by data and mechanism, one that greatly broadens research objects, deepens mechanistic analysis, and accelerates scientific progress. What exactly is the fifth paradigm? How does it differ fundamentally from those before it? And what profound changes might it bring to researchers?

Blending traditional paradigms

The concept of research paradigms was first introduced by American philosopher of science Thomas Kuhn. It refers to the worldview and methods shared by a scientific community that ensure efficient and orderly scientific research activities. Historically, scientific paradigms have evolved through four stages: the empirical paradigm, primarily based on experimental induction; the theoretical paradigm, centered on models and logical deduction; the computational paradigm, using computers to simulate and solve problems across various disciplines; and the data-driven paradigm, characterized by data-intensive research enabled by internet technologies.

With the advancement of deep learning and generative large models, AI’s capacity for data generation, analysis, pattern recognition, and causal reasoning has grown increasingly powerful. In areas such as materials science, genomics, drug discovery, and climate modeling, AI can reveal deeper natural laws and propose new avenues for discovery with efficiency far beyond human scientists, thereby advancing interdisciplinary research.

In recent interviews with CSST, several scholars agreed that AI is reshaping the process of scientific discovery primarily by transcending and integrating traditional paradigms. It accelerates the research cycle: While conventional inquiry follows a lengthy “hypothesis–experiment–verification” loop, AI can automate hypothesis generation, optimize experiment design, and execute unmanned experiments, dramatically raising efficiency. It can broaden the frontiers of scientific knowledge by generating entirely new hypotheses and even addressing long-standing mathematical problems. By integrating experimental, theoretical, computational, and data paradigms, AI enables multi-paradigm collaboration and fosters interdisciplinary exchange.

Wang Dong, an associate professor at the School of Marxism, Beijing Technology and Business University, explained that AI possesses inherent advantages in domains beyond human intuitive grasp, such as quantum phenomena at microscopic scales, high-dimensional spatial structures, and nonlinear systems. In fields involving massive data and complex variables—including climate science, ecosystems, life sciences, and complex diseases—AI can efficiently process and analyze vast datasets, detecting patterns that traditional methods cannot. It also opens promising new frontiers in cutting-edge interdisciplinary domains such as systems biology, brain-inspired research combining neuroscience and AI, and the integrated design of new materials and chemistry.

Extension of the fourth paradigm

But does AI’s rise amount to a true paradigm shift? Many scholars urge caution.

Wu Libo, executive director of the Comprehensive Laboratory of National Development and Intelligent Governance at Fudan University, remarked that technological revolutions drive cognitive evolution. In today’s “data + mechanism” era, AI models are advancing from “black-box” prediction tools to “grey-box” systems with reasoning capabilities. This shift allows researchers not only to describe the world through data but also to understand it through scientific mechanisms.

However, while scholars argue that AI-driven research enhances efficiency, whether it merits recognition as a fifth paradigm remains uncertain. Key questions include: Has it created entirely new research pathways? Has it solved problems unsolvable by traditional paradigms? Has it transformed the roles and collaborations of research subjects? Has it spawned a new generation of research infrastructure? Such uncertainties have led many in academia to see AI as an extension of the fourth paradigm, still reliant on human-labeled data.

Chen Yinzheng, deputy secretary-general of the Engineering History Committee of the Chinese Society for the History of Science and Technology, similarly believes that although AI is profoundly transforming research models, it has not yet triggered a full paradigm shift. AI is most effective in interdisciplinary settings, but traditional experimental and theoretical approaches remain indispensable. Results obtained via AI still require validation through conventional methods. Presently, AI is better at following specific research paths but is less suited to deriving entirely new principles or conducting free, original theoretical exploration. Especially in social sciences such as archaeology and the history of science and technology, deep expertise and judgment from scholars remain fundamental. Moreover, AI raises questions of integrity and ethics: Its reliability and stability remain limited, and algorithmic biases can even lead it to generate high-risk experimental plans. Such issues continue to demand oversight from traditional models.

Refreshing disciplinary development

Despite these reservations, scholars acknowledge AI’s extraordinary capacity for deep learning and exploration, seeing in it the potential to open knowledge domains far beyond human reach. The interplay between generative and discriminative models enables rapid probing of vast unknowns, injecting fresh vitality into disciplinary development.

Wu predicted that future social science research will increasingly emphasize cross-disciplinary data integration and methodological innovation, with AI as a central driver. Through advanced techniques such as multi-agent modeling, researchers can simulate complex social dynamics, reveal subtle human interactions, and deeply probe the mechanisms underlying social phenomena. Techniques such as deep learning and reinforcement learning will further enhance analytical and creative capacities, enabling more precise and profound insights into social realities.

Wang foresaw that at the stage of cognitive substitution and transcendence, AI will evolve into autonomous research agents capable of actively identifying problems, generating scientific hypotheses, and autonomously conducting observations and experiments, with humans shifting into supervisory roles. The future research landscape will move from human-human teamwork to AI collaboration under human guidance. Research will dramatically expand in both scale and depth, disciplinary boundaries will blur, and cross-field collaboration and knowledge integration will become normalized, giving rise to new interdisciplinary frontiers.

Yet vigilance is essential. Chen cautioned that while AI can enhance research efficiency within traditional disciplines, free researchers from repetitive tasks, and expand traditional research boundaries, it cannot replace traditional inquiry altogether. Over-reliance on AI risks homogenizing innovation and stifling creativity. “The sustainable development of science and research is closely tied to human creativity, which AI can hardly replace,” Chen emphasized.

Scholars concurred that although the so-called fifth paradigm introduces new trends—such as the automation of scientific research and the “agentification” of research subjects—true breakthroughs still depend on human scientific intuition, discernment of fact, and sensitivity to context and cultural background. The fifth paradigm may open a new cognitive landscape where research objects surpass the capacity of ordinary scientists. Yet as humanity steps into this new stage of exploration, it is equally urgent to reflect on the boundaries and responsibilities of technology.