Using AI to Study AI: The Intelligent Turn in Social Science Research

The advent of the era of artificial intelligence is bringing about a broad and profound reshaping effect on human society, and thereby driving an iterative development of social science at the levels of ontology, epistemology, and methodology. With the rapid iteration and widespread application of digital-intelligent technologies, AI technology is promoting the rise of the fifth paradigm—intelligent science (AI for Science, AI4S). This paradigm emphasizes the use of artificial intelligence to automatically extract features, mine patterns, conduct reasoning analysis, and generate knowledge from multi-source heterogeneous data.

人工智能时代的到来给人类社会带来广泛而深刻的重塑效应，进而带来社会科学在本体论、认识论和方法论层面的迭代发展。随着数智技术快速迭代和普及应用，人工智能技术正在推动第五范式——智能科学（AI for Science，AI4S）的兴起。这一范式强调利用人工智能从多源异构数据中自动提取特征、挖掘规律、推理分析并生成知识。

From AI4S to AI4SS

从AI4S到AI4SS

The intelligent science paradigm is bringing revolutionary changes to social science. Its breakthrough lies not only in the enhanced efficiency of processing massive heterogeneous data, but also, through simulating human cognitive mechanisms and rules of social interaction, in opening up a two-way cognitive path of "social system computation—computational model socialization." From this perspective, generative artificial intelligence represented by large language models can be understood as the collision and fusion of social intelligence and machine intelligence. The driving force of technological innovation comes not only from machine intelligence with high-performance computing and robustness, but also depends on social intelligence that generates and understands massive amounts of human social data, high-quality annotated data, and achieves human-machine value alignment, as well as the continuous shaping and integration of the two through human-machine interaction mechanisms. In short, this paradigm revolution provides an epochal opportunity to integrate social systems and computational systems, stimulate social intelligence and machine intelligence, and explore a paradigm of social knowledge production through human-machine interaction.

智能科学范式正为社会科学带来革命性变革。其突破性不仅在于海量异构数据处理效能的提升，更在于通过模拟人类认知机理与社会互动规则，开创了“社会系统计算化—计算模型社会化”的双向认知路径。从这一视角，以大语言模型为代表的生成式人工智能可被理解为社会智能与机器智能的碰撞与融合，技术革新动力不仅来自具有高性能计算、具有鲁棒性的机器智能（Machine Intelligence），更依赖于生成和理解海量人类社会数据、高质量标注数据和实现人机价值对齐的社会智能（Social Intelligence），以及通过人机交互机制实现二者的不断形塑与融合。总之，这场范式革命为融合社会系统与计算系统、激发社会智能与机器智能、探索人机交互的社会知识生产范式提供了时代机遇。

The State Council's "Opinions on Deeply Implementing the 'AI+' Action" [1]1 A policy document issued by the State Council of the People's Republic of China in 2024, formally titled "Opinions on Deeply Implementing the 'AI+' Action," which outlines measures to integrate AI into various sectors including scientific research. explicitly proposes promoting the transformation of philosophy and social science research methods toward a human-machine collaborative model, exploring the establishment of new organizational forms for philosophy and social science research adapted to the era of artificial intelligence, and expanding research horizons and observational perspectives. AI empowering social science has become a new direction, new paradigm, and new trend for social science development. China should leverage its first-mover advantage in AI and its applications to accelerate the advancement of AI for Social Science (AI4SS), achieving paradigm innovation in social science.

国务院发布《关于深入实施“人工智能+”行动的意见》明确提出，推动哲学社会科学研究方法向人机协同模式转变，探索建立适应人工智能时代的新型哲学社会科学研究组织形式，拓展研究视野和观察视域。人工智能赋能社会科学成为社会科学发展的新方向、新范式和新趋势，我国应依托人工智能及其应用的先发优势，加快推进人工智能赋能社会科学（AI for Social Science，AI4SS），实现社会科学的范式创新。

Using AI to Study AI: Measurement, Inference, and Prediction

以人工智能研究人工智能：测量、推断与预测

"Using AI to study AI" emphasizes that artificial intelligence possesses a dual attribute as both research object and research method in social science. As a research object, AI has a broad and profound impact on economic and social operations, national governance, and social culture. As a research method, large language models provide entirely new methodological paradigms and technical tools for social science in areas such as measurement, inference, and prediction. As a completely new methodological paradigm, AI is empowering all stages of social science research, injecting new vitality into research stages such as problem discovery, theory construction, research design, data collection, and data analysis. This paradigm change is reflected not only at the level of research means and tools, but also in the major transformation of human-machine relationships and human-machine interaction in the process of knowledge production in social science, promoting the deep integration of theory-driven paradigms, data-driven paradigms, and model-driven paradigms in the social science field.

“以AI研究AI”强调人工智能在社会科学中同时具有研究对象与研究方法的双重属性。作为研究对象，人工智能对经济社会运行、国家治理、社会文化产生广泛而深远的影响。作为研究方法，大语言模型为社会科学在测量、推断与预测等方面提供了全新的方法范式和技术工具。作为全新的方法范式，人工智能正在赋能社会科学研究的各环节，在问题发现、理论构建、研究设计、数据采集、数据分析等研究环节注入新的活力。这种范式变革不仅体现在研究手段和工具层面，更反映为社会科学在知识生产过程中人机关系、人机交互的重大变革，促进了社会科学领域理论驱动范式、数据驱动范式和模型驱动范式的深度融合。

In terms of measurement, compared with traditional supervised or unsupervised machine learning methods, large language models show clear advantages in text understanding and generation tasks. They do not require constructing large-scale annotated datasets in advance; they can directly complete classification and recognition tasks through natural language instructions, thereby reducing labor costs and minimizing human annotation errors. Large language models can also be used to assist questionnaire surveys and experimental research, for example, answering subjects' questions in real time, generating multi-version contextualized questions, and conducting pilot surveys to calculate statistical power, thereby improving measurement validity and subject participation.

在测量方面，相较于传统的有监督或无监督机器学习方法，大语言模型在文本理解与生成任务中表现出明显的优势。其无需事先构建大规模标注数据集，可以直接通过自然语言指令完成分类与识别任务，从而降低人力成本并减少人为标注误差。大语言模型还可用于辅助问卷调查与实验研究，例如实时解答被试疑问、生成多版本情境化问题、进行试调查测算统计力等，从而提升测量效度和被试参与度。

In terms of inference, large language models bring "silicon-based samples" to social science research. During the training process, large language models absorb a large amount of information about human social structures, behavioral preferences, and values, thus possessing the potential to simulate human cognition and behavior. By constructing silicon-based samples with specific social attributes through prompt engineering, it is possible to break through the limitations of traditional surveys and experimental research in terms of time, cost, and scale to a certain extent, and they can be used for pre-testing, experimental design optimization, and effect testing.

在推断方面，大语言模型为社会科学研究带来了“硅基样本”。大语言模型在训练过程中吸收了大量关于人类社会结构、行为偏好和价值观的信息，因而具备模拟人类认知与行为的潜力。通过提示词工程构建具有特定社会属性的硅基样本，可以在一定程度上突破传统调查与实验研究在时间、成本与规模上的限制，可用于预测试、实验设计优化及效应检验。

In terms of prediction, the continuous learning ability and reasoning ability of large language models provide a new path for social science to move from static prediction to dynamic prediction. By combining data-driven prediction and simulation modeling paradigms, developing intelligent agent clusters based on massive human social data, and conducting multi-agent simulations, it helps to capture the complexity and uncertainty in social evolution and provides inspiration for understanding group behavior and social evolution.

在预测方面，大语言模型的持续学习能力和推理能力，为社会科学从静态预测走向动态预测提供了新路径。通过结合数据驱动预测和仿真模拟范式，基于海量人类社会数据开发智能体集群，通过多智能体模拟有助于捕捉社会演化中的复杂性和不确定性，并为理解群体行为和社会演化提供启发。

Artificial intelligence is driving social science research toward a new paradigm of human-machine collaboration. This paradigm revolution presents three synchronously advancing new directions of methodological paradigms: first, the combination of data-driven and model-driven paradigms, i.e., using AI to enhance computational social science research; second, the combination of model-driven and theory-driven paradigms, embodied centrally in AI-generated intelligent agent clusters that demonstrate the heterogeneity and emergent capabilities of social evolution through multi-agent simulation; third, the combination of theory-driven, data-driven, and model-driven paradigms, embodied centrally in the intellectualization of the entire process of social science research—posing questions, literature analysis, data collection, data analysis, and analytical interpretation—providing new opportunities for the paradigm transformation of social science.

人工智能推动社会科学研究迈向人机协同新范式。这场范式革命呈现出三个同步推进的方法范式新方向：一是数据驱动与模型驱动范式的结合，即以AI增强计算社会科学的研究；二是模型驱动与理论驱动范式的结合，集中体现为AI生成智能体集群，通过多智能体仿真模拟展现社会演化的异质性和涌现能力；三是理论驱动、数据驱动与模型驱动的结合，集中体现为社会科学研究提出问题、文献分析、数据采集、数据分析和分析阐释全过程的智能化，为社会科学范式转型提供了新契机。

AI Empowering Social Science Is Becoming a Global Trend

全球范围内人工智能赋能社会科学蔚然成风

In recent years, the U.S. National Science Foundation has focused on laying out the field of social computing, with its Social Intelligence Systems program funding dozens of disciplinary projects cumulatively. Under the framework of the EU's "Horizon Europe" program, a flagship project on AI empowering scientific research has been specifically established. Japan's "Society 5.0" strategy [2]2 A national strategy proposed by Japan in 2016, envisioning a "super-smart society" that uses digital technologies (AI, IoT, big data) to solve social problems and drive economic growth, following hunting, agrarian, industrial, and information societies. integrates billion-scale multi-source data such as mobile positioning and consumer behavior, hoping to solve social problems through this. The UK's Policy Lab has taken the lead in introducing generative AI into public policy evaluation, constructing virtual simulation environments for different institutional reforms. At the same time, the era of AI has also given rise to many entirely new issues that urgently require social science intervention. For example, Canada has promoted an AI ethics certification system, requiring that research projects involving intelligent algorithmic decisions affecting humans must pass an algorithmic impact assessment. The Max Planck Institute in Germany is leading the development of an explainable social computing framework, focusing on evaluating dimensions such as fairness, accountability, transparency, and ethical compliance.

美国国家科学基金会近年来重点布局社会计算领域，其社会智能系统专项累计资助数十个学科项目。欧盟“地平线欧洲”计划框架下，专门设立人工智能赋能科学研究旗舰项目。日本推进“社会5.0”战略则整合十亿级规模的移动定位、消费行为等多源数据，期望以此解决社会问题。英国政策实验室率先将生成式人工智能引入公共政策评估，构建不同制度改革的虚拟推演环境。与此同时，人工智能时代也涌现出诸多迫切需要社会科学介入的全新议题。譬如，加拿大推行人工智能伦理认证体系，要求涉及人类的智能算法决策的研究项目必须通过算法影响评估。德国马克斯·普朗克研究所主导开发可解释社会计算框架，重点评估公平性、可问责性、透明性和伦理合规性等维度。

Globally, the rapid development of AI-driven social science has formed diverse tensions. Among them, the theoretical debate between technological empowerment and disciplinary autonomy continues to intensify: some existing research overly relies on algorithmic black boxes, leading to weakened explanatory power, while traditional resistance to technological change causes innovation stagnation. The institutional game between data openness and privacy protection is becoming increasingly acute, and the balancing mechanism between personal information protection and scientific research data utilization remains imperfect. The structural gap between innovation speed and governance capacity is gradually widening, with explainable models and algorithmic auditing mechanisms becoming key supplementary measures. The bottleneck in transforming academic value into social impact urgently needs to be broken, and new models of academic organization for AI-empowered social science remain to be explored.

全球范围内人工智能驱动社会科学快速发展已形成多样化的张力。其中，技术赋能与学科本位的理论争议持续凸显，既有研究过度依赖算法黑箱导致解释力弱化，亦有传统抗拒技术变革造成创新迟滞；数据开放与隐私保护的制度博弈日益尖锐，个人信息保护与科研数据利用平衡机制仍未健全；创新速度与治理能力的结构性落差逐渐扩大，可解释性模型与算法审计机制成为关键补位手段；学术价值与社会影响的转化瓶颈亟待突破，以人工智能赋能社会科学的学术组织新模式有待探索。

A Chinese Approach to the Intelligent Turn in Social Science Research

社会科学研究智能化转向的中国方案

China is a pioneering country in global digital-intelligent development, possessing unique advantages for AI-empowered social science. China's treasure trove of behavioral data generated by its super-large population, its institutional advantage of concentrating resources to accomplish major tasks [3]3 A standard Chinese political phrase referring to the ability of the Party-state to mobilize and allocate resources effectively for national priorities, often contrasted with the perceived fragmentation of decision-making in multiparty systems., its technological breakthroughs in autonomous open-source large models, and its rich scenarios of digital transformation and social governance innovation constitute the strategic depth for theoretical breakthroughs and technological innovation. Therefore, it is necessary to base ourselves on the pioneering practice of digital-intelligent development and governance, vigorously promote AI-empowered social science in terms of infrastructure, disciplinary construction, talent cultivation, and academic ecology, construct an independent knowledge system of China at the international academic frontier, and explore a new model of human-machine collaborative knowledge production.

我国是全球数智化发展的先发国家，具有人工智能赋能社会科学的独特优势。我国超大规模人口产生的行为数据富矿、集中力量办大事的体制优势、自主开源大模型的技术突破以及数字化转型与社会治理创新的丰富场景，构成了理论突破与技术创新的战略纵深。因此，需立足数智化发展与治理的先发实践，大力推进人工智能赋能社会科学的基础设施、学科建设、人才培养、学术生态等，在国际学术前沿构建中国自主的知识体系，探索人机协同的知识生产新模式。

At the level of research conditions, accelerate the construction of AI4SS research infrastructure, integrate data resources, computing power resources, and knowledge base resources in the social sciences, build an AI4SS research service platform open to the whole country, form an AI4SS research infrastructure system covering theoretical modeling, data mining, simulation and deduction, and knowledge generation, and develop general-purpose intelligent tools to empower the broad community of researchers.

科研条件层面，加快建设AI4SS科研基础设施，整合社会科学领域的数据资源、算力资源、知识库资源，打造面向全国开放的AI4SS科研服务平台，形成覆盖理论建模、数据挖掘、仿真推演、知识生成的AI4SS科研基础设施体系，开发通用性智能工具赋能广大研究者。

At the level of discipline construction, accelerate the cross-fertilization of artificial intelligence and the social sciences. First, use AI methods to study complex social systems, empower innovation in traditional research methods, and achieve more efficient, more integrated, and more interactive social science knowledge production; second, empower AI technology innovation based on social science knowledge, especially by building corpora, databases, and knowledge bases with a social science knowledge foundation, thereby strongly supporting the development of specialized or vertical models applicable to specific scenarios; third, conduct social science research on the mechanisms and effects of AI’s impact on social systems, providing a knowledge reserve for the development and governance of intelligent society.

学科建设层面，加快推进人工智能与社会科学的交叉融合。一是利用人工智能方法研究复杂社会系统，赋能传统研究方法创新，实现更高效、更集成、更交互式的社会科学知识生产；二是基于社会科学知识赋能人工智能技术革新，尤其是建设具有社会科学知识基础的语料库、数据库、知识库，有力支撑适用于特定场景的专用或垂类模型开发；三是对人工智能对社会系统的影响机制和效应开展社会科学研究，为智能社会发展与治理提供知识储备。

At the level of talent cultivation, accelerate the training of interdisciplinary composite talents in “AI + social science.” Promote the establishment of a deeply integrated talent cultivation system between universities and technology enterprises, accelerate the development of intelligent agents for social science disciplines, cultivate composite talents in “social science + data science + AI ethics,” and explore a collaborative innovation model of “social scientist + AI engineer + domain expert.”

人才培养层面，加快培养“人工智能+社会科学”交叉复合型人才。推动高校与科技企业构建深度融合的人才培养体系，加快社会科学学科智能体开发，培育“社会科学+数据科学+AI伦理”的复合型人才，探索“社会科学家+AI工程师+领域专家”的协同创新模式。

In terms of the academic ecosystem, build an open and shared AI-empowered social science academic ecosystem. Issue AI ethics guidelines for the social sciences, standardize academic norms such as virtual experiment design, silicon-based sample application, and generative output review; optimize the academic evaluation mechanism, establish incentives for code open-sourcing and model sharing; build a developer community for social science large models, forming a new type of academic network for data annotation crowdsourcing, algorithm optimization collaboration, and code and data sharing.

学术生态方面，构建开放共享的人工智能赋能社会科学学术生态。出台社会科学领域人工智能伦理指南，规范虚拟实验设计、硅基样本应用、生成式成果审查等学术规范；优化学术评价机制，建立代码开源与模型共享激励机制；建设社会科学大模型开发者社区，形成数据标注众包、算法优化协作、代码数据共享的新型学术网络。

(The author is Executive Director and Professor of the Laboratory of Computational Social Science and State Governance at Tsinghua University.)

（作者系清华大学计算社会科学与国家治理实验室执行主任、教授）