The large-scale social experiment of “AI as the First Author” has concluded, yet the reflections it has inspired are far from over. From four dimensions, namely the experiment’s impacts on academic norms and ethics, knowledge production modes and knowledge power, scientific research systems and the identification of research achievements, as well as educational systems and teacher-student relationships, this paper puts forward nine thought-provoking questions. What is the most practical challenge posed by the “AI as the First Author” experiment? What impact does it exert on knowledge production? What impact does it exert on knowledge power? What impact will knowledge equalization have on society? What impact does it have on academic papers and academic journals? How should knowledge and individuals’ academic contributions be evaluated in the future? What inevitable transformations will higher education have to undergo? What is the “core competence” that is the most difficult for AI to replace? What kind of new teacher-student relationship should be constructed? among others. This paper proposes important viewpoints and concepts including “AI hegemony”, “human guarantor system” and “value of trust”. In particular, it puts forward the concept of the Human-Machine Collaboration Quotient (C-Quotient), and holds that C-Quotient, together with Intelligence Quotient (IQ) and Emotional Quotient (EQ), will jointly constitute the core competencies of human beings in the era of artificial intelligence. It also envisions a future where carbon-based life and silicon-based life dance together on the same stage, and humans and machines co-evolve.

In September 2025, East China Normal University launched the world’s first large-scale social experiment of “AI as the First Author”. Using a quasi-field experimental method, it carried out an essay-soliciting activity on “AI-Driven Educational Research Paper Writing”, requiring AI to be the first author and humans to play the roles of collaborators and reviewers. Over a period of half a year, 724 valid submissions were received from both domestic and international sources. Based on this, the university explored the fifth paradigm of AI-empowered research in philosophy and social sciences, academic ethical norms, and the path to constructing an independent knowledge system in education. The experiment went through eight stages: intervention design, response monitoring, expert seminars, data collection, AI-based manuscript review, human-machine consistency testing, data analysis, and result publication. A mixed-research method was adopted to reveal the core findings. AI has significant advantages in aspects such as inspiration generation, information processing, and text polishing, but it has limitations such as fictional literature, logical hollowness, insufficient innovation, and ethical risks; efficient AI application can significantly enhance academic contributions, forming six innovative patterns and five human-AI collaboration models; AI-based manuscript review is reliable and complementary to the evaluation by human experts, and there are obvious differences in the research tastes of different large language models; there is a significant “AI generation gap” in the academic community, with the younger generation being more adaptable to human-AI collaboration, and AI has, to some extent, promoted intellectual equality. The research proposed that in the future, efforts should be made to promote the transformation of research paradigms, reform of evaluation systems, development of research-specific AI agents, reconstruction of educational models, and innovation of diploma certification. Findings provide empirical support and practical inspiration for academic innovation and educational transformation in the intelligent era.

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research. Traditional research on education policy often has limitations in evaluating the long-term dynamics and systemic impacts of macro-level policies. To overcome these constraints, this study adopts a computational experiment/simulation research method based on Agent-Based Modeling (ABM) and constructs a computational laboratory named “EduEcosystem”. The model defines three types of heterogeneous agents: students, teachers, and schools, whose attributes and interaction rules are deeply rooted in sociological and psychological mechanisms. Through simulating and deducing the teacher rotation policy, this study reveals a series of complex effects. In the initial stage of policy implementation, it can indeed promote educational equity, with a significant decline in the knowledge Gini coefficient. However, this equity benefit comes at a high cost: the teacher turnover rate rises sharply by about 82.5%, accompanied by a synchronous decline in the overall academic performance level. From the perspective of long-term dynamic evolution, due to the cumulative effect of teacher burnout and adaptive changes within the system, the equity effect of the policy fails to sustain and gradually weakens, eventually giving rise to the counterintuitive phenomenon of “equity rebound”. This study confirms that multi-agent simulation can serve as an effective policy laboratory for understanding complex education systems, and can also preview the long-term dynamic changes and nonlinear chain reactions that policies may trigger. This methodological innovation provides a new approach for the construction of the pedagogical knowledge system and scientific educational decision-making. This study is a computational simulation thought experiment rather than an empirical research, and its conclusions are only for policy reference.

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research. The rapid advancement of Generative Artificial Intelligence (GenAI) signifies a transition toward a new phase of “human-AI collaboration” in scientific research and educational practice. However, existing research predominantly focuses on the accuracy of content generated by GenAI, lacking empirical investigation into its mechanism for fostering deep learning as a cognitive scaffold within extended interactions. Based on 12,824 log entries of multi-turn dialogues between learners and GenAI concerning algorithmic learning, this study employs Natural Language Processing (NLP) techniques to construct an automated coding system and utilizes Lag Sequential Analysis (LSA) to deeply examine the temporal interaction characteristics and cognitive iteration patterns in the human-AI collaborative problem-solving process. The findings reveal that: (1) learner-GenAI interactions exhibit significant long-term and asymmetric characteristics, with GenAI providing continuous scaffolding support through an asymmetric discourse pattern of “few questions, many answers”; (2) regarding behavioral sequences, there exists a high-frequency closed-loop between learners’ self-correction and GenAI’s corrective guidance, demonstrating that deep learning does not occur in single Q&A exchanges but is achieved through a spiral process of “trial-and-error, feedback, and re-correction”; (3) micro-case analysis reveals that when GenAI falls into algorithmic fixation, learners swiftly shift from being questioners to strategy formulators, implementing key interventions through metacognitive monitoring. GenAI should not be regarded merely as an information retrieval tool in future education; instead, it should be positioned as a “Socratic tutor” that inspires thinking, reshapes learner agency, and jointly constructs a new educational ecology of human-AI symbiosis.

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research. The emergence of the AI-driven fifth paradigm of scientific research presents unprecedented challenges to the cognitive division of labor in educational research. Grounded in a meta-research perspective, this study designs and executes an “Adversarial AI-Delphi Method” by constructing a “Silicon-based Expert Panel” comprising heterogeneous large language models (LLMs). Through a three-stage dialectical deduction, it maps the evolutionary landscape of educational research paradigms from the bottom up, taking the vantage point of the AI collective mind. The findings reveal a tripartite structure: (1) morphologically, the role of AI undergoes a progressive leap along the “acceleration–simulation–generation” trajectory, evolving from data insight to autonomous inquiry; (2) critically, AI-driven research is entangled in structural traps, including the systematic forgetting of “incomputable” dimensions, the neglect of educational “slow variables,” and the “banalization” of academic innovation; and (3) axiologically, rather than replacing human researchers, AI acts as a “Sacrificial Epistemic Other.” Through extreme formalized computation, it inversely confirms the boundaries of incomputable educational meaning, thereby compelling the ethical return of human subjective responsibility. Ultimately, within the symbiotic dialectic between the computable and the incomputable, this study constructs a “Typological Matrix of Human-Machine Collaborative Educational Research,” offering a theoretical coordinate—at once epistemologically deep and practically navigable—for the paradigm shift that navigates the tensions between computability and value risk.

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research.The success of education reform depends both on the scientific rigor of the policy and on whether stakeholders can reach broad cognitive consensus. However, traditional social surveys struggle to capture the dynamic process of opinion evolution, and rule-based simulation models lack the depiction of complex semantics and cognitive mechanisms. This study introduces the “Generative Agent-based Social Simulation” paradigm, constructing a virtual education community “The Ville” containing 25 Generative Agents with independent personalities, memories, and reflection capabilities. By simulating the implementation process of a “Project-Based Learning (PBL)” policy over 7 days, this study reveals the micro-macro emergence mechanism of education reform opinions moving from “cognitive islands” to “social consensus.” The study finds that, first, without mandatory intervention, the community eventually formed a high support consensus of 92%, with no group polarization. Second, the aggregation of physical space and a high-density weak tie network provided the structural basis for breaking information cocoons. Finally, the transformation of “rational skeptics” is a key turning point in consensus formation, where deep persuasion mechanisms based on problem-solving are more influential than mere emotional appeals. This study provides a new “policy evolution laboratory” method for computational education science, demonstrating the potential of using large language models to explore complex educational social issues.

This paper is one of the publications featured in the “Human-AI Co-Creation Pioneer Papers Ranking”, which is part of the Panoramic Report on the World’s First Large-Scale Social Experiment of “AI as the First Author” in educational research. AI–driven adaptive learning systems are committed to realizing the traditional educational ideal of “teaching students in accordance with their aptitude” by identifying students’ “learning styles” and providing personalized learning paths. Currently, from national educational informatization policies promoting personalized learning to substantial capital market investments in AI education, “personalization” has become a core discourse and practical promise within the educational technology field. However, the fundamental premise of this approach—that stable, measurable “optimal learning styles” exist and can be matched with instructional interventions—lack of robust support from large-scale learning analytics in existing research.Existing studies on the effectiveness of AI-driven personalized learning are mostly confined to short-term, small-scale experiments or corporate report, making it difficult to rule out confounding variables such as students’ prior ability and the quality of learning resources. Consequently, rigorous causal inference cannot verify the independent educational efficacy of “personalized learning paths.” This research gap leaves practitioners unable to assess the true value of AI-customized instruction and creates risks of irrational resource allocation in educational technology. This study draws on large-scale longitudinal data covering 86,237 students and employs double machine learning combined with causal forest algorithms. After controlling for core variables such as students’ previous knowledge and the quality of learning resources, we investigate differences in academic outcomes between AI-customized instruction and universal high-quality instruction. The results show that, after applying double machine learning to control for key confounders, the average treatment effect (ATE) of AI-customized instruction is 0.01 standard deviations (95% CI [−0.01, 0.03], p=0.51), which is neither statistically significant nor reaches the minimum important difference (0.2 SD) for educational interventions. Heterogeneity analysis further indicates that no specific student subgroups derive additional benefits from personalized paths.

To accelerate the exploration of new research paradigms for philosophy and social science in the age of intelligence, the Faculty of Education at East China Normal University and other institutions, have launched a large-scale social experiment called “AI as the First Author”. The experiment conducted in-depth research on fundamental norms for the use of AI, diverse models of human-machine collaboration, the capability boundaries of AI, and new mechanisms for academic evaluation in philosophy and social science research, yielding numerous empirical findings and clear conclusions. The experiment has generated a tremendous social response, deepened academic discussions on relevant issues, and played a positive role in exploring new research paradigms in the age of intelligence.