Large-scale group decision making, or LSGDM, is the formal discipline behind many high-stakes choices that involve twenty or more experts with different backgrounds, risk tolerances, and information. In emergencies, the stakes are extreme: hesitation or misjudgment can cost lives, yet forcing rapid agreement risks steamrolling minority opinions that may contain crucial insights. The traditional remedy is the consensus reaching process, an iterative loop in which experts whose views stray too far from the group are asked to adjust them until a preset consensus threshold is met. The trouble, the authors argue, is that the feedback parameters governing how strongly experts are pushed to compromise have historically been fixed by hand, based on experience. Too large a step and expert preferences get bulldozed; too small a step and the discussion drags on for rounds that an unfolding disaster cannot spare.

The core innovation is to model the entire consensus process as a Markov decision process and let a Deep Deterministic Policy Gradient, or DDPG, agent learn the feedback strategy. In this formulation, the state at each round captures the group’s evolving condition: the vector of opinion deviations between each expert cluster and the group center, the current group consensus level, the historical effect of previous feedback adjustments, and the trend in how quickly opinions are changing. Because these variables contain the essential statistics of the consensus environment, the next state depends only on the current state and the action taken, satisfying the Markov property that makes reinforcement learning well-posed. The action is deceptively simple: a single continuous value between zero and one, the feedback parameter that determines how far identified experts should move their opinions toward the group position.

What makes the problem genuinely hard is the reward. The authors designed a bi-objective, piecewise reward function that pays the agent for improving group consensus, penalizes it for each additional discussion round, and rewards it for preserving what they call the opinion retention degree, a measure of how much of each expert’s original judgment survives the adjustment. The result is an agent that learns to walk a narrow line: converge quickly, but not by erasing the diversity of expert judgment that makes large-group decisions robust in the first place. Sensitivity analysis over reward weight configurations confirmed the trade-off is real and tunable. A consensus-heavy weighting cut the rounds to agreement to 5.2 but dropped preference preservation to 0.6498, while a preference-oriented configuration raised preservation to 0.7235 at the cost of more rounds. The default weights of 0.5, 0.3, and 0.2 struck the best balance.

Before the learning agent ever sees a decision problem, the raw expert input must be made computable. Experts express preferences using hesitant fuzzy linguistic term sets, which allow a decision maker to offer several linguistic evaluations at once, for instance rating an alternative as both good and very good when uncertain. Conventional score functions for these sets can misjudge cases where a multi-element set and a single-element set yield identical scores. The team introduced a hesitation degree into the score function, quantifying how much uncertainty an expert’s linguistic expression carries, so that more hesitant judgments are weighted accordingly. In the case study, this refinement changed the final ranking of the middle alternatives, evidence that the hesitation information embedded in expert language genuinely matters to the outcome rather than being mathematical decoration.

Clustering twenty or more experts into manageable subgroups poses its own trap. Grouping by preference similarity alone ignores the trust relationships that determine who actually influences whom; grouping by trust alone can lump together mutually trusting experts who hold wildly divergent views. The method fuses both signals into a comprehensive relationship matrix, blending preference similarity and normalized trust values with a balance coefficient, then applies K-Means clustering with the number of clusters chosen by the silhouette coefficient. Sensitivity tests across coefficient values from 0.0 to 1.0 showed that equal weighting produced the best clustering, the fewest consensus rounds, and the highest final consensus level, while relying on either signal alone degraded performance significantly. Expert and cluster weights are then computed from network topology, in-degree and out-degree in the trust graph, and credibility, using a Hamacher t-norm to fuse scale and credibility weights.

The real-world test replayed the emergency decision scenario sparked by the 2023 rainstorm, in which 33 people died in Beijing and more than one million people across the region were evacuated. Twenty simulated experts evaluated four emergency alternatives, spanning headcounts and medical treatment, safe transfer of trapped residents, infrastructure repair, and water rescue with secondary disaster management, across four attributes including personnel search and rescue, resource allocation, public opinion management, and timeliness. The initial group consensus level came out at 0.8599, below the emergency threshold of 0.9. The DDPG agent then generated a sequence of feedback parameters, starting conservatively at 0.2195 and escalating round by round to 0.5946, reaching the threshold in six iterations with a final consensus level of 0.9043.

Notably, the attribute weights were not set by the experts themselves. The team crawled more than 50,000 Weibo posts about the rainstorm and applied the TF-IDF algorithm to extract keyword frequencies, matching terms like search and rescue, relief supplies, and rumor control to the decision attributes. Personnel search and rescue emerged as the heaviest weighted attribute at 0.3051, followed by public opinion management at 0.2542. This grounds the final ranking of alternatives in the aggregate concern of the affected public, an unusual touch of democratic signal in a mathematical procedure. The comprehensive scores ranked the fourth alternative, focused on water rescue and secondary disaster control, as optimal, and the refined score function preserved that top ranking while reshuffling the middle options compared with the original function.

Head-to-head comparisons against three benchmark paradigms, a fixed-parameter feedback method, a random small-step adjustment method without reinforcement learning, and a traditional linearly increasing consensus approach, showed the DDPG method reaching consensus in 6.2 rounds for 50 experts versus 8.4, 12.4, and 10.5 respectively, with the gap widening as groups grew to 200 experts. The random method did preserve slightly more of the experts’ original preferences, but needed roughly twice as many rounds, a cost the authors argue is unacceptable when disaster conditions are changing by the hour. Simulated experiments with 20 to 200 experts, and with uniform, normal, and multimodal preference distributions, all reached the 0.9 threshold within eight rounds, with preference preservation rates between 0.65 and 0.73, suggesting the approach generalizes beyond the showcase case.

The authors are candid about the model’s central assumption: that experts adjust their opinions gradually according to the feedback intensity the agent prescribes. Real experts accept, reject, negotiate, or partially adopt suggestions, and the team flags modeling those behaviors as the next frontier of human-machine collaborative decision making. Even so, the study marks a meaningful shift in how consensus might be engineered, from static rules tuned by intuition to a learned policy that reads the live state of a debating group and modulates its pressure in real time. If such agents mature, the quiet work of reconciling expert disagreement in the next flood, wildfire, or pandemic could begin not in a tense conference room, but inside a neural network that has already rehearsed thousands of arguments.

Subject of Research: Reinforcement learning-based adaptive feedback consensus for large-scale group emergency decision making

Article Title: Machine-assisted adaptive feedback consensus method for large-scale group decision making based on reinforcement learning

Article References: Yu, H., Xu, X., & Zhang, W. (2026). Machine-assisted adaptive feedback consensus method for large-scale group decision making based on reinforcement learning. Frontiers in Big Data, 9, Article 1921037. https://doi.org/10.3389/fdata.2026.1921037

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1921037

Keywords: reinforcement learning, DDPG, large-scale group decision making, consensus reaching, hesitant fuzzy linguistic term sets, emergency decision making, Markov decision process, social network analysis, K-Means clustering, TF-IDF, human-machine collaboration, disaster response

The core argument is deceptively simple but far-reaching. AI is transforming how the social sciences observe, model, simulate, and govern human systems, from financial markets to classrooms. Yet at the same time, some of AI’s most stubborn problems—interpretability, controllability, legitimacy, and value alignment—are fundamentally social and institutional questions, not merely technical ones. Why did a model reach a decision? Who is accountable when an autonomous agent acts? What values should an algorithm optimize, and who decides? These are questions that social science has grappled with for centuries, and the authors contend that answering them requires social scientific theory to be built into AI systems from the ground up.

To formalize this two-way relationship, the paper introduces a framework the authors call AI × SS, which unifies two complementary research directions under a single system architecture. The first direction, AI4SS, uses artificial intelligence to advance social science research: machine learning to detect patterns in massive behavioral datasets, natural language processing to analyze text at scale, and simulation to test hypotheses about collective behavior. The second direction, SS4AI, runs the other way, importing insights from sociology, economics, psychology, political science, and law to make AI systems more transparent, governable, and aligned with human values. The multiplication sign in the framework’s name is deliberate: the authors see the combination as producing something qualitatively new rather than a simple sum of the two fields.

A central theoretical pillar of the framework is Parallel Intelligence, or PI, an approach associated with Fei-Yue Wang’s long-running research program on complex systems. In the PI paradigm, real social systems are paired with artificial, computational counterparts—virtual societies, synthetic populations, and simulated institutions—that run in parallel with reality. Data flows between the real and artificial systems, allowing researchers to conduct experiments that would be impossible or unethical in the real world, to predict the consequences of policy interventions before deploying them, and to prescribe better decisions by learning from the parallel worlds. Generative AI and foundation models have dramatically strengthened this vision, because large models can now generate plausible synthetic agents, scenarios, and social dynamics at unprecedented scale and fidelity.

The review organizes the current research landscape into four methodological pathways. The first is data representation: how human behavior, institutions, and social relations are encoded so that machines can process them, ranging from network representations of societies to multimodal embeddings of text, images, and interaction traces. The second is analysis and modeling, where techniques such as graph learning, causal inference, and large language model-based agents are used to extract structure and explanation from social data. The third is simulation and experimentation, encompassing agent-based modeling, generative agent societies, and the construction of computational laboratories for testing social theories. The fourth is decision and optimization, where models are turned into actionable guidance for governance, resource allocation, and policy design under uncertainty.

Along each of these pathways, the authors survey representative studies that illustrate how the frontier is moving. In data representation, the explosion of digital trace data—from social media, mobility records, and transaction logs—has created both opportunity and risk, demanding new methods for privacy preservation and bias correction. In analysis and modeling, foundation models pretrained on vast corpora are being repurposed as social simulators, capable of reproducing survey responses, opinion dynamics, and market behavior, though the authors caution that their fidelity and faithfulness remain open empirical questions. In simulation, generative agents that plan, remember, and interact are enabling synthetic societies in which researchers can stress-test interventions against thousands of virtual lives before touching the real world.

The paper then grounds these abstractions in concrete domains. In finance, AI-driven models now read markets, news flows, and investor sentiment in real time, while social science supplies the institutional understanding of regulation, trust, and systemic risk. In communication, recommendation algorithms and large language models are reshaping public discourse, and computational social science is racing to measure effects such as polarization and misinformation. In management, agentic AI systems are beginning to participate in organizational decision-making, raising questions about delegation, oversight, and human–machine collaboration that draw directly on organizational theory. In art, generative models are transforming creative production and raising questions of authorship and value. In education, adaptive AI tutors promise personalization at scale, but their legitimacy and fairness depend on pedagogical and ethical frameworks that only the social sciences can supply.

What emerges across all of these domains, the authors argue, is a new research structure rather than a loose collection of applications. Institutions, governance mechanisms, and human–machine collaboration are not afterthoughts bolted onto AI systems; they are constitutive components of any AI that operates in society. An autonomous agent acting in a market, a classroom, or a city is embedded in norms, incentives, and legal regimes, and its behavior cannot be predicted or controlled without modeling those social layers. Conversely, the arrival of agentic AI—systems that pursue goals, coordinate with other agents, and act over long horizons—creates genuinely novel social phenomena, such as economies of interacting algorithms, that demand new social scientific theory in their own right.

The review also reflects an unusually broad collaborative effort. The author team spans institutions including Macau University of Science and Technology, the Chinese Academy of Sciences, Beijing Jiaotong University, Tianjin University, the University of Science and Technology of China, Zhejiang Lab, Beijing Institute of Technology, the University of Chinese Academy of Sciences, and the University of Glasgow, and grew out of a series of workshops organized under the Artificial Intelligence for Autonomous Science initiative between 2023 and 2026, covering topics from social transportation to intelligent art and parallel decision theaters. That breadth mirrors the paper’s central claim: no single discipline owns the AI–society problem anymore.

For researchers and policymakers alike, the practical message is that the next generation of breakthroughs will come from teams that treat AI and the social sciences as a single coupled system. Building interpretable models requires theories of explanation that humans actually find meaningful. Building controllable agents requires institutional design, incentive analysis, and accountability structures. Building value-aligned systems requires explicit, contestable accounts of whose values are being encoded. The AI × SS framework offers a map of that territory, organizing data, modeling, simulation, and decision-making into a coherent architecture in which artificial and human systems learn from each other continuously. If the authors are right, the most important new social science of the coming decade may be the one that is co-written with machines.

Subject of Research: The bidirectional co-evolution of artificial intelligence and the social sciences through a unified AI × SS framework based on parallel intelligence

Article Title: Ai and social sciences: new AI, new studies, and new social sciences

Article References: Ni, Q., Lin, F., Zhang, H., Xue, X., Zhang, T., Lü, L., Zhang, B., Lu, Y., Li, D., Yang, L., Ye, P., Li, X., Guo, C., Huang, J., Zheng, X., Yu, H., & Wang, F.-Y. (2026). Ai and social sciences: new AI, new studies, and new social sciences. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11693-5

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11693-5

Keywords: artificial intelligence, social sciences, parallel intelligence, AI4SS, SS4AI, computational social science, agentic AI, generative AI, foundation models, value alignment, human-machine collaboration, AI governance