Towards a Science of Collective AI: LLM-based Multi-Agent Systems Need a Transition from Blind Trial-and-Error to Rigorous Science

Recent advancements in Large Language Models (LLMs) have greatly extended the capabilities of Multi-Agent Systems (MAS), demonstrating significant effectiveness across a wide range of complex and open-ended domains. However, despite this rapid progress, the field still relies heavily on empirical trial-and-error. It lacks a unified and principled scientific framework necessary for systematic optimization and improvement. This bottleneck stems from the ambiguity of attribution: first, the absence of a structured taxonomy of factors leaves researchers restricted to unguided adjustments; second, the lack of a unified metric fails to distinguish genuine collaboration gain from mere resource accumulation. In this paper, we advocate for a transition to design science through an integrated framework. We advocate to establish the collaboration gain metric ($Γ$) as the scientific standard to isolate intrinsic gains from increased budgets. Leveraging $Γ$, we propose a factor attribution paradigm to systematically identify collaboration-driving factors. To support this, we construct a systematic MAS factor library, structuring the design space into control-level presets and information-level dynamics. Ultimately, this framework facilitates the transition from blind experimentation to rigorous science, paving the way towards a true science of Collective AI.

AgentXRay: White-Boxing Agentic Systems via Workflow Reconstruction

Large Language Models have shown strong capabilities in complex problem solving, yet many agentic systems remain difficult to interpret and control due to opaque internal workflows. While some frameworks offer explicit architectures for collaboration, many deployed agentic systems operate as black boxes to users. We address this by introducing Agentic Workflow Reconstruction (AWR), a new task aiming to synthesize an explicit, interpretable stand-in workflow that approximates a black-box system using only input--output access. We propose AgentXRay, a search-based framework that formulates AWR as a combinatorial optimization problem over discrete agent roles and tool invocations in a chain-structured workflow space. Unlike model distillation, AgentXRay produces editable white-box workflows that match target outputs under an observable, output-based proxy metric, without accessing model parameters. To navigate the vast search space, AgentXRay employs Monte Carlo Tree Search enhanced by a scoring-based Red-Black Pruning mechanism, which dynamically integrates proxy quality with search depth. Experiments across diverse domains demonstrate that AgentXRay achieves higher proxy similarity and reduces token consumption compared to unpruned search, enabling deeper workflow exploration under fixed iteration budgets.

Topology-Aware Revival for Efficient Sparse Training

Static sparse training is a promising route to efficient learning by committing to a fixed mask pattern, yet the constrained structure reduces robustness. Early pruning decisions can lock the network into a brittle structure that is difficult to escape, especially in deep reinforcement learning (RL) where the evolving policy continually shifts the training distribution. We propose Topology-Aware Revival (TAR), a lightweight one-shot post-pruning procedure that improves static sparsity without dynamic rewiring. After static pruning, TAR performs a single revival step by allocating a small reserve budget across layers according to topology needs, randomly uniformly reactivating a few previously pruned connections within each layer, and then keeping the resulting connectivity fixed for the remainder of training. Across multiple continuous-control tasks with SAC and TD3, TAR improves final return over static sparse baselines by up to +37.9% and also outperforms dynamic sparse training baselines with a median gain of +13.5%.

SwiftVLM: Efficient Vision-Language Model Inference via Cross-Layer Token Bypass

Visual token pruning is a promising approach for reducing the computational cost of vision-language models (VLMs), and existing methods often rely on early pruning decisions to improve efficiency. While effective on coarse-grained reasoning tasks, they suffer from significant performance degradation on tasks requiring fine-grained visual details. Through layer-wise analysis, we reveal substantial discrepancies in visual token importance across layers, showing that tokens deemed unimportant at shallow layers can later become highly relevant for text-conditioned reasoning. To avoid irreversible critical information loss caused by premature pruning, we introduce a new pruning paradigm, termed bypass, which preserves unselected visual tokens and forwards them to subsequent pruning stages for re-evaluation. Building on this paradigm, we propose SwiftVLM, a simple and training-free method that performs pruning at model-specific layers with strong visual token selection capability, while enabling independent pruning decisions across layers. Experiments across multiple VLMs and benchmarks demonstrate that SwiftVLM consistently outperforms existing pruning strategies, achieving superior accuracy-efficiency trade-offs and more faithful visual token selection behavior.

Small-Margin Preferences Still Matter-If You Train Them Right

Preference optimization methods such as DPO align large language models (LLMs) using paired comparisons, but their effectiveness can be highly sensitive to the quality and difficulty of preference pairs. A common heuristic treats small-margin (ambiguous) pairs as noisy and filters them out. In this paper, we revisit this assumption and show that pair difficulty interacts strongly with the optimization objective: when trained with preference-based losses, difficult pairs can destabilize training and harm alignment, yet these same pairs still contain useful supervision signals when optimized with supervised fine-tuning (SFT). Motivated by this observation, we propose MixDPO, a simple yet effective difficulty-aware training strategy that (i) orders preference data from easy to hard (a curriculum over margin-defined difficulty), and (ii) routes difficult pairs to an SFT objective while applying a preference loss to easy pairs. This hybrid design provides a practical mechanism to leverage ambiguous pairs without incurring the optimization failures often associated with preference losses on low-margin data. Across three LLM-judge benchmarks, MixDPO consistently improves alignment over DPO and a range of widely-used variants, with particularly strong gains on AlpacaEval~2 length-controlled (LC) win rate.

Interpreting Emergent Extreme Events in Multi-Agent Systems

Large language model-powered multi-agent systems have emerged as powerful tools for simulating complex human-like systems. The interactions within these systems often lead to extreme events whose origins remain obscured by the black box of emergence. Interpreting these events is critical for system safety. This paper proposes the first framework for explaining emergent extreme events in multi-agent systems, aiming to answer three fundamental questions: When does the event originate? Who drives it? And what behaviors contribute to it? Specifically, we adapt the Shapley value to faithfully attribute the occurrence of extreme events to each action taken by agents at different time steps, i.e., assigning an attribution score to the action to measure its influence on the event. We then aggregate the attribution scores along the dimensions of time, agent, and behavior to quantify the risk contribution of each dimension. Finally, we design a set of metrics based on these contribution scores to characterize the features of extreme events. Experiments across diverse multi-agent system scenarios (economic, financial, and social) demonstrate the effectiveness of our framework and provide general insights into the emergence of extreme phenomena.

AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security

The rise of AI agents introduces complex safety and security challenges arising from autonomous tool use and environmental interactions. Current guardrail models lack agentic risk awareness and transparency in risk diagnosis. To introduce an agentic guardrail that covers complex and numerous risky behaviors, we first propose a unified three-dimensional taxonomy that orthogonally categorizes agentic risks by their source (where), failure mode (how), and consequence (what). Guided by this structured and hierarchical taxonomy, we introduce a new fine-grained agentic safety benchmark (ATBench) and a Diagnostic Guardrail framework for agent safety and security (AgentDoG). AgentDoG provides fine-grained and contextual monitoring across agent trajectories. More Crucially, AgentDoG can diagnose the root causes of unsafe actions and seemingly safe but unreasonable actions, offering provenance and transparency beyond binary labels to facilitate effective agent alignment. AgentDoG variants are available in three sizes (4B, 7B, and 8B parameters) across Qwen and Llama model families. Extensive experimental results demonstrate that AgentDoG achieves state-of-the-art performance in agentic safety moderation in diverse and complex interactive scenarios. All models and datasets are openly released.

The Why Behind the Action: Unveiling Internal Drivers via Agentic Attribution

Large Language Model (LLM)-based agents are widely used in real-world applications such as customer service, web navigation, and software engineering. As these systems become more autonomous and are deployed at scale, understanding why an agent takes a particular action becomes increasingly important for accountability and governance. However, existing research predominantly focuses on \textit{failure attribution} to localize explicit errors in unsuccessful trajectories, which is insufficient for explaining the reasoning behind agent behaviors. To bridge this gap, we propose a novel framework for \textbf{general agentic attribution}, designed to identify the internal factors driving agent actions regardless of the task outcome. Our framework operates hierarchically to manage the complexity of agent interactions. Specifically, at the \textit{component level}, we employ temporal likelihood dynamics to identify critical interaction steps; then at the \textit{sentence level}, we refine this localization using perturbation-based analysis to isolate the specific textual evidence. We validate our framework across a diverse suite of agentic scenarios, including standard tool use and subtle reliability risks like memory-induced bias. Experimental results demonstrate that the proposed framework reliably pinpoints pivotal historical events and sentences behind the agent behavior, offering a critical step toward safer and more accountable agentic systems.

LogicEnvGen: Task-Logic Driven Generation of Diverse Simulated Environments for Embodied AI

Simulated environments play an essential role in embodied AI, functionally analogous to test cases in software engineering. However, existing environment generation methods often emphasize visual realism (e.g., object diversity and layout coherence), overlooking a crucial aspect: logical diversity from the testing perspective. This limits the comprehensive evaluation of agent adaptability and planning robustness in distinct simulated environments. To bridge this gap, we propose LogicEnvGen, a novel method driven by Large Language Models (LLMs) that adopts a top-down paradigm to generate logically diverse simulated environments as test cases for agents. Given an agent task, LogicEnvGen first analyzes its execution logic to construct decision-tree-structured behavior plans and then synthesizes a set of logical trajectories. Subsequently, it adopts a heuristic algorithm to refine the trajectory set, reducing redundant simulation. For each logical trajectory, which represents a potential task situation, LogicEnvGen correspondingly instantiates a concrete environment. Notably, it employs constraint solving for physical plausibility. Furthermore, we introduce LogicEnvEval, a novel benchmark comprising four quantitative metrics for environment evaluation. Experimental results verify the lack of logical diversity in baselines and demonstrate that LogicEnvGen achieves 1.04-2.61x greater diversity, significantly improving the performance in revealing agent faults by 4.00%-68.00%.

Teaching LLMs to Learn Tool Trialing and Execution through Environment Interaction

Equipping Large Language Models (LLMs) with external tools enables them to solve complex real-world problems. However, the robustness of existing methods remains a critical challenge when confronting novel or evolving tools. Existing trajectory-centric paradigms primarily rely on memorizing static solution paths during training, which limits the ability of LLMs to generalize tool usage to newly introduced or previously unseen tools. In this paper, we propose ToolMaster, a framework that shifts tool use from imitating golden tool-calling trajectories to actively learning tool usage through interaction with the environment. To optimize LLMs for tool planning and invocation, ToolMaster adopts a trial-and-execution paradigm, which trains LLMs to first imitate teacher-generated trajectories containing explicit tool trials and self-correction, followed by reinforcement learning to coordinate the trial and execution phases jointly. This process enables agents to autonomously explore correct tool usage by actively interacting with environments and forming experiential knowledge that benefits tool execution. Experimental results demonstrate that ToolMaster significantly outperforms existing baselines in terms of generalization and robustness across unseen or unfamiliar tools. All code and data are available at https://github.com/NEUIR/ToolMaster.