What Counts as Strategic Reasoning? A Systematic Mapping of Chess Research on Humans, Engines, and Language Models
研究通过系统映射84个核心研究,探讨了人类、传统引擎、神经和强化学习系统以及大语言模型在国际象棋中的策略推理,强调了状态表示、一般化及人机协作的重要性。
研究通过系统映射84个核心研究,探讨了人类、传统引擎、神经和强化学习系统以及大语言模型在国际象棋中的策略推理,强调了状态表示、一般化及人机协作的重要性。
本文通过49个指标开发了适用于意大利所有市镇的公平可持续福祉指数(MESWI),揭示了地区间福祉水平的显著差异,为地方政策提供信息支持。
Traditional abductive reasoning struggles to model interactions among hypotheses and risk-sensitive timing of commitments, limiting its applicability in high-stakes decision-making. This work proposes the first logical framework—denoted κ–τ—that explicitly integrates commitment-timing control by introducing hypothesis interaction parameters (κ) and normative commitment thresholds (τ), thereby distinguishing between “highly plausible” and “worthy of commitment.” The framework incorporates a dual-mode mechanism of synthesis and analysis to support governable reasoning. Implemented as a neurosymbolic architecture, it enables neural components to estimate cognitive parameters while allowing human agents to specify normative ones. A preliminary computational implementation demonstrates its potential to deliver transparent, auditable, and formally grounded abductive tools tailored for high-risk scenarios.
This work addresses the tendency of large language models (LLMs) to generate outputs lacking verifiable syntactic structure, often resulting in structural errors and hallucinations. The authors propose a neurosymbolic framework that, for the first time, dynamically aligns the incremental derivation mechanism of Combinatory Categorial Grammar (CCG) with the prefix-driven generation process of LLMs. By leveraging the Curry–Howard isomorphism, the approach lifts model outputs into typed compositional derivations. This enables unified structural reconstruction across both natural language and formal languages—including SQL, Solidity, and OWL—and incorporates a two-tier verification mechanism to enforce structural consistency and enable early detection of factual inaccuracies in generated content.
This work addresses the challenges of premature convergence and insufficient interpretability in human-AI collaboration under complex observations by proposing a non-greedy, risk-sensitive abductive reasoning framework. The approach leverages causal cluster structures and a dual-level κ-architecture (κ* and κ**) to enable accurate causal decomposition while avoiding misattribution. A novel κ–τ mechanism is introduced, wherein κ models cognitive interactions among competing hypotheses and τ dynamically adjusts commitment thresholds based on decision risk. Undetermined decompositions serve as shareable coordination artifacts to enhance transparency. Empirical validation in epidemic crisis and adversarial cyber threat scenarios demonstrates that the method generates multiple coexisting, evidence-supported explanatory pathways, thereby facilitating robust decision-making under ambiguity.
研究通过系统映射84个核心研究,探讨了人类、传统引擎、神经和强化学习系统以及大语言模型在国际象棋中的策略推理,强调了状态表示、一般化及人机协作的重要性。
本文通过49个指标开发了适用于意大利所有市镇的公平可持续福祉指数(MESWI),揭示了地区间福祉水平的显著差异,为地方政策提供信息支持。
Traditional abductive reasoning struggles to model interactions among hypotheses and risk-sensitive timing of commitments, limiting its applicability in high-stakes decision-making. This work proposes the first logical framework—denoted κ–τ—that explicitly integrates commitment-timing control by introducing hypothesis interaction parameters (κ) and normative commitment thresholds (τ), thereby distinguishing between “highly plausible” and “worthy of commitment.” The framework incorporates a dual-mode mechanism of synthesis and analysis to support governable reasoning. Implemented as a neurosymbolic architecture, it enables neural components to estimate cognitive parameters while allowing human agents to specify normative ones. A preliminary computational implementation demonstrates its potential to deliver transparent, auditable, and formally grounded abductive tools tailored for high-risk scenarios.
This work addresses the tendency of large language models (LLMs) to generate outputs lacking verifiable syntactic structure, often resulting in structural errors and hallucinations. The authors propose a neurosymbolic framework that, for the first time, dynamically aligns the incremental derivation mechanism of Combinatory Categorial Grammar (CCG) with the prefix-driven generation process of LLMs. By leveraging the Curry–Howard isomorphism, the approach lifts model outputs into typed compositional derivations. This enables unified structural reconstruction across both natural language and formal languages—including SQL, Solidity, and OWL—and incorporates a two-tier verification mechanism to enforce structural consistency and enable early detection of factual inaccuracies in generated content.
This work addresses the challenges of premature convergence and insufficient interpretability in human-AI collaboration under complex observations by proposing a non-greedy, risk-sensitive abductive reasoning framework. The approach leverages causal cluster structures and a dual-level κ-architecture (κ* and κ**) to enable accurate causal decomposition while avoiding misattribution. A novel κ–τ mechanism is introduced, wherein κ models cognitive interactions among competing hypotheses and τ dynamically adjusts commitment thresholds based on decision risk. Undetermined decompositions serve as shareable coordination artifacts to enhance transparency. Empirical validation in epidemic crisis and adversarial cyber threat scenarios demonstrates that the method generates multiple coexisting, evidence-supported explanatory pathways, thereby facilitating robust decision-making under ambiguity.