Ideology is a powerful tool for parties in armed conflicts, as it provides a source of motivation for combatants to stay in group under difficult circumstances and to perform actions that put them at risk or defy their personal ethical codes. But once in peacetime, besides the effects of past negative intergroup experiences, radical beliefs may become an obstacle to reconciliation and prolong the confrontation in the minds of ex-combatants. An examination of 484 recently decommissioned soldiers and insurgents in Colombia shows how the persistent ideological differences among former enemies help us explain postconflict intergroup bias beyond the effects of wartime victimization. We conclude that addressing the ideological radicalization that prolongs confrontation after armed conflict ceases is fundamental to creating proper conditions for reconciliation, and it offers a viable policy alternative to the much-needed healing from wartime-related trauma.

The Ming dynasty’s survival depended on locating and employing men with the ability to direct military forces, and contemporary observers were deeply concerned with the nexus of command, troop morale, and dynastic fighting capacity. This essay focuses on the years following the Tumu Crisis of 1449, a time when dynastic authorities were particularly alive to issues of military ability, and it draws on the perspectives of two men, the Minister of War, Yu Qian 于謙 (1398–1457), and another more junior official, Ye Sheng 葉盛 (1420–72). The essay offers a snapshot of how military ability was defined, cultivated, assessed, and rewarded. Further, it suggests that, read carefully, the writings of Ye Sheng and Yu Qian not only offer insight into the views of elite civil officials but also shed light, however faint and wavering, on military labor and working conditions for those who fought and commanded for a living.

The conventional literature suggests that the Chinese party-state has further strengthened social control and reinforced stability maintenance through expanded grassroots delegation. However, drawing on fieldwork interviews, government reports and media coverage, this article demonstrates that initiatives aimed at delegating power may actually weaken the government’s substantive responsiveness, thereby hindering the everyday management of disputes. The inherent tension of decentralization within a centralized political system leads to an uneven distribution of incentives and resources among agents at various levels. While more logistical powers (such as surveillance and mundane daily services) are allocated to grassroots governments, most decision-making and coercive powers (law enforcement and court rulings) remain in the hands of district-level functional departments. Grassroots officials are increasingly required to take broad responsibility for resolving citizen complaints, yet they face significant obstacles in mobilizing the relevant functional departments to address these issues. The reduced efficiency of problem-solving at the grassroots level not only increases the burden on grassroots bureaucrats to appease aggrieved citizens but also diminishes the effectiveness of initial efforts to contain routine grievances and prevent their escalation. This poses greater challenges for higher-level governments in balancing control and inclusivity, as well as in maintaining the legitimacy of state-sanctioned participatory institutions and the regime.

The early development of Artificial Intelligence (AI) in the latter half of the twentieth century was marked by limited, hand-crafted systems and fluctuating perceptions of the field’s potential. Early research explored a range of paradigms – including symbolic, neural and probabilistic approaches – constrained by severe hardware and data limitations. Key technological advances, such as the invention of microchips, GPUs and later TPUs, significantly enhanced computational capacity, enabling more complex AI experimentation. Concurrently, the proliferation of digital data through the internet addressed longstanding bottlenecks in data availability. The most transformative shift, however, came from architectural innovations in neural networks, culminating in the deep learning revolution. This unfolded in two phases: the emergence of Recurrent and Convolutional Neural Networks, followed by the development of transformer-based models, which underpin today’s Large Language Models (LLMs).