As large language models (LLMs) increasingly act as autonomous agents in markets and organizations, their behavior in strategic environments becomes economically consequential. We document that off-the-shelf LLM agents exhibit systematic deviations from payoff-sensitive behavior in canonical economic games, including excessive cooperation and limited responsiveness to incentives. We introduce a supervised fine-tuning approach that aligns agent behavior with explicit economic preferences. Specifically, we generate optimal strategies under two stylized utility specifications, homo economicus, which maximizes self-interest, and homo moralis, which incorporates Kantian universalizability, and use these utility-implied reasoning and strategies to guide fine-tuning. Fine-tuning on a small, theory-driven synthetic dataset induces persistent and interpretable shifts in strategic behavior. In applications to moral dilemmas and repeated duopoly pricing, agents aligned to different preference structures produce systematically distinct equilibrium outcomes and pricing dynamics. These results frame AI alignment in multi-agent settings as an objective-design problem and illustrate how economic theory can guide the design of strategically coherent AI agents.
Elena Moral Sánchez, Martin Campos Pinto, Yaman Güçlü
et al.
In this work we propose a high-order structure-preserving discretization of the cold plasma model which describes the propagation of electromagnetic waves in magnetized plasmas. By utilizing B-Splines Finite Elements Exterior Calculus, we derive a space discretization that preserves the underlying Hamiltonian structure of the model, and we study two stable time-splitting geometrical integrators. We approximate an incoming wave boundary condition in such a way that the resulting schemes are compatible with a time-harmonic / transient decomposition of the solution, which allows us to establish their long-time stability. This approach readily applies to curvilinear and complex domains. We perform a numerical study of these schemes which compares their cost and accuracy against a standard Crank-Nicolson time integrator, and we run realistic simulations where the long-term behaviour is assessed using frequency-domain solutions. Our solvers are implemented in the Python library Psydac which makes them memory-efficient, parallel and essentially three-dimensional.
As military organisations consider integrating large language models (LLMs) into command and control (C2) systems for planning and decision support, understanding their behavioural tendencies is critical. This study develops a benchmarking framework for evaluating aspects of legal and moral risk in targeting behaviour by comparing LLMs acting as agents in multi-turn simulated conflict. We introduce four metrics grounded in International Humanitarian Law (IHL) and military doctrine: Civilian Target Rate (CTR) and Dual-use Target Rate (DTR) assess compliance with legal targeting principles, while Mean and Max Simulated Non-combatant Casualty Value (SNCV) quantify tolerance for civilian harm. We evaluate three frontier models, GPT-4o, Gemini-2.5, and LLaMA-3.1, through 90 multi-agent, multi-turn crisis simulations across three geographic regions. Our findings reveal that off-the-shelf LLMs exhibit concerning and unpredictable targeting behaviour in simulated conflict environments. All models violated the IHL principle of distinction by targeting civilian objects, with breach rates ranging from 16.7% to 66.7%. Harm tolerance escalated through crisis simulations with MeanSNCV increasing from 16.5 in early turns to 27.7 in late turns. Significant inter-model variation emerged: LLaMA-3.1 selected an average of 3.47 civilian strikes per simulation with MeanSNCV of 28.4, while Gemini-2.5 selected 0.90 civilian strikes with MeanSNCV of 17.6. These differences indicate that model selection for deployment constitutes a choice about acceptable legal and moral risk profiles in military operations. This work seeks to provide a proof-of-concept of potential behavioural risks that could emerge from the use of LLMs in Decision Support Systems (AI DSS) as well as a reproducible benchmarking framework with interpretable metrics for standardising pre-deployment testing.
In this study, machine learning models were tested to predict whether or not a customer of an insurance company would purchase a travel insurance product. For this purpose, secondary data provided by an open-source website that compiles databases from statistical modeling competitions were used. The dataset used presents approximately 2,700 records from an unidentified company in the tourism insurance sector. Initially, the feature engineering stage was carried out, which were selected through regularized models: Ridge, Lasso and Elastic-Net. In this phase, gains were observed not only in relation to dimensionality, but also in the maintenance of interpretative capacity, through the coefficients obtained. After this process, five classification models were evaluated (Random Forests, XGBoost, H2O GBM, LightGBM and CatBoost) separately and in a hybrid way with the previous regularized models, all these stages using the k-fold stratified cross-validation technique. The evaluations were conducted by traditional metrics, including AUC, precision, recall and F1 score. A very competitive hybrid model was obtained using CatBoost combined with Lasso feature selection, achieving an AUC of 0.861 and an F1 score of 0.808. These findings motivate us to present the effectiveness of using hybrid models as a way to obtain high predictive power and maintain the interpretability of the estimation process
Identifying human morals and values embedded in language is essential to empirical studies of communication. However, researchers often face substantial difficulty navigating the diversity of theoretical frameworks and data available for their analysis. Here, we contribute MoVa, a well-documented suite of resources for generalizable classification of human morals and values, consisting of (1) 16 labeled datasets and benchmarking results from four theoretically-grounded frameworks; (2) a lightweight LLM prompting strategy that outperforms fine-tuned models across multiple domains and frameworks; and (3) a new application that helps evaluate psychological surveys. In practice, we specifically recommend a classification strategy, all@once, that scores all related concepts simultaneously, resembling the well-known multi-label classifier chain. The data and methods in MoVa can facilitate many fine-grained interpretations of human and machine communication, with potential implications for the alignment of machine behavior.
Natalie Tirabassi, Sathish A. P. Kumar, Sumit Jha
et al.
We propose MORAL (a multimodal reinforcement learning framework for decision making in autonomous laboratories) that enhances sequential decision-making in autonomous robotic laboratories through the integration of visual and textual inputs. Using the BridgeData V2 dataset, we generate fine-tuned image captions with a pretrained BLIP-2 vision-language model and combine them with visual features through an early fusion strategy. The fused representations are processed using Deep Q-Network (DQN) and Proximal Policy Optimization (PPO) agents. Experimental results demonstrate that multimodal agents achieve a 20% improvement in task completion rates and significantly outperform visual-only and textual-only baselines after sufficient training. Compared to transformer-based and recurrent multimodal RL models, our approach achieves superior performance in cumulative reward and caption quality metrics (BLEU, METEOR, ROUGE-L). These results highlight the impact of semantically aligned language cues in enhancing agent learning efficiency and generalization. The proposed framework contributes to the advancement of multimodal reinforcement learning and embodied AI systems in dynamic, real-world environments.
W artykule przedstawiono projekt diecezjalnego programu budzenia powołań kapłańskich. Sytuuje się go w kontekście radykalnego spadku liczby kandydatów do sakramentu święceń. Niniejsze rozważania nie sprowadzają się jedynie do
przedstawienia technik pastoralnych, mających na celu zachęcenie młodych ludzi do otwarcia serc na Boże wezwanie. Prezentują przede wszystkim wagę ponownego odkrycia piękna i wartości dwóch powiązanych sakramentów: Eucharystii i kapłaństwa sakramentalnego. Z głębokich analiz teologicznych nad tymi sakramentami oraz refleksji duchowej nad nimi może zrodzić się entuzjazm wspólnoty wierzących odnośnie
do troski o powołania kapłańskie. Entuzjazm ten może prowadzić do nowego zaangażowania diecezjalnej wspólnoty kapłańskiej, wspólnot parafialnych i żyjących w nich rodzin, a także ludzi dotkniętych chorobą, ofiarujących swoje cierpienia w intencji powołań. Zaangażowania te, przekute na konkretne inicjatywy pastoralne (zaczerpnięte z dotychczasowych doświadczeń polskich i amerykańskich), być może otworzą serca ludzi młodych na przyjęcie daru powołania kapłańskiego. Najważniejszy wniosek z niniejszych analiz jest następujący: troska o powołania to działania szerokie, czyli diecezjalne, długodystansowe, trwające wiele lat, a przede wszystkim pogłębione, zbudowane na teologii sakramentów ustanowionych w Wieczerniku podczas Ostatniej Wieczerzy przez Pana Jezusa.
Daniel Fleming, James Keenan, SJ, Hans Zollner, SJ
There is a certain urgency about this volume, which is not often reflected in works of theology or theological ethics. The sheer scale of the undermining of human dignity through sexual abuse that has occurred within the Church asks questions of these disciplines and scholars within them: to what extent have we been blind to these issues? Why have our efforts in theology and theological ethics been so slow to wrestle with this crisis? How are theology and theological ethics implicated in the crisis? And how might the disciplines be constructive in responding? In this volume, we encounter a diverse range of scholars from all around the world wrestling with these and other questions.
İslâm düşüncesinde farklı alanlarda disiplinlerin/ilimlerin teşekkülüne müteakip disiplinlerin/ilimlerin kendilerine özgü kavramları ve terminolojileri de oluşmaya başlamış ve daha sonraki süreçte bu kavramsallaştırmaya katkı sunacak müstakil tanım (ḥudûd) risâleleri telif edilerek her disiplindeki/ilimdeki kavramın anlam ve kullanım çerçevesi daha da netleştirilmiştir. Temel olarak kullanılan kavramın ıstılahî anlamının belirlenmesi ve böylece takip edilen kelâmî temayülün desteklenmesi adına, kelâm ilmi özelinde de bu türe ait eserler kaleme alınmıştır. Kelâm ilmi özelinde bu alana en büyük katkıyı ise şüphesiz Mu‘tezilî kelâmcılar vermiştir. Nitekim başta Kādî Abdülcebbâr (ö. 415/1025) olmak üzere öğrencileri Ebû Reşîd en-Nîsâbûrî (ö. 420/1029’dan sonra) ve İbn Şervîn (V./XI. yüzyılın ilk yarısı) gibi isimler tarafından kelâmî kavramları ihtiva eden özgün risâleler telif edilmiştir. Zeydî kelâmcı ve fakih Ebü’l-Kāsım el-Büstî’nin (ö. 420/1029 civarı) henüz bir nüshasına ulaşılamayan Tehẕîbü’l-ḥudûd’u da bahsedilen eserlerle eş zamanlıdır. Bu gelenek, İmâmî-Mu‘tezilî çizgide Ebû Ca‘fer et-Tûsî’nin (ö. 460/1067), kurgusu ve yer yer yaptığı açıklamalar itibariyle mûcez bir kelâm eseri hüviyeti de taşıyan el-Muḳaddime fi’l-medḫal ilâ ṣınâʿati ʿilmi’l-kelâm’ı ve Şerîf el-Murtazâ’ya (ö. 436/1044) nispet edilen el-Ḥudûd ve’l-ḥaḳāʾiḳ isimli eserde yansımasını bulmuştur. Ebû Ca‘fer el-Mukrî en-Nîsâbûrî’nin (ö. VI./XII. yy.’ın ortaları [?]) el-Ḥudûd’u ise bu gelenek içerisindeki en hacimli örnek olarak göze çarpmaktadır.
Mu‘tezilî gelenek içerisinde mevcut eserler arasında ilki olması açısından bu türe ait en önemli eser, yakın zamana kadar kayıp olduğu düşünülen ve tarafımızdan keşfedilen bir mecmua içerisinde, önemli bazı diğer Zeydî-Mu‘tezilî eserlerle birlikte bulunan, Kādî Abdülcebbâr’ın Ḥudûdü’l-elfâẓ isimli risâlesidir. Genel olarak kelâmî kavramların yer aldığı ve kısmen fıkha ve mezhepler tarihine ilişkin kavramların da bulunduğu bu risâlede Kādî Abdülcebbâr dile dayalı bir tanım teorisi ortaya koymuştur. Buradaki yaklaşımı, el-Muğnî’sinde tanım ve tanımlamaya yönelik ortaya koyduğu görüşlerle uyumludur. Nitekim klasik mantıkta beş tümel üzerinden yürütülen tanım (ḥad) ve betim (resm) yerine, ele alınan kavramın manasına mutabık ve ondan daha açık olan bir lafız üzerinden yapılan lügavî tanım metodunu benimsemiş ve bu doğrultuda ele aldığı kavramın kısa bir tanım ve açıklamasına yer vermiştir. Bu durum, önceki Mu‘tezilî kelâmcılar tarafından felsefeye mesafeli yaklaşmaları nedeniyle Aristotelesçi tanım anlayışının alternatifi olarak kullanılan dile dayalı tanım anlayışının, Kādî Abdülcebbâr tarafından da tevarüs edildiğini açıkça göstermektedir. Ebû Reşîd en-Nîsâbûrî’nin el-Ḥudûd’u gibi Kādî’yi takip eden dönem bazı kaynaklarda görülen yaklaşım da onun etkisini imlemektedir.
21st Century war is increasing in speed, with conventional forces combined with massed use of autonomous systems and human-machine integration. However, a significant challenge is how humans can ensure moral and legal responsibility for systems operating outside of normal temporal parameters. This chapter considers whether humans can stand outside of real time and authorise actions for autonomous systems by the prior establishment of a contract, for actions to occur in a future context particularly in faster than real time or in very slow operations where human consciousness and concentration could not remain well informed. The medical legal precdent found in 'advance care directives' suggests how the time-consuming, deliberative process required for accountability and responsibility of weapons systems may be achievable outside real time captured in an 'advance control driective' (ACD). The chapter proposes 'autonomy command' scaffolded and legitimised through the construction of ACD ahead of the deployment of autonomous systems.
To introduce the volume, Andrea Vicini, SJ, stresses the need to address the global cancer pandemic, while the world struggles with the ongoing pandemic caused by COVID-19. Cancer affects millions of people. It is the first or second cause of death in 134 countries, the leading cause of death in most high-income countries (i.e., 10 million deaths in 2020), and the leading cause of death by disease in American children. Cancer is also unjust. Striking inequities can be traced within and between countries in cancer incidence and survival by race, ethnicity, and socio-economic-status. Survival is much higher among the wealthy than among the poor. An overview of the volume introduces the following chapters.
Customer churn prediction is a valuable task in many industries. In telecommunications it presents great challenges, given the high dimensionality of the data, and how difficult it is to identify underlying frustration signatures, which may represent an important driver regarding future churn behaviour. Here, we propose a novel Bayesian hierarchical joint model that is able to characterise customer profiles based on how many events take place within different television watching journeys, and how long it takes between events. The model drastically reduces the dimensionality of the data from thousands of observations per customer to 11 customer-level parameter estimates and random effects. We test our methodology using data from 40 BT customers (20 active and 20 who eventually cancelled their subscription) whose TV watching behaviours were recorded from October to December 2019, totalling approximately half a million observations. Employing different machine learning techniques using the parameter estimates and random effects from the Bayesian hierarchical model as features yielded up to 92\% accuracy predicting churn, associated with 100\% true positive rates and false positive rates as low as 14\% on a validation set. Our proposed methodology represents an efficient way of reducing the dimensionality of the data, while at the same time maintaining high descriptive and predictive capabilities. We provide code to implement the Bayesian model at https://github.com/rafamoral/profiling_tv_watching_behaviour.
This paper presents the design process and empirical evaluation of a new tool for enabling ethics by design: The Moral-IT Cards. Better tools are needed to support the role of technologists in addressing ethical issues during system design. These physical cards support reflection by technologists on normative aspects of technology development, specifically on emerging risks, appropriate safeguards and challenges of implementing these in the system. We discuss how the cards were developed and tested within 5 workshops with 20 participants from both research and commercial settings. We consider the role of technologists in ethics from different EU/UK policymaking initiatives and disciplinary perspectives (i.e. Science and Technology Studies (STS), IT Law, Human Computer Interaction (HCI), Computer/Engineering Ethics). We then examine existing ethics by design tools, and other cards based tools before arguing why cards can be a useful medium for addressing complex ethical issues. We present the development process for the Moral-IT cards, document key features of our card design, background on the content, the impact assessment board process for using them and how this was formulated. We discuss our study design and methodology before examining key findings which are clustered around three overarching themes. These are: the value of our cards as a tool, their impact on the technology design process and how they structure ethical reflection practices. We conclude with key lessons and concepts such as how they level the playing field for debate; enable ethical clustering, sorting and comparison; provide appropriate anchors for discussion and highlighted the intertwined nature of ethics.
Valerio Capraro, Ismael Rodriguez-Lara, Maria J. Ruiz-Martos
Recent work highlights that cooperation in the one-shot Prisoner's dilemma (PD) is primarily driven by moral preferences for doing the right thing, rather than social preferences for equity or efficiency. By contrast, little is known on what motivates cooperation in the Stag-Hunt Game (SHG). Cooperation in the SHG fundamentally differs from cooperation in the PD in that it is not costly, but risky: players have no temptation to deviate from the cooperative outcome, but cooperation only pays off if the other player cooperates. Here, we provide data from a large (N=436), pre-registered, experiment. Contrary to what has been observed for the PD, we find that SHG cooperation is primarily driven by preferences for efficiency, rather than preferences for doing the right thing.
This article presents and discusses recent Norwegian research on Bible Didactics, using the Jarle and Karina Waldemars’ Norwegian version of Carolyn Larsen’s book Princess Stories. Real Bible Stories of God’s Princesses as a starting point and case. The emphasis of the discussion is on theology, gender and learning discourses, and is being done in relation to a certain field of practice, namely the Christian Education Reform of the Church of Norway. The author shows how the curriculum/learning plan of Chrisitian Eduation seems to carry a bricolaged and accidental understanding of Bible Didactics. Recent Norwegian research on Bible Didactic, mostly done by Old Testament scholars, is being discus-
sed. The research in question shows how a moral, instrumental didactisism often dominates bible mediation for children. Our case is an excellent example of how this is gendered differently for girls and boys. The methodology of the book used as a case also relies upon an understanding as learning as acquisition less than participation and knowledge creation.
Keywords: Bible, didactics, gender, theology, learning, children
Nøkkelord: Bibel, didaktikk, kjønn, teologi, læring, barn