Generative AI (GenAI) presents significant opportunities, particularly for creative work like visual design. GenAI can effectively address creative challenges like the “blank page” problem by enabling rapid visual conceptualisation, thereby enhancing productivity in visual design tasks. However, organisations face significant challenges in integrating off-the-shelf, untamed foundation models, whose complex user interfaces are misaligned with production-oriented workflows, exacerbating issues such as AI illiteracy, employee resistance, and job displacement. To facilitate GenAI integration into creative workflows while addressing these challenges, we conduct an action design research study at Ubisoft, where we develop and introduce a GenAI-enabled creativity support tool (CST)—comprising workflow guidance and simplified user interface on Stable Diffusion models—as a bridging tool in concept art creation processes. Our user studies and evaluations indicate that the artefact increases user acceptance, improves productivity, and adds value to both concept artists and their studios. In addition to detailing the artefact’s design and evaluation, we present seven design principles for text-to-image GenAI CSTs for visual design. These principles contribute to the academic discourse on human-AI collaboration by filling a critical gap in prescriptive design knowledge and emphasising the role of GenAI in augmenting, rather than replacing, human skill and ingenuity.

This paper presents a structured review of explainable stream reasoning with Graph Neural Networks (GNNs) in settings where graph structures and background knowledge evolve over time. Although prior work has advanced GNN modeling, temporal reasoning, and Knowledge Graph (KG) integration, the literature remains fragmented regarding how explanations should be generated, semantically grounded, and evaluated in knowledge-enriched graph streams. Existing studies provide limited guidance on how explanations should align with ontologies, preserve relational coherence, and remain stable under temporal evolution. As a review article, this paper provides a taxonomy-driven synthesis of GNN explanation methods, semantic integration patterns, and stream-oriented constraints rather than a new benchmark or deployment system. It analyzes how established explanation families apply to evolving, knowledge-enriched graphs and systematizes KG integration strategies for reasoning and explanation, including embedding-based, message-passing, neuro-symbolic, and KG-assisted approaches. The review also presents an implementation-oriented architectural roadmap illustrating how semantic constraints can be incorporated into temporal message passing. Building on this synthesis, the paper proposes knowledge-aware evaluation dimensions, including semantic fidelity, relational coherence, temporal semantic stability, and human/domain-centered alignment. Where possible, these dimensions are accompanied by illustrative formal metric definitions, while their standardization remains open. A bounded empirical feasibility illustration on temporally ordered healthcare graph data demonstrates the computability of selected dimensions. Representative use cases in healthcare, security, social media, and autonomous systems further illustrate how the proposed perspective can guide future research on trustworthy and semantically grounded GNN-based stream reasoning.

The emergence of dynamic, high-volume data streams demands advanced reasoning frameworks to capture complex spatio-temporal relationships that are essential for enabling contextual understanding. However, current approaches often lack scalable and adaptable semantic representations in dynamic and spatio-temporal scenarios. To answer this need, we introduce a novel Spatio-Temporal Knowledge approach based on Graph Neural Networks (STKGNN) for activity recognition. This framework performs graph-based reasoning over semantically enriched Spatio-Temporal Knowledge Graphs (STKGs) constructed from open-source video datasets. By leveraging these custom STKGs, we propose three advanced Graph Neural Network (GNN) based architectures to recognize various activities. Accordingly, we establish a comprehensive approach for spatio-temporal reasoning that adapts to diverse Knowledge Graph structures by addressing adaptability, scalability, and temporal complexities. This framework enhances activity recognition and provides a foundation for wider dynamic or real-time applications in different domains including healthcare, autonomous systems, video surveillance, and various other fields.

Large language models (LLMs) are increasingly used for social-science simulations, yet most evaluations target task optimality rather than the variability and adaptation characteristic of human decision-making.We propose a process-oriented evaluation framework with progressive interventions (Intrinsicality, Instruction, and Imitation), and apply it to two classic economics tasks: the second-price auction and the newsvendor inventory problem.By default, LLMs adopt stable, conservative strategies that diverge from observed human behavior.Giving LLMs risk-framed instructions makes them behave more like humans.However, this also causes complex irregularities.Incorporating human decision trajectories via in-context learning further narrows distributional gaps, indicating that models can absorb human patterns.However, across all interventions, LLMs underexpress round-toround variability relative to humans, revealing a persistent alignment gap in behavioral fidelity.Future evaluations of LLM-based social simulations should prioritize processlevel realism.Our code and data are available here

Real-time video streams present unique challenges for continual learning systems, demanding models that can incrementally update representations, preserve past knowledge, and reason over complex semantic relationships without sacrificing efficiency. In this paper, we introduce CAST-GNN, the first unified Graph Neural Network architecture expressly designed for continual adaptation on streaming Spatio-Temporal Knowledge Graphs (STKGs) derived from open-source video benchmarks. CAST-GNN integrates dynamic temporal embedding layers, adaptive self-attention, episodic graph pattern memory, and a novel hybrid selective replay buffer with Fisher-based regularization and knowledge distillation to mitigate catastrophic forgetting. Through comprehensive experiments on four diverse STKG benchmarks (UCF-101, HMDB-51, Kinetics-400, and SomethingSomething), our model achieves 96-97% accuracy, between 0.13-0.31 % forgetting, by consistently outperforming re-implemented continual-learning baselines under identical conditions. Ablation studies confirm the critical synergy between temporal embeddings and adaptive attention. We further demonstrate XAI-driven interpretability by aligning global distributional shifts with local node-level attributions. CAST-GNN not only advances robust semantic reasoning and knowledge retention but also provides a scalable, explainable framework applicable to a wide array of real-world streaming scenarios.

Artificial intelligence (AI) applications have proliferated, garnering significant interest among information systems (IS) scholars.AI-powered analytics, promising effective and low-cost decision augmentation, has become a ubiquitous aspect of contemporary organisations.Unlike traditional decision support systems (DSS) designed to support decisionmakers with fixed decision rules and models that often generate stable outcomes and rely on human agentic primacy, AI systems learn, adapt, and act autonomously, demanding recognition of IS agency within AI-augmented decision making (AIADM) systems.Given this fundamental shift in DSS; its influence on autonomy, responsibility, and accountability in decision making within organisations; the increasing regulatory and ethical concerns about AI use; and the corresponding risks of stochastic outputs, the extrapolation of prescriptive design knowledge from conventional DSS to AIADM is problematic.Hence, novel design principles incorporating contextual idiosyncrasies and practice-based domain knowledge are needed to overcome unprecedented challenges when adopting AIADM.To this end, we conduct an action design research (ADR) study within an e-commerce company specialising in producing and selling clothing.We develop an AIADM system to support marketing, consumer engagement, and product design decisions.Our work contributes to theory and practice with a set of actionable design principles to guide AIADM system design and deployment.

Abstract The rapid expansion of decentralized, complex streaming data across diverse domains such as the Internet of Things, healthcare, and smart cities presents significant technical challenges. These challenges–data heterogeneity (integration of diverse formats and sources), dynamicity (handling real-time data evolution), and high-volume throughput (efficient processing of large, rapidly arriving data)–are the central focus of this study and are examined in depth. To address these critical issues necessitates advanced methods capable of seamless integration, effective real-time reasoning, and continuous learning from heterogeneous streaming data, thus enhancing real-time decision-making capabilities. This study provides an extensive review of existing research at the intersection of streaming data, machine learning, and reasoning. The literature review categorizes Stream Reasoning approaches into three key groups: Streaming Machine Learning, Streaming Linked Data, and Streaming Knowledge Graphs. Each category is critically examined in terms of strengths, limitations, ongoing challenges, and future opportunities identified in recent studies. Additionally, potential integrative solutions that leverage Knowledge Graph structures and advanced Stream Reasoning techniques are highlighted, illustrating how state-of-the-art modeling methods can effectively address Stream Reasoning related challenges. The analysis concludes that combining Knowledge Graph and Machine Learning approaches significantly enhances the capability to manage and overcome complex Stream Reasoning challenges.

Learning a new language can be challenging. To help learners, we built a recommendation system that suggests texts and videos based on the learners’ skill level of the language and topic interests. Our system analyzes content to determine its difficulty and topic, and, if needed, can simplify complex texts while maintaining semantics. Our work explores the holistic use of Large Language Models (LLMs) for the various sub-tasks involved for accurate recommendations: difficulty estimation and simplification, graph recommender engine, topic estimation. We present a comprehensive evaluation comparing zero-shot and fine-tuned LLMs, demonstrating significant improvements in French content difficulty prediction (18-56%), topic prediction accuracy (27%), and recommendation relevance (up to 18% NDCG increase).

Generative artificial intelligence (AI) presents large risks for society when it is used to create fake news. A crucial factor for fake news to go viral on social media is that users share such content. Here, we aim to shed light on the sharing behavior of users across human-generated vs. AI-generated fake news. Specifically, we study: (1) What is the perceived veracity of human-generated fake news vs. AI-generated fake news? (2) What is the user's willingness to share human-generated fake news vs. AI-generated fake news on social media? (3) What socio-economic characteristics let users fall for AI-generated fake news? To this end, we conducted a pre-registered, online experiment with N= 988 subjects and 20 fake news from the COVID-19 pandemic generated by GPT-4 vs. humans. Our findings show that AI-generated fake news is perceived as less accurate than human-generated fake news, but both tend to be shared equally. Further, several socio-economic factors explain who falls for AI-generated fake news.

We study the complexity of the bribery problem with distance restrictions. In particular, in the bribery problem, we are given an election and a distinguished candidate p, and are asked whether we can make p win/not win the election by bribing at most k voters to recast their votes. In the bribery problem with distance restrictions, we require that the votes recast by the bribed voters are close to their original votes. To measure the closeness between two votes, we adopt the prevalent Kendall-Tau distance and the Hamming distance. We achieve a wide range of complexity results for this problem under a variety of voting correspondences, including the Borda, Condorcet, Copelandαfor every 0≤α≤1 and Maximin.

We study the computational complexity of control problems for two-stage voting rules. An example of a two-stage voting rule is the Black's procedure. The first stage of the Black's procedure selects the Condorcet winner if one exists; otherwise, in the second stage the Borda winner is selected. The computational complexity of the manipulation problem of two-stage voting rules has recently been studied by Narodytska and Walsh [20] and Fitzsimmons et al. [14]. Extending their work, we consider the control problems for similar scenarios, focusing on constructive control by adding or deleting votes, denoted as CCAV and CCDV, respectively.