This chapter explores the emergence of agentic AI in architectural design – systems that move beyond generative assistance to act as active collaborators. Building on the transition from expert to learning systems and from retrieval to generation, it proposes Creative AI as a framework integrating intuition and reasoning across multi-agent, multimodal networks. Unlike traditional generative models that interpolate within data, agentic systems negotiate constraints, critique outcomes, and evolve strategies through cooperative and adversarial interactions. Drawing from projects such as Gaudí + Neural Networks, DeepHimmelblau, and Semantic Universes, the chapter illustrates how distributed, node-based ecologies of specialiszed and generalist agents can co-create, evaluate, and refine architectural ideas. This evolution redefines authorship: the architect shifts from directing tools to orchestrating intelligences, guiding dialogues among human and machine agents. Agentic AI thus marks a new paradigm in design – one of negotiation, reflection, and shared creativity across human--machine networks.
This dataset comprises empirical survey responses collected from 315 smartphone users in Indonesia to examine the behavioral and psychological determinants influencing the continuous use of Generative AI (GenAI) features on mobile devices. The dataset captures both respondent profiles and multi-item measurement scales adapted from extended technology acceptance and post-adoption continuance frameworks. The profile variables encompass core demographic attributes, such as age and gender, alongside mobile platform preferences (Android and iOS), usage frequency, and specific GenAI application modalities utilized by respondents, including text generation, summarization, proofreading, and image creation. The structural evaluation items employ a standard 5-point Likert scale to operationalize key theoretical constructs: Perceived Usefulness (PU1–PU4), Perceived Ease of Use (PEOU1–PEOU4), Confirmation (CONF1–CONF3), Satisfaction (SAT1–SAT4), Attitude towards Success (ATS1–ATS3), Attitude towards Process (ATP1–ATP3), Trust (TR1–TR3), Perceived Intelligence (PI1–PI4), and Continuance Intention (CI1–CI3). This comprehensive structure makes the data suitable for structural equation modeling (SEM), partial least squares (PLS-SEM), and multivariate statistical analysis exploring human-AI interaction dynamics in consumer technology.
Naufal R Pratama· Zenodo (CERN European Organi...· 0 citations
Swing the Otter is a proposed black-box behavioral audit for generative AI systems. It evaluates whether a system’s semantic position on a fixed proposition changes in proportion to changes in independently verifiable evidence. The protocol separates source-derived evidence, user assertions, and model self-report; includes neutral and leading-pressure conditions, fresh-session replication, and retrieval-manifest controls; and introduces diagnostic constructs including Evidence-Normalized Oscillation (ENO) and Unsupported Flip Count (UFC). The paper explicitly treats these constructs as proposed methods requiring empirical calibration rather than validated universal metrics.
Andrew Paul Roebuck· Zenodo (CERN European Organi...· 0 citations
Generative Artificial Intelligence is reshaping marketing practice, yet little is known about whether Large Language Models (LLMs) can produce email templates that are directly editable inside campaign builders. This study reports a technical feasibility evaluation of AI-generated email templates compatible with the Stripo template builder, conducted with a hospitality CRM partner. Few-Shot Prompting and Fine-Tuning of an LLM were compared on the generation of Stripo-compatible, non-standard HTML components and of complete templates. Fine-Tuning outperformed Few-Shot Prompting in Stripo compatibility (88%vs. 69%, p = 0.002), generation speed, and cost per output. End-to-end template generation proved feasible but not production-ready, confirming that deployment requires human review of every campaign-critical element. The study serves as an evaluation protocol for platform compatibility of AI-generated email HTML, providing an empirical comparison of adaptation strategies for this niche code-generation task and design implications for CRM vendors and hospitality marketing teams.
Maria Inês Mendes, Nuno António, Sérgio Guerreiro· International Journal of Inf...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
This paper presents the Non-Euclidean Geometry Synchronization Algorithm (NESA), a novel algorithm designed to automatically generate and validate geometric theorems within complex non-Euclidean geometries. The core aim is to provide a verifiable framework for mathematical proof and abstraction, moving beyond manual construction and leveraging the generative capabilities of a deep learning model. NESA utilizes a trained generative AI model to construct geometric constructions and proofs, creating a dynamically evolving system capable of discovering novel theorems. We detail the algorithm's architecture, training methodology, and the resulting validation process, emphasizing the potential for automated theorem generation and rigorous mathematical verification. The system's design incorporates a mechanism for continuous refinement based on established geometric principles, ensuring the generated theorems are demonstrably valid.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces M.E.S.H., a gestural musical instrument that reclaims human agency and physical engagement in AI-assisted music creation. While large generative text-to-audio models often restrict individual artistic expression, M.E.S.H. utilizes neural networks in Max MSP as a co-creative mediator between bodily intent and sound. By placing the performer’s body at the center of the workflow, our research-creation project wishes to establish a novel relationship between biotechnology and algorithmic generation, one that actively confronts the ethical implications of biological data use within the corporate technology ecosystem. We evaluate the system across creative and ethical axes, using a live performance as one case study to demonstrate the dynamic negotiation of agency between musician and AI. Ultimately, we are sharing this framework as open software so that other artists can reappropriate their biometric data to create music that is highly personalized and physically engaged with the assistance of a small AI system.
Generative AI increasingly mediates human cognitive work by automating substantial portions of textual production, analysis, reformulation, and decision support. Much of the literature on prompt engineering, prompt literacy, and evaluative judgement has focused on how to formulate effective requests and how to inspect or revise generated outputs. Recent work has begun to establish that framing before generation is itself an important competence. This article develops a more specific question within that emerging space: what happens when the categories through which a cognitive request is framed are treated as objects of critical inquiry rather than as neutral components of a “good prompt”?The article proposes frame opacity as an epistemic asymmetry between a realised output and the unrealised alternative framings that were available before generation. It introduces ex ante evaluative judgement as the capacity to examine and justify the epistemic constitution of a request before delegation, and develops categorial decompression as an instructional operation organised around eight dimensions: role, operation, object, genre, criterion, context, addressee, and purpose. A cross-cutting category of demonstrative material is also examined because examples, templates, and exemplars can silently define what counts as an acceptable response.The framework is explicitly sociocritical. Drawing on Freirean problem-posing and praxis, Habermasian justification and communicative rationality, and critical educational theory, it treats prompting as a site in which authority, relevance, standards, and responsibility can become naturalised or contested. The article distinguishes descriptive specificity from epistemic specification and proposes a non-monotonic gradient in which greater detail is not inherently better. It concludes with nine falsifiable propositions linking framing, delegation, exemplars, metacognitive calibration, cognitive residue, and adjudication. The paper reports no original empirical study; its contribution is a conceptual synthesis designed to generate testable research.
Adalberto Hernández Santos, Lisett D. Páez Cuba· Zenodo (CERN European Organi...· 0 citations
This CETaS Briefing Paper explores the potential of generative artificial intelligence (GenAI) in creating malicious software and is aimed at informing risk management and supporting the AI-cybersecurity evaluation community. The cybersecurity field is divided, with some fearing GenAI could lead to novel threats while others believe it merely automates existing malicious code. Despite the release of GPT-4 in March 2023 there has been no noticeable increase in novel malware detected. GenAI currently lacks the capabilities to independently create operational malware and autonomously identify and exploit vulnerabilities, but its future impact on cybersecurity could be profound, especially as models and datasets improve, leading to potential scenarios where AI-created malware and AI-based defence systems continuously evolve. Effective use of GenAI in cybersecurity requires leveraging its strengths in pattern recognition and natural language processing, and fostering collaboration between AI and cybersecurity communities to ensure cyber defence stays ahead of the AI-enabled game.
Sarah Mercer, Tim Watson· Alan Turing Institute Resear...· 0 citations
This paper explains why organizations that adopt the same generative AI/LLM technology end up with different results. It argues the real driver isn't the technology itself but "appropriation" — how employees actually use it day to day, which can differ from how it was intended. Using Adaptive Structuration Theory (DeSanctis & Poole, 1994) — originally built for older, simpler group technologies — the paper adapts it to fit modern AI's more flexible, unpredictable, conversational nature, and connects it to how organizations choose to automate, augment, or redesign work around AI. It offers a model, seven research propositions, a plan for future testing, and practical guidance for managing AI use responsibly.
Likhitha Macha· Zenodo (CERN European Organi...· 0 citations
Smart speakers may support independent living among individuals with visual disabilities, yet evidence on everyday use remains limited. This study examined user experience among 79 participants (41 visually impaired, 38 sighted) over 7 days of naturalistic in-home Amazon Echo use. Interviews, analyzed through content analysis, yielded five themes: accessibility, effectiveness, enjoyment, efficiency, and privacy. Participants with visual disabilities valued voice-based autonomy and emotional engagement, describing Alexa as an “e-friend,” but encountered speech recognition errors. Sighted participants questioned information accuracy relative to generative AI systems and reported setup difficulties, with limited emotional engagement. Privacy concerns were more pronounced among participants with visual disabilities. Findings indicate that smart speaker design should balance these five dimensions, with implications for speech recognition robustness, onboarding, and conversational personalization.
Vincent Ogunmwonyi, Hyung Nam Kim· Proceedings of the Human Fac...· 0 citations
This work presents twenty conceptual formulas for examining how a civilization can transform energy, matter, knowledge, attention, technology, and human relationships into lasting capabilities that support life. The term “White-Hole Civilization” is used as a systems metaphor rather than as a claim about the physical existence or engineering of astrophysical white holes. It describes a civilization that does not merely consume and concentrate resources, but continually converts its inputs into accessible energy, material circulation, basic-needs security, individual autonomy, distributed resilience, knowledge inheritance, and future possibilities. The twenty formulas are organized into five interconnected domains: the foundations of survival; production and abundance; cognition and human–AI co-evolution; institutions and everyday life; and long-term civilizational continuity. They explore subjects including usable energy flow, material circularity, household self-reliance, resilient networks, technological learning curves, post-scarcity thresholds, anti-monopoly structures, automation dividends, attention and meaning, knowledge preservation, bounded freedom, grounded warmth, minimal sufficient governance, peaceful system transition, future seeding, relational networks, and intergenerational handoff. These formulas are not proposed as immutable natural laws or experimentally validated predictive equations. They function as transparent and revisable design languages for identifying bottlenecks, hidden dependencies, externalized costs, concentration of power, failure boundaries, and long-term trade-offs. The central proposition is that an advanced civilization should not be evaluated solely by how much energy or technology it controls, but by how effectively it transforms its capabilities into dignity, resilience, freedom, diversity, and future choice—without consuming the conditions that allow life and goodness to continue.
政恩 馮· Zenodo (CERN European Organi...· 0 citations
Artificial intelligence (AI) is increasingly reshaping higher education, influencing how students access knowledge, engage with learning tasks, and regulate their learning processes. While existing research has largely focused on efficiency and performance, comparatively limited attention has been given to how AI affects metacognitive processes such as planning, monitoring, and evaluation. This study addresses this gap through a qualitative systematic literature review with thematic synthesis of peer-reviewed studies published between 2018 and 2026, supplemented by foundational theoretical work on metacognition. The findings indicate that AI does not exert a uniform effect on metacognition. Instead, its impact is dynamic, context-dependent, and shaped by the interaction between technological, pedagogical, and socio-economic conditions. AI can enhance metacognition by externalising cognitive processes and supporting feedback, but it can also enable cognitive offloading, reducing learners’ engagement in regulation. A third mode, hybrid regulation, emerges in which metacognitive processes are co-constructed between learners and AI systems. These findings are particularly relevant to AI-supported and technology-rich learning environments, including STEM education contexts where intelligent tutoring systems and generative AI tools increasingly mediate learning processes. The study proposes a conceptual framework that positions AI along a continuum from support to offloading, shaped by contextual conditions. The findings reframe AI as a system that reorganises, rather than simply enhances or erodes, metacognition in higher education.
Thasmai Dhurumraj, Celeste Labuschagne· Frontiers in Education· 0 citations
A new method, called CW-Net, translates the reasoning process of an autonomous vehicle’s AI system into understandable concepts that explain its behavior.
