Murat Sariyar

Scarce medical images often prompt data augmentation. We ask when it earns its opportunity cost under a complete information budget. Across four collections, classical and GAN candidates are compared with real-only training, full-pool exposure, score adjustment and direct label use. Transformations reduce within-patient variation; source mismatch and between-patient uncertainty remain. Fixed 50-model ResNet50 ensembles with ranked classical candidates and probability averaging outperform equally adjusted real-only and full-pool ensembles at five sizes. We evaluate procedures in 50 paired draws at 50, 100, 150, 200 and 250 image-training patients. The design counts 200 regularization patients plus disjoint 100-patient selection and adjustment sets. Evaluation includes 38,743 CheXpert patients. At 250 patients, guidance reduces maximum group cross-entropy by 0.0442 versus real-only training (simultaneous 95% interval, 0.0048–0.0836). Adding the 100 selection patients to each original training cohort outperforms using their labels only to guide augmentation at 50–200 patients. Thus, use independent labels first when cohorts are smallest; consider selected augmentation once patient evidence, strong representation and aggregation are available. The 250-patient result marks the upper tested setting, not a universal threshold. AUROC remains stable while sensitivity and specificity shift, requiring operating-point review.

Medical large language models are evaluated through factuality, physician preference, source support, retrieval quality, safety, and clinical utility. These measures describe content and performance while leaving unresolved whether a generated claim meets the requirements for a specified clinical use. This article proposes clinical warrant as a use-specific judgment for a particular output–use pair. Clinical use includes reliance within a patient-related encounter, task, or care pathway, including education that can shape understanding or later action. The proposed use and the standards used to judge it are documented separately. For action-guiding outputs, clinical appropriateness is a gateway. Five further conditions examine role, case information, evidence, preserved limits, and authority/accountability. The framework asks whether these requirements are met at the point of proposed reliance, records failed or unresolved conditions and the resulting disposition, and tests whether a relevant change stops or redirects use. Fifteen constructed examples across five domains distinguish established, not established, and indeterminate warrant.

Background: Removing explicit protected health information (PHI) does not necessarily make a clinical narrative non-identifiable. Rare diagnoses, social circumstances, locations, temporal patterns, and treatment trajectories may still permit re-identification. Methods: This PRISMA-ScR-informed scoping review combined PubMed/MEDLINE, OpenAlex, and targeted NLP searches with OpenAlex title-and-abstract, Web of Science, and Scopus sensitivity checks. Eligible publications addressed clinical or biomedical text de-identification from 1 January 2018 to 31 March 2026. Results: The searches yielded 275 records and 258 unique records after deduplication. Eligibility assessment produced 110 core records, 11 background or review records, and 22 exclusions. All 143 eligibility-stage records underwent independent double screening and reconciliation; structured extraction covered all 110 core records. Abstract screening of 47 additional Scopus candidates found no task function or implementation paradigm outside the proposed classification. The resulting two-axis framework separates a method’s role in the workflow from its technical implementation. Conclusions: Clinical text de-identification cannot be reduced to named-entity recognition. Local hybrid or transformer pipelines remain the most defensible baseline for routine PHI detection. LLMs are better suited to defined tasks in augmentation, transformation, and assurance, with local validation and explicit control of data exposure.

Healthcare processes combine stable workflow segments with open, case-dependent work. This makes the use of large language models (LLMs) and more autonomous AI systems less a question of technical capability alone than of orchestration: different degrees of AI autonomy fit different degrees of process structure. Based on a conceptual review, this paper develops a BPMN–CMMN lens for distinguishing between structured tasks that support tightly bounded LLM-guided execution and case-dependent segments in which more autonomous coordination may become relevant. The literature suggests that LLMs fit best in documentation, summarization, communication, referral support, and other bounded workflow steps, whereas more autonomous systems become more plausible where progression depends on changing context, exceptions, and discretionary judgment. The main challenge lies in evaluation. Classical process indicators remain necessary, but they become harder to interpret once work is distributed across humans and AI. In addition to operational KPIs, healthcare organizations need measures that capture orchestration quality, supervision burden, context carryover, and the stability of outcomes across variable cases.

Introduction: Large Language Models (LLMs) are often framed through metaphors such as “bullshit” or “stochastic parrots,” emphasizing missing grounding, belief, or intention. While rhetorically powerful, these framings obscure how LLMs are used for sense-making, ideation, and communication. We reframe LLMs as Operators for General Cognitive Shortcuts (GECOS) within techno-semiotic assemblages. Methods: We develop a functional model by integrating concepts from Luhmannian systems theory, Deleuzian ontology, and minimally from Husserlian phenomenology. Using conceptual analysis as functional–comparative synthesis, we analyze human–LLM interaction without attributing agency, belief, or understanding to the model. Results: GECOS explains LLM usefulness as communicative complexity reduction: models generate connectable continuations by approximating second-order expectations (“what is expected to be expected”), enabling interactional continuity without reference to truth or intention. Via Luhmann’s contingency formula, LLMs help users navigate uncertainty through procedurally plausible coherence. Discussion: The framework shifts attention from ontological debates about “understanding” to the operational role of LLMs in distributed sense-making. It also highlights risks: overreliance, emotional projection, and normative flattening when connectability substitutes for justification. Conclusion: GECOS offers a non-anthropomorphic alternative to deficit metaphors by modeling LLMs as pragmatic operators that sustain communicative momentum and enable workable continuations in complex socio-technical environments.

Purpose: This paper reinterprets “bullshit” according to Harry Frankfurt using Luhmannian systems theory. Instead of treating bullshit as a moral failing or a defect in speaker intention, we analyze it as a communicative form that keeps communication going, especially when incompatible expectations have to be met. Design/methodology/approach: We develop a functional, systems-theoretical reconstruction that looks at observable patterns in communication rather than inner motives. To avoid moralizing vocabulary, we introduce the term performative decoupling (PD) for cases where value-laden talk is only loosely linked to concrete rules, procedures, or verification. We retain the term “bullshit” only when engaging with existing philosophical literature. Findings: PD is proposed as a neutral label for utterances that sound meaningful and inclusive but postpone or minimize binding commitments. PD emerges where functionally differentiated subsystems (politics, science, economy and organizations) generate conflicting demands, while the tools that would tightly couple values to norms and rules are missing or contested. In such settings, PD preserves connectivity: communication can continue without forcing premature, exclusionary decisions. The same mechanism, however, can become destructive when semantic promises are not followed by any tightening of couplings over time, leading to persistent gaps between talk and implementation. Social implications: The framework suggests a non-moral way of evaluating public and organizational discourse: not by asking whether speakers are “authentic”, but by examining whether value-heavy communication is followed by clear targets, timelines, responsibilities and verification. It also points to design levers, such as cheaper audits, reversible commitments and institutionalized contestation, that can turn PD from a dead end into a bridge between values and rules. Originality/value: The paper shifts the analysis of bullshit from sincerity and intention to semantics, coupling and connectivity. It shows how PD can be observed and graded without moral diagnosis by tracking the relation between value language and concrete commitments. This makes it possible to distinguish temporary, functional decoupling (as a buffer) from entrenched, credibility-eroding forms.

This article asks what grounds novelty once temporal passage no longer carries metaphysical priority. It approaches this question through a parallax comparison of Ibn Arabi and Gilles Deleuze. At its centre lies the problem of productivity: what makes the new real when time is no longer conceived as a self-moving medium? A brief heuristic reference to selected interpretations of modern physics helps suspend the intuitive privilege of a moving present and sharpen the problem. The article then develops two distinct ontologies of renewal. In Ibn Arabi, novelty is grounded in continuous divine actualisation: each moment receives existence through ongoing bestowal, though immersion in the world covers over this dependence and makes renewal appear as self-sustaining continuity. In Deleuze, novelty is grounded in an immanent temporality structured by contraction, the pure past, and the caesural force of the future, even as habit and recognition can absorb the event back into the familiar and blunt the force of the new. Read together, these accounts show that transcendence and immanence are not rival doctrines added to a shared ontology of time, but competing accounts of how renewal and transformation become real. The comparison further suggests that immanence can account for novelty only by including within itself an irreducible virtual dimension that cannot be reduced to the actual, thereby complicating any simple opposition between immanence and transcendence. The article closes by drawing out the ethical and religious stakes of these ontologies for freedom, responsibility, practice, and memory.

Background: Synthetic data generation (SDG) has emerged as a critical enabler for data-driven healthcare research, offering privacy-preserving alternatives to real patient data. Temporal health data – ranging from physiological signals to electronic health records (EHRs) – pose unique challenges for SDG due to their complexity, irregularity, and clinical sensitivity. Objective: This review systematically examines SDG methods for longitudinal and time-series health data. Its aims are to (1) propose a lightweight taxonomy to support orientation across the SDG landscape along five structural dimensions, (2) characterize the major synthesis techniques and their alignment with temporal structures and data modalities, and (3) synthesize the utility and privacy evaluation strategies used in practice. Methods: A systematic literature review was conducted following PRISMA guidelines across four major databases (ACM, arXiv, IEEE Xplore, Europe PMC) for publications from 2017 to 2025. Eligible studies proposed or applied SDG techniques to healthcare-relevant temporal data with sufficient methodological transparency. Structured data extraction and thematic analysis were used to identify modeling trends, evaluation metrics, and domain-specific requirements, complemented by a comparative synthesis of SDG methods. Results: A total of 115 studies were included. Deep generative models – especially Generative Adversarial Networks (GANs), Autoencoders (AEs), and diffusion-based methods – dominate the field, with increasing adoption of autoregressive and hybrid simulation approaches. Event-based EHR data are most commonly targeted, while continuous and irregular time series remain underexplored. Utility evaluations vary widely, with strong emphasis on descriptive statistics and predictive performance, but limited attention to inferential validity and clinical realism. Privacy assessments are sparse and inconsistently reported: only 30% of studies included any metric, and just around 6% implemented differential privacy (DP), often without parameter disclosure. This limited adoption may reflect technical challenges, limited expertise, and the absence of regulatory incentives. Conclusions: Synthetic temporal data play an increasingly vital role across clinical prediction, public health modeling, and Artificial Intelligence (AI) development. However, SDG research remains fragmented in terminology, evaluation practices, and privacy safeguards. Responsible-AI considerations – such as fairness, transparency, and trust – along with evidence on clinical adoption remain underexplored but are critical for future integration. This review provides a unified conceptual and methodological framework to guide future research, standardization efforts, and interdisciplinary collaboration for responsible, effective use of synthetic health data.

Debates in bioethics and healthcare governance repeatedly confront communicative practices that smooth over uncertainty, ritualise consent or protect institutions more than they inform patients and professionals. These practices can be labelled ‘bullshit’, but the term is morally charged and analytically imprecise. This paper develops a diagnostic for truth-indifferent communication in healthcare that does not depend on inferring inner motives. By juxtaposing Frankfurt-style, Cohen-style and a Luhmannian approach, it shows how a systems-theoretical lens is especially useful for thinking through ‘bullshit’ as a structural and organisational phenomenon rather than an individual moral failure. Drawing on this perspective, the analysis focuses on observable proxies for truth-indifference: verification latency, incentive alignment, format and placement (eg, slogans or ritual reassurances), and the strength of sanction coupling. Informed consent serves as the central case for demonstrating how formalisation and nudging can both discipline and foster truth-indifferent communication. The result is a second-order, non-moralising framework that distinguishes functional from dysfunctional truth-indifferent communication and translates this distinction into concrete design and governance levers for bioethics and clinical practice.