Does generative AI mean the “end of history” for pharmacovigilance automation? towards a framework for the future of human-AI systems

Possibility of using artificial intelligence models in routine pharmacovigilance processes

Introduction  . Artificial intelligence (AI) has undergone rapid development in pharmacovigilance (PV), evolving from experimental application to being considered a key tool in daily practice. Relatively simple AI models, including statistical signal detection methods, have been used in PV for decades, while recent advances in Natural Language Processing (NLP) have significantly expanded the scope of potential applications. Objective  . This article provides a critical analysis of the potential of NLP systems to optimize routine PV tasks, taking into account data protection requirements. The application of semantic search AI models based on alternative architectural approaches, specifically embedding models and retrieval-augmented generation (RAG), is examined separately.  Main points  . The authors distinguish between processes that do not involve personal data and allow the use of open AI solutions (searching and systematizing scientific literature, generating publication summaries), and processes involving the handling of confidential information (e. g., data extraction from Individual Case Safety Reports (ICSRs), automated generation of clinical case descriptions, benefit-risk analysis, and compliance with regulatory requirements and reporting standards), which require the use of corporate AI systems deployed within a secure infrastructure. The article also discusses limitations, risks, practical implementation aspects, as well as issues of ensuring the reliability, reproducibility, and transparency of the solutions used. Conclusion  . AI models, particularly NLP models, have significant potential for integration into routine PV processes. However, successful integration of AI models is impossible without a systematic approach to managing associated risks. Subject to these conditions, AI can become a robust tool for increasing the efficiency of PV processes.

An Overview of Explainable Artificial Intelligence (XAI) and Its Application

This assessment paper emphasise about newb technology of Explainable Artificial Intelligence (XAI) is an emerging and vital field of research that addresses the "black box" problem prevalent in modern machine learning. As AI systems become more complex and integrated into high-stakes domains such as healthcare, finance, and criminal justice, their inherent opacity raises critical concerns regarding transparency, trust, and accountability. The primary goal of XAI is to provide methods and techniques that enable human users to understand, interpret, and appropriately trust the decisions and predictions made by AI algorithms. While XAI provides a powerful framework for responsible AI development, challenges such as the performance-interpretability trade-off, lack of standardized evaluation metrics, and potential for human misinterpretation remain areas of active research. Ultimately, XAI is a critical step toward creating a symbiotic relationship between humans and AI, where intelligent systems operate not just with high performance but with ethical and transparent reasoning.

Baseline review of Formal Methods and Foundations of Artificial Intelligence

. Artificial Intelligence (AI) is advancing rapidly, yet many successful models remain opaque and provide limited assurance about reliability, safety, and failure modes. This motivates renewed interest in formal methods and foundational perspectives that can support trustworthy AI beyond empirical testing. This paper presents a baseline review of the selected papers volume of the inaugural International Conference on Formal Methods and Foundations of Artificial Intelligence (FMF-AI 2025), published as Annales Mathematicae et Informaticae , Vol. 61 (2025). The goal is twofold: (i) to map and summarize the first FMF-AI “snapshot” as a starting point for the Hungarian research ecosystem, and (ii) to define a reproducible baseline that can serve as a reference for measuring topical and methodological shifts in subsequent FMF-AI editions. The review clusters the twenty selected papers into five thematic groups and records their relative prevalence. In addition, it introduces simple baseline metrics that can be recomputed in future FMF-AI editions to observe structural changes in the research landscape. The main pattern is a clear imbalance between verification-oriented contributions and papers that primarily use AI methods in application or optimization contexts.

Intelligent Automation Improved Efficiency in Pharmacovigilance Safety Signal Assessment

Pharmacovigilance is the science involving the detection, assessment, understanding, and prevention of adverse events associated with drugs, biologics, or medical devices. In pharmacovigilance, information that suggests a new potential causal relationship between an intervention and an adverse event is called a safety signal. Following their detection, safety signals are assessed via a comprehensive, structured analysis to more fully elucidate whether a correlation exists. This key process often requires the manual assessment of many individual case safety report (ICSR) narratives to extract meaningful information in a labor‐intensive, time‐consuming, and variability‐prone manner. In this retrospective feasibility study, we describe the potential utility of an intelligent automation system leveraging the GPT‐4o large language model to automate the extraction of case elements of interest from a series of ICSR narratives while maintaining human expert oversight. Our proprietary platform allowed users to extract the presence or absence of risk factors and responses to dechallenge and rechallenge via instructional prompts built on a common template structure. Case elements for five historical signal assessments were selected based on need for and feasibility of artificial intelligence extraction. Performance ranged from F1 = 0.444 to 1.000 for risk factors and from F1 = 0.429 to 0.909 for responses to dechallenge and rechallenge. Even when considering the need to verify GPT‐4o outputs for accuracy, potential time savings were identified. To the best of our knowledge, our results are the first to demonstrate an intelligent automation platform that may streamline signal management workflows using a machine‐first, human‐verified operational workflow while maintaining regulatory compliance.

Causal Post-hoc XAI: Categorisation and Systematic Literature Review

Today, the use of Artificial Intelligence (AI) is rapidly increasing in many areas of society. While model performance on various tasks continue to impress, it does so at the cost of increased model complexity, such that most state-of-the-art AI models are effectively black boxes. Where human-made decisions typically are accompanied by human-understandable explanations detailing the reasoning behind the decision, incorporating advanced AI as part of a decision-making process reduces the transparency of that process significantly. Yet, the ability to explain decisions is essential for there to be understanding and trust. As a response to this, Explainable Artificial Intelligence (XAI) has emerged as a field that aims to provide explanations of model behaviour. Methods categorised as post-hoc are designed to generate explanations for black box models after training, at no cost to model performance. In parallel with this, extensive work has been done in the field of causality to formalise the structure of human-understandable, causal explanations. This work presents a comprehensive literature review of the current state of the subfield of XAI that consist of causality-motivated post-hoc XAI methods. In order to clearly define causal XAI, a causal framework for categorising XAI is introduced, and three types of post-hoc XAI methods are identified: observational methods, internally causal methods and externally causal methods. Finally, externally causal XAI is argued a promising direction for reliable and understandable post-hoc XAI, with the ability to generate counterfactual explanations using a meaningful vocabulary, in line with the definition of counterfactual used in causal theory.

Explanation-Driven Visual Interventions in Genomics