Andrea Capiluppi

While AI tools continue to evolve and are increasingly incorporated in creative domains as tools and companions, humans are becoming less accurate at distinguishing AI-generated outputs from human-generated ones in artistic domains such as visual art, poetry, and creative writing. In this context, understanding the human-centred aspects of human-AI interaction has become as important as assessing computational ability. This systematic review investigates human-centred perspectives on how AI outputs are differentiated from human-like creations in artistic domains through Turing Test-like experimental paradigms. Following the PRISMA 2020 guidelines, we searched IEEE Xplore, PubMed, Scopus, and Web of Science for empirical studies published between 2014 and 2025. From 3,461 records, 31 studies were included after screening and snowballing. We applied thematic analysis to examine methods, metrics, evaluation practices, problematizations, research gaps, and future directions. Risk of bias was assessed using the JBI Checklist, and certainty of evidence was evaluated through GRADE-CerQual. Our thematic analysis revealed several key metrics, including authorship detection, emotional reaction, aesthetic judgment, creativity, agency, empathy, attention, liking/preference, technical qualities, and value judgments. Most studies relied on self-reports and behavioral measures, with limited mixed-methods work. Gaps remain with respect to embodiment and experience, and with human-AI co-creativity. Our review underscores the importance of mixed-methods, human-centred approaches for advancing ethical and collaborative AI systems.

The current increase in the use of GenAI has raised concerns about human involvement, AIs acceptability and detectability. Although current studies stressed the importance of human involvement, most studies employing Turing Test-like approaches focus on this aspect within the human-in-the-loop (HITL) versus human-out-of-the-loop (HOTL) frameworks. In this study we go beyond the HITL vs. HOTL dichotomy and hypothesize that the outputs that are generated with different levels of interaction evoke different responses along with the information regarding the human and AI involvement in the making. We use behavioral methods and self-reports to measure the participants' reactions towards artworks that are generated with three different levels of interaction with an AI tool. We focused on different aspects of experiencing artworks through various metrics including liking, perceived beauty, meaning, creativity, self-relevance and empathy towards the author and the emotional valence with respect to the human involvement in the making the and the information that is disclosed about it. Our analyses revealed a quadratic-shaped effect on the ways humans experience artworks generated via different levels of interaction; where the outputs generated with highest interaction are similar to the human-only/no interaction and outputs that are generated with low and moderate levels of interaction are similar to one another. Disclosing the information regarding the human involvement also had an impact on the emotional valence. Our study has implications for understanding the level of acceptance of human and AI involvement in the co-creative process and contributes to the growing literature on developing ethical human-centred AI systems.

Self-Admitted Technical Debt (SATD) refers to technical compromises explicitly admitted by developers in natural language artifacts, such as code comments, commit messages, and issue trackers. Among its types, Architecture Technical Debt (ATD) is particularly difficult to detect due to its abstract and context-dependent nature. Manual annotation of ATD is costly, time-consuming, and challenging to scale. To reduce labeling effort, this study combines keyword-based filtering, active learning, and explainable AI for ATD detection. We refined an existing dataset of ATD-related Jira issues to obtain an expert-validated seed set used to extract representative keywords. These keywords were then applied to identify more than 103k candidate issues across 10 open-source projects. To assess the reliability of keyword-based filtering, we qualitatively evaluated a statistically representative sample of labeled issues. Building on the resulting dataset, we applied active learning with multiple query strategies to prioritize informative samples for annotation. The results show that Breaking Ties achieved the best performance, with an F1-score of 0.72 and a 49% reduction in annotation effort. To improve transparency, we used SHAP and LIME to explain ATD classification results. Expert evaluation showed that both methods provided useful explanations, with LIME generally preferred for its clarity and ease of use.

Technical debt (TD) refers to the long‐term costs associated with suboptimal design or code decisions in software development, often made to meet short‐term delivery goals. SATD occurs when developers explicitly acknowledge these trade‐offs in the codebase, typically through comments or annotations. SATD detection has become an increasingly important research area, particularly with the rise of learning‐based techniques that aim to streamline SATD detection. This SLR provides a comprehensive analysis of SATD detection approaches published between 2014 and early 2025, focusing on the evolution of techniques from heuristic‐based techniques to more advanced ML, DL, and Transformer‐based models. It examines key trends in SATD detection methodologies and tools, evaluates the effectiveness of different approaches using metrics like precision, recall, and F1 score, and highlights the primary challenges in this domain, including dataset heterogeneity, model generalizability, and explainability. The findings reveal that while early heuristic‐based techniques laid the foundation for SATD detection, more recent advancements in DL and Transformer models have significantly improved detection accuracy. However, challenges remain in scaling these models for broader industrial adoption. This review offers insights into current research gaps and provides directions for future work, aiming to improve the robustness and practicality of SATD detection tools.

Context: Researchers have increasingly recognised the benefit of mining software repositories to extract information. Although, the development logs of software projects, contained in Version Control (VC) systems can be severely incomplete when tracking bugs, especially in open-source software projects, resulting in reduced traceability of defects. Other times, such logs can contain bug information that is not available in the bug tracking system (BT system) repositories, and vice-versa. If the development logs and bug related-data (i.e., BT system data) were applied together, researchers and practitioners often would have a larger set of bug IDs for a software project, and a better picture of a bug life cycle, its evolution and maintenance. Method The research aims to design and implement a toolchain to support the integration of a VC and BT toolset as well to synchronise the missing development logs and Bug data of open-source software projects automatically. Considering a random sample of 344 Open Source Software (OSS) projects development logs (55,689) and Bugs data (167,800), the two objectives of this paper are (i) to determine which of the keywords 'Fix', 'Bug' or the '#' identifier provide better precision; and (ii) to analyse their respective precision and recall at locating significant amount possible of bug IDs semi-automatically. In its formulation, the SZZ algorithm looks for the terms ``Bugs'' or ``Fixed'' (case-insensitive) along with the '#' sign, that shows the ID of a bug in the development logs and Bug data respectively. Results:Overall, our results suggest that the use of the '#' identifier in conjunction with the bug ID digits (e.g., #1234) is more precise for locating bugs in the development logs than the use of the 'Bug' and 'Fix' keywords. Such keywords are indeed present in the development logs, but they are less useful when trying to connect the development actions with the bug traces in open source software project. The results indicate that the development log and Bug related data can be track and recovered with better accuracy using only a part of the SZZ algorithm. Thus 80-95\% of all The missing bug data and development logs of 344 OSS projects have been synchronised using the proposed tool-chain into Bicho and CVSAnalY database respectively. Conclusion: The presented toolchain eliminate and avoid repetitive activities in traceability tasks, software maintenance and evolution. The fact that in the past researchers have proven linking and synchronising development logs and bug data is complicated. Thus Bicho and CVSAnalY tools were developed to mine and store development logs and bug data independently. This research provides a solution towards the automation and traceability of bug data of software projects (in particular, OSS projects) using development logs to complement and track the missing bug data. The Synchronisation involves completing the missing bug data in software repositories with the development logs which details the actions of developers.