
Large-scale historical corpora present persistent challenges for research, particularly in relation to the navigation, interpretation, and extraction of meaningful information across extensive and heterogeneous collections. The Legislación Mexicana corpus, a 42-volume compilation of legal dispositions spanning more than two centuries, exemplifies these challenges due to its size, structural complexity, and evolving terminology. Traditional approaches to working with such corpora have relied on close reading and the use of indexes. While effective within defined limits, these methods constrain the scope of inquiry and require significant time and expertise to produce meaningful results. This article presents the development of LegMexIA, an AI-enabled research assistant designed to extend interaction with the corpus beyond conventional retrieval methods. The system integrates traditional text retrieval and retrieval-augmented generation via a coordinated architecture that distinguishes between different types of user queries. Structured queries are processed using Elasticsearch and BM25 ranking, while exploratory and interpretive queries are routed through a retrieval-augmented generation workflow, where relevant text fragments are retrieved using vector similarity and assembled into contextual inputs for a language model. A key aspect of the system is the agentic decision layer that analyzes user intent and dynamically selects the appropriate processing strategy, in alignment with established reference practices in academic libraries and in a way that maintains the user’s role in verification. The article also addresses key considerations related to prompt design, technological sustainability, institutional constraints, and the implications of public deployment.
Many heritage institutions hold extensive collections of show programs. Too numerous to be cataloged individually, they remain difficult to access and are still largely undigitized. This paper presents a workflow for transforming such documents into structured, interoperable data by combining vision-language models (VLMs), a custom extension of the Linked Art ontology for the performing arts, and different approaches for automatic semantic data annotation. Using the Festival d'Avignon programs (1947–present), preserved at the Bibliothèque nationale de France, as a case study, we demonstrate how VLMs achieve more than 98
This paper explores the emergence of artificial research intelligence (ARI), autonomous AI systems designed to conduct end-to-end scientific research, from ideation to manuscript production. Unlike simple AI enhancements, ARI utilizes ensembles of agentic models to automate the scientific method, potentially creating a “fifth scientific research paradigm.” While ARI offers rapid, low-cost discovery, it faces significant challenges including “epistemic capture,” where knowledge production shifts from public universities to private AI corporations. Technically, these systems currently struggle with hallucinations, a lack of deep domain expertise, and a tendency to overclaim results. However, the growing number of ARI frameworks points to an emerging research methodology of transformational possibilities. For research libraries, ARI necessitates a profound shift from managing static collections to stewarding dynamic, AI-driven knowledge systems. Libraries can lead by reimagining processes and services supporting scholarly communications, enhancing AI competencies, and conducting applied research through library-based AI labs. The combination of advanced compute and the guiding framework of the scientific method makes ARI a human–AI co-creation model where AI performs the "reckoning" (computation) and humans provide the "judgment" (ethical and intellectual oversight).
The rapid integration of generative artificial intelligence (GenAI) into academic library discovery systems offers new opportunities but also introduces challenges, as vendor tools streamline natural language searching while obscuring search logic. This article presents a case study of a librarian-designed AI discovery tool that is focused on transparency and pedagogy. The assistant converts natural language queries into structured, editable Boolean search strings; surfaces search logic; and retrieves results across library holdings. Developed through iterative prompt engineering, this tool demonstrates how GenAI can support research practices without sacrificing rigor. Findings from qualitative feedback and quantitative assessment indicate that the tool improves the consistency and quality of Boolean query construction while fostering user engagement with search as a reflective process. This case study argues that libraries can play a critical role in shaping AI discovery tools that align with information literacy goals and institutional scholarly standards. By foregrounding transparency and user agency, librarian-led AI design positions GenAI as a form of instructional scaffolding rather than a replacement for expertise.
This paper reviews Project Gutenberg’s (PG) use of artificial intelligence (AI) with the free books in its digital library. The paper covers three main projects where PG has integrated AI into its workflow. The three areas discussed are disparate in nature and include audiobook, book summary, and browsing category generation. The paper reviews the AI models and processes used, as well as how volunteers and users have responded to the organization’s use of AI. PG’s dataset is extremely large and includes more than 75,000 books. Large datasets can be challenging and costly for AI models to process. The main goal of this paper is to share how PG has incorporated AI processes into the workflow of a very large digital library while maintaining human oversight.