CINCEL libraries explore the opportunities and risks of AI in research support

Author: Heloisa Tiberio

AI is quickly reshaping how research is done, and what researchers expect from their libraries. In September/2025 a webinar hosted by Chile’s Consorcio para el Acceso a la Información Electrónica (CINCEL) and Springer Nature, librarians explored the concepts, trends, and practical use cases that matter most to academic library services today.

CINCEL is Chile’s national consortium for scientific information. Since 2003, it has helped universities and public institutions expand access to international journals, books, and digital resources, supporting more equitable access to research across the country. To continue empowering its librarian community, CINCEL partnered with Springer Nature to host a webinar focused on AI in practice. The session encouraged consortium librarians to explore how AI can support their daily activities, while supporting the research community.

The webinar began with a quick pulse check to see if the librarians were using AI in their tasks. The pull result showed that 41% had experimented but didn’t use them regularly, 39% used AI for specific tasks, and a smaller group had not used AI yet, or weren’t sure what “counts” as AI.

Building on this context, the webinar introduced core AI terminology and the technologies most relevant to academic libraries today. This foundation allowed for a broader conversation about what AI can offer, from enhancing services to supporting researchers more efficiently, while also addressing the questions and ethical considerations that come with adopting new tools.

Here’s a look at the central points raised during the webinar, providing a straightforward overview of emerging AI trends that can inform your library’s next steps.

From AI foundations to the idea of superintelligence

To raise the audience's knowledge level, the webinar began providing simple definitions of AI concepts. At its simplest, Artificial Intelligence (AI) is the ability of a system to perform tasks we associate with human intelligence, such as understanding language, recognising patterns, making predictions, or generating content. In the library context, it could be translated as faster drafting, summarisation, cataloguing, translation, and support for discovery workflows.

• Large Language Models (LLMs): LLMs are AI models trained on massive collections of text to understand natural language prompts and generate humanlike responses. Well-known LLMs include ChatGPT, Copilot, Gemini, and Claude. They perform tasks such as drafting, summarizing, translating, and answering questions.
• Machine Learning (ML) Techniques: LLMs are built using ML techniques that enable them to learn from large datasets and predict relevant outputs. The session covered four key ML approaches:
  • Supervised learning: Learns from labelled examples (e.g., correct/incorrect) and uses this knowledge to make predictions, for example, usage predictions.
  • Unsupervised learning: Finds patterns or clusters in data without labels; useful for uncovering trends such as usage behaviour.
  • Reinforcement learning: Improves through trial-and-error feedback, optimising decisions over time.
  • Deep learning: A subset of ML that uses multi‑layer neural networks to process unstructured data such as long text, images, and audio.

The webinar also highlighted a practical distinction: Generative AI creates content, while retrieval-augmented generation (RAG) combines generation with search, returning answers grounded in specific documents (often with citations). For research support, grounding and traceability are essential.

Finally, the session briefly defined two long-term concepts that often appear in AI conversations: artificial general intelligence (AGI), a hypothetical AI that could perform any intellectual task a human can, and superintelligence, which would exceed human capability across domains. These ideas remain speculative, but they continue to shape debates about AI governance and safety.

How libraries are putting AI to work

As libraries continue to navigate rapid developments in AI, the webinar highlighted several trends that are likely to be felt most directly in 2025. These include the emergence of AI agents and increased automation, the growing availability of smaller language models (SLMs), expanded data‑centre infrastructure, and rising attention to environmental impact and workforce change.

Rather than discussing these shifts in isolation, the session connected them to the practical opportunities already available to libraries. Many of these trends are influencing day‑to‑day work in small but meaningful ways, creating new options for streamlining tasks, enhancing services, and supporting decision‑making. This set the stage for a series of concrete examples showing how libraries are beginning to apply AI today.

• Content creation/revision, Translation, Data analysis (user behavior, collection usage, predictions): Draft emails, instructional materials, workshop outlines, or guides, tasks that often take librarians considerable time to refine. Tools: AI Platforms (Chat-GPT, Gemini, Copilot, Claude, Perplexity, Grok, Deep Seek).
• Synthetic Videos: Product and service demos, tutorials, educational videos on specific topics, and short videos for social media. Tools example: Prezi, Colossyan, Synthesia, HeyGen, StoryKit.
• AI Agents and Chatbots: Process automation (cataloguing, classification, metadata), virtual assistants and chatbots: 24/7 user support. Example of tools: Copilot Studio, Chatbase, Botsonic, QuickChat.  Real-world example: bot use at Hong Kong Polytechnic University Library.
• More advanced tools for Data Analytics and Design/Prototyping: Lovable (Create interactive forms, Design thematic microsites, Develop internal tools, Prototype innovation ideas) and Google Colab (Bibliographic data analysis, Information visualization, Task automation).

The takeaway: AI is already useful in small, concrete ways, especially when it reduces routine work and frees time for higher-value research support.

Ethical, responsible and transparent AI use

The webinar stressed that AI outputs can reflect bias in training data (including gaps in language and cultural coverage). The recommendation is to treat AI results as drafts and always check for: validation, contextualisation, and uneven performance across user groups.

Another essential theme: Data privacy is a key concern. Many free AI tools use submitted information to train or improve their models, which means anything entered may not stay confidential. Librarians should be cautious and ensure they do not share sensitive or restricted content when using such tools.

Throughout the discussion, one message remained consistent: Keeping a human in the loop is critical. AI systems can produce plausible but incorrect information, and expert review is essential to maintain accuracy, trust, and integrity.

SN AI capabilities supporting the research community

The webinar also shared examples of Springer Nature editorial AI tools to support researchers, improve workflows, and strengthen research integrity.

Nature Research Assistant (currently in beta) is designed to help researchers work through growing volumes of literature by supporting discovery, summarising evidence, and helping with early drafting.

For research integrity, Springer Nature has developed AI tools such as our irrelevant reference checker tool and our AI-generated nonsense-text-checking tool, previously known as Geppetto. After internal use, Springer Nature donated Geppettoto the STM Integrity Hub to support wider adoption across the community.

Springer Nature also supports global collaboration through free AI-powered translation services for book authors, and through protocols.io, which can convert existing documents into interactive, publishable protocols (with careful review of imported content).

Preparing libraries for an AI‑driven research future

The session wrapped up by looking ahead at what growing AI adoption could mean for academic libraries. Rather than diving into big projects straight away, librarians were encouraged to start small, try out simple tools, set clear goals, and gradually build confidence. Even testing an AI assistant for common questions or using a tool to summarise a long document can help teams get a feel for where AI genuinely adds value.

Collaboration came through as another key theme. Creating communities of practice, teaming up with colleagues from other departments, or joining hackathons can make it easier to learn, swap ideas, and experiment together. These kinds of shared efforts have already sparked creative solutions in other settings, including prototypes developed within Springer Nature.

Besides introducing practical ways to begin using AI in libraries, responsible adoption is key. The webinar emphasized the importance of using AI ethically and responsibly, guided by Springer Nature’s AI Principles: Dignity, Respect, and Minimizing Harm; Fairness and Inclusivity; Transparency; Accountability; Privacy and Data Governance.

Looking ahead, building AI literacy will help everyone navigate changes in research workflows. Springer Nature’s publication Perspectives on AI in Scholarly Communications brings together voices from across the ecosystem and offers helpful context on how AI is influencing the way research is produced, shared , and evaluated. These insights can support libraries as they consider how AI might fit into their own services and future planning.

Related content

• Exploring AI with research librarians at Charleston
• Exploring AI futures in libraries with the Futurescape Workshop community
• How libraries and publishers are shaping the future of AI in research
• Evolving roles of academic librarians adapting to change and new technologies
• From the floor of ALA to the future of libraries

Get the latest news & blogs, subscribe to The Link Alerts!