In the UK, libraries have always been more than places to borrow books. They are quiet infrastructure: catalogues that make knowledge findable, local collections that preserve community memory, and professional expertise that turns messy information into something people can actually use. For decades, that work has supported education, civic life and research. Now, as artificial intelligence moves from a specialist tool to a general-purpose technology, the value of that infrastructure is becoming harder to ignore.
AI systems do not “know” things in the way humans do. They learn patterns from data, and they perform best when the data they draw on is well described, reliably governed, and accessible under clear rules. That is where libraries—public, academic, and community—can play a role that is both practical and strategic. Not by trying to become tech companies, and not by pretending that libraries can solve every problem in the AI supply chain. Instead, libraries can help the UK build the conditions for trustworthy data use: the standards, stewardship practices, and access pathways that allow researchers, businesses and public bodies to work with information responsibly.
The shift is subtle but profound. In the past, the question was often whether information existed. Today, the question increasingly becomes whether it can be used—legally, ethically, technically, and at scale. Libraries sit at the intersection of those questions. They are stewards of vast quantities of data in many forms: digitised archives, local history materials, government publications, academic outputs, newspapers, oral histories, and research datasets. They also hold the metadata that makes these resources navigable. In an AI era, metadata is not a bureaucratic afterthought; it is the map that helps machines and people locate meaning.
To understand why this matters, it helps to look at what AI needs beyond raw content. Most AI applications require three things: data that is relevant, data that is usable, and data that is governed. Relevance is about coverage and context. Usability is about formats, interfaces, and documentation. Governance is about rights, consent, privacy, and accountability. Libraries are already built around all three, even if their mission statements rarely use the language of machine learning.
Consider the everyday library catalogue. It is a system for describing items so that users can discover them. In AI terms, that is a form of structured knowledge. When libraries maintain consistent subject headings, authority records, and classification schemes, they reduce ambiguity. When they link related works—editions, translations, citations, and provenance—they create relationships that can be exploited by search engines and, increasingly, by AI systems that need to understand how information connects.
Now imagine scaling that approach across digitised collections and research repositories. A library that digitises a local newspaper archive is not just scanning pages. It is deciding how to handle OCR errors, how to tag topics, how to represent dates and locations, and how to preserve original context. Those decisions determine whether the archive becomes a useful resource for historians, journalists, educators—and potentially for AI models trained to detect events, track narratives, or summarise local developments over time.
But the real opportunity lies in governance. AI projects often stall not because data does not exist, but because the rules for using it are unclear. Rights management can be complex, especially when materials include copyrighted content, personal data, or sensitive community records. Libraries have long experience navigating permissions, licensing, and ethical access. They are accustomed to balancing openness with protection, and they have established processes for handling requests, documenting decisions, and ensuring that access is appropriate.
This is particularly important as AI expands into areas where the stakes are high: health research, education, public services, and cultural heritage. In these domains, the cost of getting governance wrong is not merely reputational. It can mean violating privacy, undermining trust, or producing biased outputs that harm individuals and communities. Libraries, with their public-facing legitimacy and professional norms, can help create safer pathways for data use.
There is also a less discussed but equally significant issue: the quality of data. AI is only as good as the inputs it receives. Digitised archives can contain transcription errors, missing metadata, inconsistent naming conventions, and gaps in coverage. Libraries are trained to improve discoverability and accuracy through curation. They can apply quality assurance methods that are familiar in information science: validating metadata, correcting authority records, improving indexing, and maintaining provenance. In an AI context, these practices translate into better training data, more reliable retrieval, and fewer hallucinations driven by poor context.
Yet libraries should not be romanticised as perfect stewards. They face constraints: funding pressures, uneven digitisation progress, and varying levels of technical capacity across institutions. Some collections remain undigitised. Some metadata is incomplete. Some rights restrictions limit what can be shared openly. The point is not that libraries already have everything needed for AI. The point is that they have a foundation—and that foundation can be strengthened with targeted investment and policy support.
One unique angle in the UK is the breadth of the library ecosystem. Public libraries are embedded in local communities and can provide access to digital skills, information literacy, and trusted guidance. Academic libraries connect to research workflows and institutional repositories. National and specialist libraries hold large-scale cultural and historical collections. Community archives preserve materials that may never be prioritised by mainstream digitisation efforts. Together, they form a distributed network of knowledge stewardship.
That distribution matters because AI is not only a technical challenge; it is a social one. If AI development relies solely on data from a narrow set of sources—typically those that are easiest to license or most profitable to digitise—then the resulting systems will reflect those biases. Libraries can help broaden the data landscape by supporting access to diverse collections, including materials that represent minority communities, regional histories, and local languages. This is not charity; it is a way to improve representativeness and reduce blind spots in AI training and evaluation.
However, broadening access must be done carefully. Libraries operate under legal frameworks that protect copyright and personal data. They also have ethical responsibilities, especially when dealing with sensitive records. In practice, this means that “access” does not always mean open download. It can mean controlled access environments, secure data enclaves, mediated research services, and clear usage agreements. Libraries are well positioned to offer these models because they already manage user authentication, request workflows, and compliance processes.
In other words, libraries can help the UK move from a simplistic debate about open versus closed data to a more nuanced approach: appropriate access with accountability. That is exactly what AI developers need. Many organisations want to use data but cannot justify the risk of uncontrolled sharing. Libraries can provide the structure that allows responsible experimentation while protecting rights and privacy.
Another area where libraries can contribute is in the “last mile” of AI adoption: helping people understand what AI outputs mean and how to verify them. Information literacy has always been part of library culture. In the AI era, that becomes more urgent. AI systems can generate plausible text, images, and summaries that may be inaccurate or misleading. Users need skills to evaluate sources, check claims, and understand limitations. Libraries can support these skills through workshops, curated reading lists, guided searches, and partnerships with schools and universities.
This is not separate from data stewardship. It is part of the same mission. If libraries help people interpret information responsibly, they also strengthen the demand for trustworthy data practices. That creates a feedback loop: better-informed users ask better questions, which encourages better governance and higher-quality datasets.
There is also a workforce dimension. AI adoption requires people who can bridge disciplines: technologists who understand data governance, librarians who understand machine-readable formats, and researchers who know how to document datasets for reuse. Libraries can act as training hubs for these hybrid roles. They already employ professionals skilled in metadata, cataloguing, digital preservation, and research support. With additional training and collaboration, those skills can be extended into AI-adjacent work such as dataset documentation, annotation strategies, and evaluation of retrieval systems.
The UK’s policy environment makes this timely. The country is actively debating how to regulate AI, how to protect rights, and how to ensure that innovation benefits society. Libraries are not lawmakers, but they can influence implementation. They can pilot practical standards for data access, develop templates for licensing and consent, and collaborate with regulators and industry on workable models. When policy is abstract, pilots matter. Libraries can turn principles into procedures.
A particularly promising direction is the integration of library metadata and digital preservation practices into AI-ready infrastructures. Many AI projects struggle with interoperability: data stored in different formats, described with different schemas, and governed under different rules. Libraries can help by adopting common standards for metadata and persistent identifiers. They can also support long-term preservation, ensuring that datasets remain accessible and interpretable over time. AI is often treated as a fast-moving field, but the data it depends on must be stable. Preservation is therefore not just about archiving; it is about maintaining the reliability of the evidence base.
This is where libraries can offer a distinctive advantage over purely commercial data platforms. Commercial incentives often favour rapid acquisition and monetisation. Libraries are incentivised to maintain continuity, transparency, and public value. That does not mean they cannot innovate technologically. It means their innovation tends to be oriented toward durability and trust.
There is another layer: libraries can help ensure that AI development does not become extractive. Data extraction without meaningful benefit to communities can create resentment and reduce participation. Libraries, especially public and community institutions, can negotiate more balanced relationships. They can advocate for community consultation, fair attribution, and mechanisms for communities to see how their materials are used. Even when full control is impossible, libraries can push for transparency and respect.
This matters because AI is increasingly used to generate outputs that shape culture and decision-making. If AI systems learn from community archives, then communities should have a say in how those archives are represented. Libraries can facilitate that conversation, drawing on their experience as mediators between information and the public.
Of course, none of this happens automatically. Libraries need investment in digitisation, metadata enhancement, and technical infrastructure. They also need clarity on legal and regulatory expectations for AI